{
  "generated": true,
  "meta": {
    "repo_owner": "wjlgatech",
    "repo_name": "FM-os",
    "title": "🛠️ FM-os — the Foundation Model Operating System",
    "tagline": "The most comprehensive, community-driven, **living** map of how modern language models are actually built and shipped — pre-training · post-training · fine-tuning · RL — with a first, sharp focus on **Small Language Models (SLM)**.",
    "subtitle": "*From a 135M model you can train on one GPU to the RL recipes behind frontier reasoning — every repo, course, paper, and job worth your time, cross-linked and kept fresh automatically.*",
    "applied_banner": "🧬 **The flywheel, applied → [longevity-loop](https://github.com/wjlgatech/longevity-loop)**: an AI-native, build-in-public loop turning this SLM/FM-ops method into real aging-science results — code-only, verified, no wet lab.",
    "nav": [
      {
        "label": "Start Here",
        "anchor": "start-here"
      },
      {
        "label": "Repos",
        "anchor": "open-source-repos"
      },
      {
        "label": "Courses",
        "anchor": "courses"
      },
      {
        "label": "Papers",
        "anchor": "papers"
      },
      {
        "label": "Reading Lists",
        "anchor": "reading-lists"
      },
      {
        "label": "Jobs",
        "anchor": "jobs--careers"
      },
      {
        "label": "Roadmap",
        "anchor": "learning-roadmap"
      },
      {
        "label": "Contribute",
        "anchor": "contribute"
      }
    ],
    "why_different": [
      "**SLM-first.** Not another everything-list — organized around small, efficient, trainable-on-a-budget models and the exact ops that make them work.",
      "**Lifecycle-structured.** Everything filed under the real FM pipeline: pre-training → post-training → fine-tuning → RL → serving.",
      "**Cross-linked.** Papers point to code, code points to courses, courses point to jobs — follow a thread from idea to hire.",
      "**Auto-fresh.** A weekly GitHub Action re-checks every repo's stars, latest release, and links, then opens a PR — this list is never stale.",
      "**Data-driven & forkable.** Every entry lives in a plain `data/*.yml` file; the README is generated. Adding a resource is a two-line PR."
    ],
    "footer_note": "FM-os is maintained by [@wjlgatech](https://github.com/wjlgatech) and the community. Sibling projects: [longevity-loop](https://github.com/wjlgatech/longevity-loop) · [rsi](https://github.com/wjlgatech/rsi) · [FDE-os](https://github.com/wjlgatech/FDE-os).",
    "description": "🛠️ FM-os: the living, SLM-first map of foundation-model operations — pre-training, post-training, fine-tuning & RL. Curated repos, courses, papers & jobs, auto-refreshed weekly.",
    "topics": [
      "small-language-models",
      "slm",
      "foundation-models",
      "large-language-models",
      "llm",
      "pretraining",
      "post-training",
      "fine-tuning",
      "rlhf",
      "reinforcement-learning",
      "dpo",
      "grpo",
      "awesome",
      "awesome-list",
      "machine-learning",
      "deep-learning",
      "model-distillation",
      "quantization",
      "llmops",
      "ai-education"
    ],
    "sections": [
      {
        "id": "open-source-repos",
        "icon": "🧰",
        "title": "Open-Source Repos",
        "source": "repos",
        "key": "category",
        "groups": [
          {
            "key": "models",
            "title": "Small & Efficient Models"
          },
          {
            "key": "multimodal",
            "title": "Vision-Language & Video Models"
          },
          {
            "key": "frameworks",
            "title": "Pre-training & Training Frameworks"
          },
          {
            "key": "jax",
            "title": "JAX / TensorFlow Ecosystem"
          },
          {
            "key": "finetuning",
            "title": "Fine-tuning & PEFT"
          },
          {
            "key": "rl",
            "title": "Post-training & RL (RLHF / DPO / GRPO)"
          },
          {
            "key": "eval",
            "title": "Evaluation"
          },
          {
            "key": "serving",
            "title": "Serving, Inference & On-Device"
          },
          {
            "key": "compression",
            "title": "Distillation & Compression"
          },
          {
            "key": "retrieval",
            "title": "Retrieval & Vector Databases"
          },
          {
            "key": "distributed",
            "title": "Distributed Training & Orchestration"
          },
          {
            "key": "dataset",
            "title": "AV / Robotics / Video Datasets"
          },
          {
            "key": "research",
            "title": "Autonomous Research Agents (AI Scientists)"
          },
          {
            "key": "bayesopt",
            "title": "Bayesian Optimization & Active Learning"
          },
          {
            "key": "probabilistic",
            "title": "Probabilistic Programming & Uncertainty"
          },
          {
            "key": "bio-ml",
            "title": "ML for Molecules & Bio Design"
          },
          {
            "key": "neuro",
            "title": "Neuroscience & Brain-Computer Interfaces"
          }
        ]
      },
      {
        "id": "courses",
        "icon": "🎓",
        "title": "Courses",
        "source": "courses",
        "key": "topic",
        "groups": [
          {
            "key": "foundations",
            "title": "Foundations & From-Scratch"
          },
          {
            "key": "pretraining",
            "title": "Pre-training"
          },
          {
            "key": "posttraining",
            "title": "Post-training & Alignment"
          },
          {
            "key": "finetuning",
            "title": "Fine-tuning"
          },
          {
            "key": "rl",
            "title": "Reinforcement Learning"
          },
          {
            "key": "agents",
            "title": "Agents & Applications"
          },
          {
            "key": "vision",
            "title": "Computer Vision"
          },
          {
            "key": "multimodal",
            "title": "Multi-modal Machine Learning"
          },
          {
            "key": "vlm",
            "title": "Vision-Language Models"
          },
          {
            "key": "video",
            "title": "Video & Data Ops"
          },
          {
            "key": "bayesopt",
            "title": "Bayesian Optimization & Gaussian Processes"
          },
          {
            "key": "probabilistic",
            "title": "Probabilistic & Bayesian ML"
          },
          {
            "key": "molecular",
            "title": "ML for Molecules & Science"
          },
          {
            "key": "neuroscience",
            "title": "Neuroscience & Brain-Computer Interfaces"
          }
        ]
      },
      {
        "id": "papers",
        "icon": "📄",
        "title": "Papers",
        "source": "papers",
        "key": "topic",
        "groups": [
          {
            "key": "slm",
            "title": "Small Language Models & Surveys"
          },
          {
            "key": "pretraining",
            "title": "Pre-training & Data"
          },
          {
            "key": "scaling",
            "title": "Scaling Laws"
          },
          {
            "key": "posttraining",
            "title": "Post-training & Alignment"
          },
          {
            "key": "rl",
            "title": "RL & Reasoning"
          },
          {
            "key": "peft",
            "title": "Parameter-Efficient Fine-tuning"
          },
          {
            "key": "distillation",
            "title": "Distillation & Compression"
          },
          {
            "key": "compression",
            "title": "Distillation & Compression"
          },
          {
            "key": "vlm",
            "title": "Vision-Language Models"
          },
          {
            "key": "video",
            "title": "Video Understanding"
          },
          {
            "key": "multimodal",
            "title": "Multi-modal & Grounding"
          },
          {
            "key": "retrieval",
            "title": "Retrieval & Embeddings"
          },
          {
            "key": "eval",
            "title": "Multi-modal Evaluation"
          },
          {
            "key": "research",
            "title": "Autonomous Research & AI Scientists"
          },
          {
            "key": "bayesopt",
            "title": "Bayesian Optimization & Active Learning"
          },
          {
            "key": "active-learning",
            "title": "Bayesian Optimization & Active Learning"
          },
          {
            "key": "uncertainty",
            "title": "Probabilistic Modeling & Uncertainty"
          },
          {
            "key": "representation",
            "title": "Representation Learning"
          },
          {
            "key": "molecular",
            "title": "ML for Molecular & Bio Design"
          },
          {
            "key": "neuroscience",
            "title": "Neuroscience & Brain-Computer Interfaces"
          },
          {
            "key": "ssm",
            "title": "State-Space & Sequence Models"
          }
        ]
      },
      {
        "id": "reading-lists",
        "icon": "📚",
        "title": "Reading Lists",
        "source": "readinglists",
        "key": "focus",
        "groups": [
          {
            "key": "ai",
            "title": "AI — Deep-Learning Canon → Physical AI → AGI/ASI"
          },
          {
            "key": "ai-research",
            "title": "AI-Enabled Research — AI × Math & Science"
          },
          {
            "key": "math",
            "title": "Mathematics"
          },
          {
            "key": "physics",
            "title": "Physics"
          },
          {
            "key": "cs",
            "title": "Computer Science"
          },
          {
            "key": "bio",
            "title": "Biology · BioMed · BioTech"
          },
          {
            "key": "index",
            "title": "Track Index & Contract"
          }
        ]
      },
      {
        "id": "labs--platforms",
        "icon": "🏛️",
        "title": "Labs & Platforms",
        "source": "labs",
        "key": "category",
        "groups": [
          {
            "key": "frontier-lab",
            "title": "Frontier Labs"
          },
          {
            "key": "finetuning",
            "title": "Fine-tuning Platforms"
          },
          {
            "key": "inference-serving",
            "title": "Inference & Serving"
          },
          {
            "key": "infra",
            "title": "Compute & Infra"
          },
          {
            "key": "data-platform",
            "title": "Data & Lakehouse"
          },
          {
            "key": "open-research",
            "title": "Open Research"
          },
          {
            "key": "bio-ai",
            "title": "Bio × AI / Closed-Loop Science Labs"
          }
        ]
      },
      {
        "id": "people--researchers",
        "icon": "🧑‍🔬",
        "title": "People to Follow",
        "source": "people",
        "key": "category",
        "groups": [
          {
            "key": "bayesopt",
            "title": "Bayesian Optimization & Gaussian Processes"
          },
          {
            "key": "molecular",
            "title": "ML for Molecules & Self-Driving Labs"
          },
          {
            "key": "probabilistic",
            "title": "Probabilistic ML & Uncertainty"
          },
          {
            "key": "ssm",
            "title": "State-Space Models & Neuro-adjacent"
          }
        ]
      },
      {
        "id": "jobs--careers",
        "icon": "💼",
        "title": "Jobs & Careers",
        "source": "jobs",
        "key": "type",
        "groups": [
          {
            "key": "company",
            "title": "Frontier Labs (careers pages)"
          },
          {
            "key": "board",
            "title": "Specialized AI Job Boards"
          },
          {
            "key": "aggregator",
            "title": "Aggregators"
          },
          {
            "key": "newsletter",
            "title": "Newsletters & Signals"
          }
        ]
      }
    ]
  },
  "repos": [
    {
      "name": "SmolLM / SmolLM2 / SmolLM3",
      "repo": "huggingface/smollm",
      "url": "https://github.com/huggingface/smollm",
      "category": "models",
      "slm": true,
      "blurb": "Fully open recipes, data, and weights for the 135M-3B SmolLM family, the reference open SLM line.",
      "stars": 3844
    },
    {
      "name": "Phi Cookbook",
      "repo": "microsoft/PhiCookBook",
      "url": "https://github.com/microsoft/PhiCookBook",
      "category": "models",
      "slm": true,
      "blurb": "Microsoft's official hub for the Phi SLM family with inference, fine-tuning, quantization, and edge-deployment recipes.",
      "stars": 3767
    },
    {
      "name": "Gemma (DeepMind)",
      "repo": "google-deepmind/gemma",
      "url": "https://github.com/google-deepmind/gemma",
      "category": "models",
      "slm": true,
      "blurb": "Official JAX library for Gemma open weights including the 1B/2B and 3n on-device small variants.",
      "stars": 5555
    },
    {
      "name": "Qwen3",
      "repo": "QwenLM/Qwen3",
      "url": "https://github.com/QwenLM/Qwen3",
      "category": "models",
      "slm": true,
      "blurb": "Alibaba's Qwen series spanning 0.6B/1.7B/4B dense SLMs with strong multilingual and reasoning quality.",
      "stars": 27393
    },
    {
      "name": "gemma_pytorch",
      "repo": "google/gemma_pytorch",
      "url": "https://github.com/google/gemma_pytorch",
      "category": "models",
      "slm": true,
      "blurb": "Official PyTorch inference implementation of Gemma (incl. small text-only variants) for CPU/GPU/TPU.",
      "stars": 5711
    },
    {
      "name": "TinyLlama",
      "repo": "jzhang38/TinyLlama",
      "url": "https://github.com/jzhang38/TinyLlama",
      "category": "models",
      "slm": true,
      "blurb": "Compact 1.1B Llama pretrained on 3T tokens; a canonical, reproducible sub-2B pretraining reference.",
      "stars": 9011
    },
    {
      "name": "MobileLLM",
      "repo": "facebookresearch/MobileLLM",
      "url": "https://github.com/facebookresearch/MobileLLM",
      "category": "models",
      "slm": true,
      "blurb": "Meta research on sub-billion-parameter, deep-thin architectures optimized for on-device use (ICML 2024).",
      "stars": 1451
    },
    {
      "name": "OLMo",
      "repo": "allenai/OLMo",
      "url": "https://github.com/allenai/OLMo",
      "category": "models",
      "slm": true,
      "blurb": "AI2's fully open model+data+training stack including small 1B variants for reproducible SLM research.",
      "stars": 6587
    },
    {
      "name": "Llama Models",
      "repo": "meta-llama/llama-models",
      "url": "https://github.com/meta-llama/llama-models",
      "category": "models",
      "slm": true,
      "blurb": "Meta's official utilities and model cards for Llama, including the 1B/3B Llama 3.2 on-device SLMs.",
      "stars": 7651
    },
    {
      "name": "nanoGPT",
      "repo": "karpathy/nanoGPT",
      "url": "https://github.com/karpathy/nanoGPT",
      "category": "frameworks",
      "slm": true,
      "blurb": "Minimal ~300-line GPT training/finetuning loop; the standard starting point for training small GPTs from scratch.",
      "stars": 61156
    },
    {
      "name": "LitGPT",
      "repo": "Lightning-AI/litgpt",
      "url": "https://github.com/Lightning-AI/litgpt",
      "category": "frameworks",
      "slm": true,
      "blurb": "20+ hackable LLM implementations with pretrain/finetune/deploy recipes, including small Phi/Qwen/Gemma models.",
      "stars": 13482
    },
    {
      "name": "GPT-NeoX",
      "repo": "EleutherAI/gpt-neox",
      "url": "https://github.com/EleutherAI/gpt-neox",
      "category": "frameworks",
      "slm": false,
      "blurb": "EleutherAI's Megatron+DeepSpeed training stack for autoregressive transformers with 3D parallelism.",
      "stars": 7443
    },
    {
      "name": "Megatron-LM",
      "repo": "NVIDIA/Megatron-LM",
      "url": "https://github.com/NVIDIA/Megatron-LM",
      "category": "frameworks",
      "slm": false,
      "blurb": "NVIDIA's GPU-optimized library and building blocks for large-scale transformer pretraining.",
      "stars": 17064
    },
    {
      "name": "TorchTitan",
      "repo": "pytorch/torchtitan",
      "url": "https://github.com/pytorch/torchtitan",
      "category": "frameworks",
      "slm": false,
      "blurb": "PyTorch-native platform for generative-model pretraining with composable FSDP2/TP/PP/CP parallelism.",
      "stars": 5532
    },
    {
      "name": "Nanotron",
      "repo": "huggingface/nanotron",
      "url": "https://github.com/huggingface/nanotron",
      "category": "frameworks",
      "slm": false,
      "blurb": "Minimalistic 3D-parallelism pretraining library from Hugging Face, basis of the Ultrascale Playbook.",
      "stars": 2747
    },
    {
      "name": "PEFT",
      "repo": "huggingface/peft",
      "url": "https://github.com/huggingface/peft",
      "category": "finetuning",
      "slm": true,
      "blurb": "Reference library for LoRA/QLoRA and other parameter-efficient methods, enabling SLM tuning on consumer GPUs.",
      "stars": 21394
    },
    {
      "name": "Unsloth",
      "repo": "unslothai/unsloth",
      "url": "https://github.com/unslothai/unsloth",
      "category": "finetuning",
      "slm": true,
      "blurb": "2x-faster, ~70%-less-VRAM finetuning for small models, ideal for LoRA/QLoRA on single-GPU setups.",
      "stars": 68200
    },
    {
      "name": "Axolotl",
      "repo": "axolotl-ai-cloud/axolotl",
      "url": "https://github.com/axolotl-ai-cloud/axolotl",
      "category": "finetuning",
      "slm": true,
      "blurb": "Config-driven post-training framework covering SFT/LoRA/DPO across many small and large model families.",
      "stars": 12197
    },
    {
      "name": "LLaMA-Factory",
      "repo": "hiyouga/LLaMA-Factory",
      "url": "https://github.com/hiyouga/LLaMA-Factory",
      "category": "finetuning",
      "slm": true,
      "blurb": "Unified zero-code fine-tuning of 100+ LLMs/VLMs with LoRA/QLoRA/DPO and a web UI, common for SLM tuning.",
      "stars": 73271
    },
    {
      "name": "torchtune",
      "repo": "meta-pytorch/torchtune",
      "url": "https://github.com/meta-pytorch/torchtune",
      "category": "finetuning",
      "slm": true,
      "blurb": "PyTorch-native post-training recipes (SFT, distillation, DPO/PPO/GRPO, QAT) tuned for memory-limited hardware.",
      "stars": 5783
    },
    {
      "name": "TRL",
      "repo": "huggingface/trl",
      "url": "https://github.com/huggingface/trl",
      "category": "rl",
      "slm": true,
      "blurb": "Hugging Face post-training library with SFT/DPO/GRPO trainers widely used to align small reasoning models.",
      "stars": 18840
    },
    {
      "name": "OpenRLHF",
      "repo": "OpenRLHF/OpenRLHF",
      "url": "https://github.com/OpenRLHF/OpenRLHF",
      "category": "rl",
      "slm": true,
      "blurb": "Ray+vLLM RLHF framework (PPO/GRPO/RLOO) that scales from small models up to 70B+, agent-friendly.",
      "stars": 9788
    },
    {
      "name": "verl",
      "repo": "verl-project/verl",
      "url": "https://github.com/verl-project/verl",
      "category": "rl",
      "slm": true,
      "blurb": "ByteDance HybridFlow RL post-training (PPO/GRPO/DAPO) with vLLM/SGLang; popular for GRPO on small models.",
      "stars": 22468
    },
    {
      "name": "trlX",
      "repo": "CarperAI/trlx",
      "url": "https://github.com/CarperAI/trlx",
      "category": "rl",
      "slm": false,
      "blurb": "Distributed RLHF framework (PPO, ILQL) via Accelerate/NeMo; an early, widely-cited RLHF reference.",
      "stars": 4753
    },
    {
      "name": "lm-evaluation-harness",
      "repo": "EleutherAI/lm-evaluation-harness",
      "url": "https://github.com/EleutherAI/lm-evaluation-harness",
      "category": "eval",
      "slm": true,
      "blurb": "De-facto standard few-shot eval harness (60+ benchmarks) backing the Open LLM Leaderboard, ideal for SLM benchmarking.",
      "stars": 13279
    },
    {
      "name": "LightEval",
      "repo": "huggingface/lighteval",
      "url": "https://github.com/huggingface/lighteval",
      "category": "eval",
      "slm": true,
      "blurb": "Hugging Face all-in-one evaluator across vLLM/Accelerate/TGI backends with 1000+ tasks for small-model eval.",
      "stars": 2476
    },
    {
      "name": "llama.cpp",
      "repo": "ggml-org/llama.cpp",
      "url": "https://github.com/ggml-org/llama.cpp",
      "category": "serving",
      "slm": true,
      "blurb": "C/C++ GGUF inference engine that runs quantized SLMs efficiently on CPUs, laptops, and edge devices.",
      "stars": 120340
    },
    {
      "name": "vLLM",
      "repo": "vllm-project/vllm",
      "url": "https://github.com/vllm-project/vllm",
      "category": "serving",
      "slm": false,
      "blurb": "High-throughput PagedAttention serving engine; the default for scalable OpenAI-compatible model serving.",
      "stars": 86247
    },
    {
      "name": "Ollama",
      "repo": "ollama/ollama",
      "url": "https://github.com/ollama/ollama",
      "category": "serving",
      "slm": true,
      "blurb": "One-command local runner for small open models, the easiest path to running SLMs on a personal machine.",
      "stars": 176100
    },
    {
      "name": "MLC-LLM",
      "repo": "mlc-ai/mlc-llm",
      "url": "https://github.com/mlc-ai/mlc-llm",
      "category": "serving",
      "slm": true,
      "blurb": "ML-compilation deployment engine that compiles SLMs to iOS, Android, WebGPU, and diverse GPUs/CPUs.",
      "stars": 22948
    },
    {
      "name": "BitNet",
      "repo": "microsoft/BitNet",
      "url": "https://github.com/microsoft/BitNet",
      "category": "compression",
      "slm": true,
      "blurb": "Official 1-bit (1.58-bit) LLM inference framework with optimized CPU/GPU kernels for extreme efficiency.",
      "stars": 39724
    },
    {
      "name": "LLM-AWQ",
      "repo": "mit-han-lab/llm-awq",
      "url": "https://github.com/mit-han-lab/llm-awq",
      "category": "compression",
      "slm": true,
      "blurb": "Activation-aware INT3/4 weight quantization (MLSys 2024) plus TinyChat for on-device/edge SLM inference.",
      "stars": 3589
    },
    {
      "name": "GPTQModel",
      "repo": "ModelCloud/GPTQModel",
      "url": "https://github.com/ModelCloud/GPTQModel",
      "category": "compression",
      "slm": true,
      "blurb": "Actively maintained GPTQ quantization toolkit with HF/vLLM/SGLang support across NVIDIA/AMD/Intel/Apple hardware.",
      "stars": 1205
    },
    {
      "name": "LightCompress (LLMC)",
      "repo": "ModelTC/LightCompress",
      "url": "https://github.com/ModelTC/LightCompress",
      "category": "compression",
      "slm": true,
      "blurb": "Broad model-compression toolkit (quantization, sparsity, pruning) for shrinking LLMs/VLMs to deployable sizes.",
      "stars": 733
    },
    {
      "name": "DistillKit",
      "repo": "arcee-ai/DistillKit",
      "url": "https://github.com/arcee-ai/DistillKit",
      "category": "compression",
      "slm": true,
      "blurb": "Open toolkit for knowledge distillation, training smaller student models from larger teachers (logit + hidden-state).",
      "stars": 976
    },
    {
      "name": "Qwen2.5-VL",
      "repo": "QwenLM/Qwen2.5-VL",
      "url": "https://github.com/QwenLM/Qwen2.5-VL",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Strong open VLM family with native dynamic-resolution and long-video/temporal grounding, a common backbone for fine-tuning on driving footage."
    },
    {
      "name": "LLaVA-NeXT",
      "repo": "LLaVA-VL/LLaVA-NeXT",
      "url": "https://github.com/LLaVA-VL/LLaVA-NeXT",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Actively maintained LLaVA line (incl. OneVision and Video variants) with training and eval recipes for image, multi-image, and video."
    },
    {
      "name": "InternVL",
      "repo": "OpenGVLab/InternVL",
      "url": "https://github.com/OpenGVLab/InternVL",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Scaled open VLM series with large vision encoders and full training code, competitive on high-resolution perception and video benchmarks."
    },
    {
      "name": "VideoLLaMA3",
      "repo": "DAMO-NLP-SG/VideoLLaMA3",
      "url": "https://github.com/DAMO-NLP-SG/VideoLLaMA3",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Vision-centric image/video foundation model with released training and inference code, directly targeting long-form video understanding."
    },
    {
      "name": "Video-LLaVA",
      "repo": "PKU-YuanGroup/Video-LLaVA",
      "url": "https://github.com/PKU-YuanGroup/Video-LLaVA",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Unified image+video projection into one representation before the LLM, a compact reference for joint image/video instruction tuning."
    },
    {
      "name": "CLIP",
      "repo": "openai/CLIP",
      "url": "https://github.com/openai/CLIP",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Original contrastive image-text model; the reference whose embeddings still anchor most multimodal retrieval and probing."
    },
    {
      "name": "open_clip",
      "repo": "mlfoundations/open_clip",
      "url": "https://github.com/mlfoundations/open_clip",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Open training/eval for CLIP-style models at scale, the go-to for reproducible contrastive image-text encoders and domain pretraining."
    },
    {
      "name": "MiniCPM-V",
      "repo": "OpenBMB/MiniCPM-V",
      "url": "https://github.com/OpenBMB/MiniCPM-V",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Efficient end-side VLM series with strong image/video/OCR performance, relevant where on-vehicle or edge inference budgets are tight."
    },
    {
      "name": "Molmo",
      "repo": "allenai/molmo",
      "url": "https://github.com/allenai/molmo",
      "category": "multimodal",
      "vlm": true,
      "blurb": "Ai2's fully open VLM with training code and the PixMo data, a transparent reproducible baseline including pointing/grounding."
    },
    {
      "name": "Hugging Face Transformers",
      "repo": "huggingface/transformers",
      "url": "https://github.com/huggingface/transformers",
      "category": "frameworks",
      "vlm": true,
      "blurb": "De facto model hub and API with first-class VLM/video-LLM support, the integration surface most training and serving stacks build on."
    },
    {
      "name": "DeepSpeed",
      "repo": "deepspeedai/DeepSpeed",
      "url": "https://github.com/deepspeedai/DeepSpeed",
      "category": "frameworks",
      "vlm": false,
      "blurb": "ZeRO sharding, offload, and pipeline/tensor parallelism that make large VLM training fit real GPU budgets; wired into most trainers."
    },
    {
      "name": "ms-swift",
      "repo": "modelscope/ms-swift",
      "url": "https://github.com/modelscope/ms-swift",
      "category": "finetuning",
      "vlm": true,
      "blurb": "Unified SFT/DPO/GRPO toolkit covering 300+ multimodal models (Qwen-VL, InternVL, LLaVA), a fast path to fine-tune VLMs on custom data."
    },
    {
      "name": "XTuner",
      "repo": "InternLM/xtuner",
      "url": "https://github.com/InternLM/xtuner",
      "category": "finetuning",
      "vlm": true,
      "blurb": "Memory-efficient LLM/VLM fine-tuning engine (LLaVA-style pipelines, large-MoE support) for constrained or very large setups."
    },
    {
      "name": "lmms-eval",
      "repo": "EvolvingLMMs-Lab/lmms-eval",
      "url": "https://github.com/EvolvingLMMs-Lab/lmms-eval",
      "category": "eval",
      "vlm": true,
      "blurb": "One-command multimodal eval harness across image/video/audio benchmarks, the standard for consistent VLM regression testing."
    },
    {
      "name": "VLMEvalKit",
      "repo": "open-compass/VLMEvalKit",
      "url": "https://github.com/open-compass/VLMEvalKit",
      "category": "eval",
      "vlm": true,
      "blurb": "Broad LMM evaluation toolkit (220+ models, 80+ benchmarks) with unified data prep, complementary to lmms-eval for coverage."
    },
    {
      "name": "SGLang",
      "repo": "sgl-project/sglang",
      "url": "https://github.com/sgl-project/sglang",
      "category": "serving",
      "vlm": true,
      "blurb": "Fast serving runtime with RadixAttention and structured decoding plus VLM support, strong for high-concurrency multimodal endpoints."
    },
    {
      "name": "LMDeploy",
      "repo": "InternLM/lmdeploy",
      "url": "https://github.com/InternLM/lmdeploy",
      "category": "serving",
      "vlm": true,
      "blurb": "Compression + serving toolkit with a dedicated VLM pipeline, for quantized, low-latency deployment of vision-language models."
    },
    {
      "name": "FAISS",
      "repo": "facebookresearch/faiss",
      "url": "https://github.com/facebookresearch/faiss",
      "category": "retrieval",
      "vlm": false,
      "blurb": "Battle-tested library for billion-scale similarity search over image/video embeddings, the baseline for mining and nearest-neighbor lookup."
    },
    {
      "name": "Milvus",
      "repo": "milvus-io/milvus",
      "url": "https://github.com/milvus-io/milvus",
      "category": "retrieval",
      "vlm": false,
      "blurb": "Distributed vector database for large multimodal embedding corpora, used when single-node indices no longer fit."
    },
    {
      "name": "Qdrant",
      "repo": "qdrant/qdrant",
      "url": "https://github.com/qdrant/qdrant",
      "category": "retrieval",
      "vlm": false,
      "blurb": "Rust vector DB with payload filtering and good ergonomics, common for production embedding search over image/video/text."
    },
    {
      "name": "LanceDB",
      "repo": "lancedb/lancedb",
      "url": "https://github.com/lancedb/lancedb",
      "category": "retrieval",
      "vlm": false,
      "blurb": "Embedded columnar vector store on the Lance format, well suited to versioned multimodal datasets and fast on-disk embedding queries."
    },
    {
      "name": "Ray",
      "repo": "ray-project/ray",
      "url": "https://github.com/ray-project/ray",
      "category": "distributed",
      "vlm": false,
      "blurb": "Distributed compute for data loading, training, and batch multimodal inference, the orchestration layer for scaling VLM pipelines across a cluster."
    },
    {
      "name": "MLflow",
      "repo": "mlflow/mlflow",
      "url": "https://github.com/mlflow/mlflow",
      "category": "distributed",
      "vlm": false,
      "blurb": "Experiment tracking, model registry, and artifact logging for reproducible large-scale training and eval runs."
    },
    {
      "name": "nuScenes devkit",
      "repo": "nutonomy/nuscenes-devkit",
      "url": "https://github.com/nutonomy/nuscenes-devkit",
      "category": "dataset",
      "vlm": false,
      "blurb": "Official devkit for the multimodal nuScenes AD dataset (camera, lidar, radar), the standard toolkit for sensor+video data loading and eval."
    },
    {
      "name": "Waymo Open Dataset",
      "repo": "waymo-research/waymo-open-dataset",
      "url": "https://github.com/waymo-research/waymo-open-dataset",
      "category": "dataset",
      "vlm": false,
      "blurb": "Large-scale AD perception/motion/end-to-end datasets with eval code, a primary source of camera+lidar video for driving models."
    },
    {
      "name": "BDD100K",
      "repo": "SysCV/bdd100k-models",
      "url": "https://github.com/SysCV/bdd100k-models",
      "category": "dataset",
      "vlm": false,
      "blurb": "Model zoo and tooling for the diverse BDD100K driving-video dataset, useful for detection/segmentation/tracking baselines and labels."
    },
    {
      "name": "Ego4D",
      "repo": "facebookresearch/Ego4d",
      "url": "https://github.com/facebookresearch/Ego4d",
      "category": "dataset",
      "vlm": false,
      "blurb": "Massive egocentric video dataset with download, feature-extraction, and API tooling, relevant for first-person video understanding and robotics."
    },
    {
      "name": "Argoverse 2",
      "repo": "argoverse/av2-api",
      "url": "https://github.com/argoverse/av2-api",
      "category": "dataset",
      "vlm": false,
      "blurb": "Next-gen self-driving datasets (sensor, lidar, motion forecasting) with a maintained Python API and HD maps for multimodal AD research."
    },
    {
      "name": "JAX",
      "repo": "jax-ml/jax",
      "url": "https://github.com/jax-ml/jax",
      "category": "jax",
      "vlm": false,
      "blurb": "Composable NumPy with autodiff, XLA compilation, and pmap/shard_map, the base for large-scale research training on TPUs/GPUs."
    },
    {
      "name": "Flax",
      "repo": "google/flax",
      "url": "https://github.com/google/flax",
      "category": "jax",
      "vlm": false,
      "blurb": "Neural-network library for JAX (the NNX API) used across DeepMind/Google research models, including many multimodal architectures."
    },
    {
      "name": "Keras",
      "repo": "keras-team/keras",
      "url": "https://github.com/keras-team/keras",
      "category": "jax",
      "vlm": false,
      "blurb": "Multi-backend (JAX / TensorFlow / PyTorch) high-level API, handy for portable model code across the three frameworks this role expects."
    },
    {
      "name": "MaxText",
      "repo": "AI-Hypercomputer/maxtext",
      "url": "https://github.com/AI-Hypercomputer/maxtext",
      "category": "jax",
      "vlm": false,
      "blurb": "High-performance, scalable JAX LLM reference (Google) for TPU/GPU pods, a clean example of large-scale distributed training in JAX."
    },
    {
      "name": "Levanter",
      "repo": "marin-community/levanter",
      "url": "https://github.com/marin-community/levanter",
      "category": "jax",
      "vlm": false,
      "blurb": "JAX/Equinox framework for legible, scalable, reproducible foundation-model training with bitwise determinism across hardware."
    },
    {
      "name": "Penzai",
      "repo": "google-deepmind/penzai",
      "url": "https://github.com/google-deepmind/penzai",
      "category": "jax",
      "vlm": false,
      "blurb": "DeepMind JAX toolkit for building and visualizing/interpreting models as legible pytrees, useful for research-grade experimentation."
    },
    {
      "name": "AI-Scientist",
      "repo": "SakanaAI/AI-Scientist",
      "url": "https://github.com/SakanaAI/AI-Scientist",
      "category": "research",
      "vlm": true,
      "blurb": "Runs a full loop that generates ideas, writes and executes experiment code, plots results, and drafts a paper with an automated reviewer."
    },
    {
      "name": "AI-Scientist-v2",
      "repo": "SakanaAI/AI-Scientist-v2",
      "url": "https://github.com/SakanaAI/AI-Scientist-v2",
      "category": "research",
      "vlm": true,
      "blurb": "End-to-end agentic system using progressive agentic tree search and VLM feedback on figures to produce workshop-level manuscripts."
    },
    {
      "name": "Agent Laboratory",
      "repo": "SamuelSchmidgall/AgentLaboratory",
      "url": "https://github.com/SamuelSchmidgall/AgentLaboratory",
      "category": "research",
      "vlm": false,
      "blurb": "Multi-agent pipeline that takes a human research idea through literature review, experimentation, and report writing."
    },
    {
      "name": "STORM",
      "repo": "stanford-oval/storm",
      "url": "https://github.com/stanford-oval/storm",
      "category": "research",
      "vlm": false,
      "blurb": "LLM knowledge-curation system that researches a topic via multi-perspective question asking and writes a cited, Wikipedia-style report."
    },
    {
      "name": "GPT-Researcher",
      "repo": "assafelovic/gpt-researcher",
      "url": "https://github.com/assafelovic/gpt-researcher",
      "category": "research",
      "vlm": false,
      "blurb": "Autonomous agent that plans sub-queries, searches and scrapes sources, and synthesizes a cited research report."
    },
    {
      "name": "deep-research",
      "repo": "dzhng/deep-research",
      "url": "https://github.com/dzhng/deep-research",
      "category": "research",
      "vlm": false,
      "blurb": "Minimal agent that runs iterative search-and-reason loops with configurable breadth and depth to produce a report."
    },
    {
      "name": "smolagents",
      "repo": "huggingface/smolagents",
      "url": "https://github.com/huggingface/smolagents",
      "category": "research",
      "vlm": false,
      "blurb": "Barebones code-acting agent library; its examples include Hugging Face's open reproduction of Deep Research."
    },
    {
      "name": "ADAS",
      "repo": "ShengranHu/ADAS",
      "url": "https://github.com/ShengranHu/ADAS",
      "category": "research",
      "vlm": false,
      "blurb": "Meta-agent that iteratively programs and evaluates new agent designs in code, automating the search over agentic systems."
    },
    {
      "name": "DSPy",
      "repo": "stanfordnlp/dspy",
      "url": "https://github.com/stanfordnlp/dspy",
      "category": "research",
      "vlm": false,
      "blurb": "Define LLM pipelines as modules and optimize their prompts/weights against a metric rather than hand-prompting — the rigor layer for agent pipelines."
    },
    {
      "name": "AI-Researcher",
      "repo": "HKUDS/AI-Researcher",
      "url": "https://github.com/HKUDS/AI-Researcher",
      "category": "research",
      "vlm": false,
      "blurb": "Automates the research pipeline from literature analysis through algorithm implementation to paper generation."
    },
    {
      "name": "Curie",
      "repo": "Just-Curieous/Curie",
      "url": "https://github.com/Just-Curieous/Curie",
      "category": "research",
      "vlm": false,
      "blurb": "Experimentation agent that enforces methodological rigor (controlled setup, reproducibility) when running and analyzing experiments."
    },
    {
      "name": "MLAgentBench",
      "repo": "snap-stanford/MLAgentBench",
      "url": "https://github.com/snap-stanford/MLAgentBench",
      "category": "eval",
      "vlm": false,
      "blurb": "Benchmark of end-to-end ML experimentation tasks for measuring how well agents can improve models from a starting codebase."
    },
    {
      "name": "Aviary",
      "repo": "Future-House/aviary",
      "url": "https://github.com/Future-House/aviary",
      "category": "eval",
      "vlm": false,
      "blurb": "Gym-style environment framework for training and evaluating language agents on challenging scientific tasks."
    },
    {
      "name": "PaperQA",
      "repo": "Future-House/paper-qa",
      "url": "https://github.com/Future-House/paper-qa",
      "category": "retrieval",
      "vlm": false,
      "blurb": "Retrieval-augmented QA engine that answers questions over scientific PDFs with grounded in-text citations."
    },
    {
      "name": "Verifiers",
      "repo": "PrimeIntellect-ai/verifiers",
      "url": "https://github.com/PrimeIntellect-ai/verifiers",
      "category": "rl",
      "slm": true,
      "blurb": "Framework (on TRL) for multi-turn RL with verifiable rewards; v1 adds DAG-branching environments that exceed the model context window."
    },
    {
      "name": "SkyRL",
      "repo": "NovaSky-AI/SkyRL",
      "url": "https://github.com/NovaSky-AI/SkyRL",
      "category": "rl",
      "slm": true,
      "blurb": "Berkeley's flexible RL library focused on multi-turn, long-horizon agentic training."
    },
    {
      "name": "Open-AgentRL (RLAnything / AutoTool)",
      "repo": "Gen-Verse/Open-AgentRL",
      "url": "https://github.com/Gen-Verse/Open-AgentRL",
      "category": "rl",
      "slm": true,
      "blurb": "Open RL for LLMs + agentic scenarios (ICML 2026); RLAnything closed-loop-optimizes each component of the training pipeline."
    },
    {
      "name": "InternBootcamp",
      "repo": "InternLM/InternBootcamp",
      "url": "https://github.com/InternLM/InternBootcamp",
      "category": "rl",
      "slm": true,
      "blurb": "Scalable framework of 1000+ verifiable reasoning tasks (code, logic, games) behind one interface for RL-with-verifiable-rewards."
    },
    {
      "name": "Gymnasium (Farama)",
      "repo": "Farama-Foundation/Gymnasium",
      "url": "https://github.com/Farama-Foundation/Gymnasium",
      "category": "rl",
      "slm": false,
      "blurb": "The maintained successor to OpenAI Gym — the standard environment API most RL training stacks (incl. RLlib) build on."
    },
    {
      "name": "MORPHEUS evals (Skyfall)",
      "repo": "Skyfall-Research/morpheus-evals",
      "url": "https://github.com/Skyfall-Research/morpheus-evals",
      "category": "eval",
      "slm": false,
      "blurb": "Open eval code for MORPHEUS, a persistent enterprise simulation for CONTINUAL RL: no episode resets, structured non-stationarity (failure-injection + config shifts), composite verifier reward."
    },
    {
      "name": "BoTorch",
      "repo": "meta-pytorch/botorch",
      "url": "https://github.com/meta-pytorch/botorch",
      "category": "bayesopt",
      "slm": false,
      "blurb": "The reference PyTorch-native Bayesian-optimization library — Monte-Carlo acquisition (qEI/qNEI), multi-objective (qEHVI/qNEHVI), constrained + batch/async BO; the exact stack the Merge JD names."
    },
    {
      "name": "Ax (Adaptive Experimentation)",
      "repo": "facebook/Ax",
      "url": "https://github.com/facebook/Ax",
      "category": "bayesopt",
      "slm": false,
      "blurb": "High-level adaptive-experimentation platform over BoTorch — orchestrates sequential DBTL campaigns with a service API, multi-objective goals, and constraints."
    },
    {
      "name": "GPyTorch",
      "repo": "cornellius-gp/gpytorch",
      "url": "https://github.com/cornellius-gp/gpytorch",
      "category": "bayesopt",
      "slm": false,
      "blurb": "Scalable Gaussian-process inference in PyTorch — the surrogate engine under BoTorch; core to probabilistic modeling and UQ over sparse, noisy experimental data."
    },
    {
      "name": "Trieste",
      "repo": "secondmind-labs/trieste",
      "url": "https://github.com/secondmind-labs/trieste",
      "category": "bayesopt",
      "slm": false,
      "blurb": "Production-oriented BO toolbox (TensorFlow) with batch, multi-objective, and active-learning loops — a clean reference for closed-loop design patterns."
    },
    {
      "name": "scikit-optimize",
      "repo": "scikit-optimize/scikit-optimize",
      "url": "https://github.com/scikit-optimize/scikit-optimize",
      "category": "bayesopt",
      "slm": false,
      "blurb": "Lightweight sequential model-based optimization with a scipy-style API — the classic BO baseline (now archived/read-only, cite as legacy)."
    },
    {
      "name": "HEBO",
      "repo": "huawei-noah/HEBO",
      "url": "https://github.com/huawei-noah/HEBO",
      "category": "bayesopt",
      "slm": false,
      "blurb": "Heteroscedastic, robust BO that won the NeurIPS 2020 black-box optimization challenge — strong on noisy real-world objectives (bundles MCBO, NAP)."
    },
    {
      "name": "Optuna",
      "repo": "optuna/optuna",
      "url": "https://github.com/optuna/optuna",
      "category": "bayesopt",
      "slm": false,
      "blurb": "The most-adopted define-by-run black-box optimizer with pruning and dashboards — a strong engineering reference and baseline."
    },
    {
      "name": "Pyro",
      "repo": "pyro-ppl/pyro",
      "url": "https://github.com/pyro-ppl/pyro",
      "category": "probabilistic",
      "slm": false,
      "blurb": "Deep probabilistic programming on PyTorch (SVI, MCMC) — the JD's named tool for probabilistic modeling and uncertainty quantification."
    },
    {
      "name": "NumPyro",
      "repo": "pyro-ppl/numpyro",
      "url": "https://github.com/pyro-ppl/numpyro",
      "category": "probabilistic",
      "slm": false,
      "blurb": "JAX-backed Pyro — fast NUTS/HMC and SVI for Bayesian models on GPU/TPU when sampling speed matters."
    },
    {
      "name": "Laplace",
      "repo": "aleximmer/Laplace",
      "url": "https://github.com/aleximmer/Laplace",
      "category": "probabilistic",
      "slm": false,
      "blurb": "Post-hoc Laplace approximations for neural nets — cheap Bayesian uncertainty over deep representation-learning models."
    },
    {
      "name": "Uncertainty Toolbox",
      "repo": "uncertainty-toolbox/uncertainty-toolbox",
      "url": "https://github.com/uncertainty-toolbox/uncertainty-toolbox",
      "category": "probabilistic",
      "slm": false,
      "blurb": "Metrics, calibration, and visualization for predictive uncertainty — validates that a surrogate's error bars are trustworthy."
    },
    {
      "name": "GAUCHE",
      "repo": "leojklarner/gauche",
      "url": "https://github.com/leojklarner/gauche",
      "category": "bio-ml",
      "slm": false,
      "blurb": "A Gaussian-process library for chemistry (30+ molecule/reaction/protein kernels) that plugs into GPyTorch/BoTorch — the most on-point repo for BO over molecules."
    },
    {
      "name": "RDKit",
      "repo": "rdkit/rdkit",
      "url": "https://github.com/rdkit/rdkit",
      "category": "bio-ml",
      "slm": false,
      "blurb": "The standard cheminformatics toolkit (descriptors, fingerprints, conformers) — the featurization layer for any molecular surrogate."
    },
    {
      "name": "REINVENT 4",
      "repo": "MolecularAI/REINVENT4",
      "url": "https://github.com/MolecularAI/REINVENT4",
      "category": "bio-ml",
      "slm": false,
      "blurb": "RL-based de novo molecular design (scaffold hopping, R-group, linker) — the canonical 'RL frameworks guiding molecular engineering' reference from AstraZeneca."
    },
    {
      "name": "Chemprop",
      "repo": "chemprop/chemprop",
      "url": "https://github.com/chemprop/chemprop",
      "category": "bio-ml",
      "slm": false,
      "blurb": "Message-passing neural networks for molecular property prediction — a leading learned surrogate for objective functions in a DBTL loop."
    },
    {
      "name": "DeepChem",
      "repo": "deepchem/deepchem",
      "url": "https://github.com/deepchem/deepchem",
      "category": "bio-ml",
      "slm": false,
      "blurb": "Batteries-included deep learning for drug discovery, materials, and quantum chemistry — featurizers, models, and dataset splitters."
    },
    {
      "name": "Therapeutics Data Commons (PyTDC)",
      "repo": "mims-harvard/TDC",
      "url": "https://github.com/mims-harvard/TDC",
      "category": "bio-ml",
      "slm": false,
      "blurb": "Curated therapeutic ML datasets and benchmarks with a Python API — ready-made sparse/noisy bio tasks for prototyping active-learning loops."
    },
    {
      "name": "SELFIES",
      "repo": "aspuru-guzik-group/selfies",
      "url": "https://github.com/aspuru-guzik-group/selfies",
      "category": "bio-ml",
      "slm": false,
      "blurb": "A 100%-robust molecular string representation — enables valid generative and latent-space optimization of molecules."
    },
    {
      "name": "Olympus",
      "repo": "aspuru-guzik-group/olympus",
      "url": "https://github.com/aspuru-guzik-group/olympus",
      "category": "bio-ml",
      "slm": false,
      "blurb": "Benchmark surfaces and planners for experiment planning / self-driving labs — simulate a closed loop without wet-lab cost."
    },
    {
      "name": "Atlas",
      "repo": "aspuru-guzik-group/atlas",
      "url": "https://github.com/aspuru-guzik-group/atlas",
      "category": "bio-ml",
      "slm": false,
      "blurb": "A 'brain for self-driving labs' — BO over categorical/discrete/mixed spaces with multi-objective and constraints; tightly matched to the JD."
    },
    {
      "name": "ESM (EvolutionaryScale)",
      "repo": "evolutionaryscale/esm",
      "url": "https://github.com/evolutionaryscale/esm",
      "category": "bio-ml",
      "slm": false,
      "blurb": "The current ESM3/ESMC generative protein language models — sequence representation learning for protein design campaigns."
    },
    {
      "name": "BioNeMo Framework",
      "repo": "NVIDIA/bionemo-framework",
      "url": "https://github.com/NVIDIA/bionemo-framework",
      "category": "bio-ml",
      "slm": false,
      "blurb": "NVIDIA's scalable framework for biomolecular foundation models (MolMIM latent generative model, ESM-2) — property-guided molecule optimization at scale."
    },
    {
      "name": "Mamba",
      "repo": "state-spaces/mamba",
      "url": "https://github.com/state-spaces/mamba",
      "category": "bayesopt",
      "slm": true,
      "blurb": "Selective state-space architecture (Mamba/Mamba-2) — linear-time long-sequence modeling; the JD's state-space nice-to-have, relevant to neural time-series."
    },
    {
      "name": "S4 (Structured State Spaces)",
      "repo": "state-spaces/s4",
      "url": "https://github.com/state-spaces/s4",
      "category": "bayesopt",
      "slm": true,
      "blurb": "The foundational structured state-space sequence models (S4/HiPPO/DSS) — efficient modeling of 10k+ step dependencies in signals."
    },
    {
      "name": "MNE-Python",
      "repo": "mne-tools/mne-python",
      "url": "https://github.com/mne-tools/mne-python",
      "category": "neuro",
      "slm": false,
      "blurb": "The standard toolkit for analyzing human neurophysiology data (MEG, EEG, iEEG) — the data layer for non-invasive brain-signal modeling."
    },
    {
      "name": "Braindecode",
      "repo": "braindecode/braindecode",
      "url": "https://github.com/braindecode/braindecode",
      "category": "neuro",
      "slm": false,
      "blurb": "Deep learning for EEG/neural-signal decoding on PyTorch — models, datasets, and training loops for brain-computer-interface decoding."
    },
    {
      "name": "Neural Latents Benchmark (nlb_tools)",
      "repo": "neurallatents/nlb_tools",
      "url": "https://github.com/neurallatents/nlb_tools",
      "category": "neuro",
      "slm": false,
      "blurb": "Tools for the Neural Latents Benchmark — evaluating latent-variable models of neural population dynamics; the eval harness for neural representation learning."
    },
    {
      "name": "Tinker Cookbook",
      "repo": "thinking-machines-lab/tinker-cookbook",
      "url": "https://github.com/thinking-machines-lab/tinker-cookbook",
      "category": "finetuning",
      "slm": true,
      "blurb": "Apache-2.0 post-training recipe library on the Tinker fine-tuning API — SFT, DPO, RLHF, and custom RL environments composed from four low-level training primitives over LoRA adapters."
    }
  ],
  "courses": [
    {
      "title": "Neural Networks: Zero to Hero",
      "org": "Eureka Labs",
      "instructor": "Andrej Karpathy",
      "year": 2023,
      "free": true,
      "url": "https://karpathy.ai/zero-to-hero.html",
      "topic": "foundations",
      "blurb": "Code-along video series building neural nets from backprop up to a GPT, following Attention Is All You Need and GPT-2/3."
    },
    {
      "title": "CS224N: NLP with Deep Learning",
      "org": "Stanford",
      "instructor": "Christopher Manning",
      "year": 2024,
      "free": true,
      "url": "https://web.stanford.edu/class/cs224n/",
      "topic": "foundations",
      "blurb": "Foundational NLP-with-deep-learning course covering word vectors, attention, transformers, and pretraining; lecture videos are public."
    },
    {
      "title": "CS 11-711: Advanced NLP",
      "org": "Carnegie Mellon University",
      "instructor": "Graham Neubig",
      "year": 2024,
      "free": true,
      "url": "https://www.phontron.com/class/anlp-fall2024/",
      "topic": "foundations",
      "blurb": "Graduate NLP course rebuilt around LLMs, including a build-your-own-LLaMa assignment; slides and videos are public."
    },
    {
      "title": "6.S191: Introduction to Deep Learning",
      "org": "MIT",
      "instructor": "Alexander Amini, Ava Soleimany",
      "year": 2025,
      "free": true,
      "url": "https://introtodeeplearning.com/",
      "topic": "foundations",
      "blurb": "Fast-paced intro to deep learning with labs, now including large language models and generative AI."
    },
    {
      "title": "The Full Stack LLM Bootcamp",
      "org": "Full Stack Deep Learning",
      "instructor": "Charles Frye, Sergey Karayev, Josh Tobin",
      "year": 2023,
      "free": true,
      "url": "https://fullstackdeeplearning.com/llm-bootcamp/",
      "topic": "foundations",
      "blurb": "Recorded bootcamp on building LLM applications: prompt engineering, LLMOps, augmented models, and shipping an app."
    },
    {
      "title": "Generative AI with Large Language Models",
      "org": "DeepLearning.AI & AWS",
      "instructor": "Antje Barth, Chris Fregly, et al.",
      "year": 2023,
      "free": true,
      "url": "https://www.deeplearning.ai/courses/generative-ai-with-llms/",
      "topic": "foundations",
      "blurb": "Covers the LLM lifecycle: pretraining, scaling laws, instruction tuning, and RLHF (free to audit on Coursera)."
    },
    {
      "title": "CS336: Language Modeling from Scratch",
      "org": "Stanford",
      "instructor": "Percy Liang, Tatsunori Hashimoto",
      "year": 2025,
      "free": true,
      "url": "https://cs336.stanford.edu/spring2025/",
      "topic": "pretraining",
      "blurb": "Implementation-heavy course that builds a language model end to end: tokenization, transformer, training, systems, scaling, data, and alignment."
    },
    {
      "title": "LLM101n: Let's build a Storyteller",
      "org": "Eureka Labs",
      "instructor": "Andrej Karpathy",
      "year": 2024,
      "free": true,
      "url": "https://github.com/karpathy/LLM101n",
      "topic": "pretraining",
      "blurb": "Public syllabus/repo (in development) for building a Storyteller LLM end to end in Python, C, and CUDA."
    },
    {
      "title": "Post-training of LLMs",
      "org": "DeepLearning.AI",
      "instructor": "Banghua Zhu",
      "year": 2025,
      "free": true,
      "url": "https://www.deeplearning.ai/courses/post-training-of-llms/",
      "topic": "posttraining",
      "blurb": "When and how to apply SFT, DPO, and online RL, including data curation for post-training."
    },
    {
      "title": "Fine-tuning & RL for LLMs: Intro to Post-training",
      "org": "DeepLearning.AI (with AMD)",
      "instructor": "Sharon Zhou",
      "year": 2025,
      "free": true,
      "url": "https://www.deeplearning.ai/courses/fine-tuning-and-reinforcement-learning-for-llms-intro-to-post-training/",
      "topic": "posttraining",
      "blurb": "Covers fine-tuning, reward modeling, RLHF, and RL algorithms (PPO, GRPO) for shaping behavior and reasoning."
    },
    {
      "title": "Hugging Face LLM Course",
      "org": "Hugging Face",
      "instructor": "Hugging Face team",
      "year": 2024,
      "free": true,
      "url": "https://huggingface.co/learn/llm-course",
      "topic": "finetuning",
      "blurb": "Free hands-on course on transformers, tokenizers, fine-tuning pretrained models, and building LLM applications."
    },
    {
      "title": "CS234: Reinforcement Learning",
      "org": "Stanford",
      "instructor": "Emma Brunskill",
      "year": 2024,
      "free": true,
      "url": "https://web.stanford.edu/class/cs234/",
      "topic": "rl",
      "blurb": "Graduate RL course spanning tabular methods, deep RL, policy gradients, and the basics of RL from human feedback."
    },
    {
      "title": "Deep Reinforcement Learning Course",
      "org": "Hugging Face",
      "instructor": "Thomas Simonini",
      "year": 2023,
      "free": true,
      "url": "https://huggingface.co/learn/deep-rl-course",
      "topic": "rl",
      "blurb": "Free self-paced Deep RL course with practical training in Stable-Baselines3, CleanRL, and Sample Factory; optional certificate."
    },
    {
      "title": "Reinforcement Learning from Human Feedback",
      "org": "DeepLearning.AI",
      "instructor": "Nikita Namjoshi",
      "year": 2024,
      "free": true,
      "url": "https://learn.deeplearning.ai/courses/reinforcement-learning-from-human-feedback/",
      "topic": "rl",
      "blurb": "Short course on the RLHF pipeline, tuning an open model with reward and preference data."
    },
    {
      "title": "Reinforcement Fine-Tuning LLMs with GRPO",
      "org": "DeepLearning.AI (with Predibase)",
      "instructor": "Travis Addair, Arnav Garg",
      "year": 2025,
      "free": true,
      "url": "https://www.deeplearning.ai/short-courses/reinforcement-fine-tuning-llms-grpo/",
      "topic": "rl",
      "blurb": "Short course on using GRPO with programmable reward functions to improve LLM reasoning."
    },
    {
      "title": "CS294/194-196: Large Language Model Agents",
      "org": "UC Berkeley",
      "instructor": "Dawn Song, Xinyun Chen",
      "year": 2024,
      "free": true,
      "url": "https://rdi.berkeley.edu/llm-agents/f24",
      "topic": "agents",
      "blurb": "MOOC-available course on LLM agent foundations, reasoning, tool use, and applications, with frontier-lab guest lectures."
    },
    {
      "title": "CS294/194-280: Advanced LLM Agents",
      "org": "UC Berkeley",
      "instructor": "Dawn Song, Xinyun Chen",
      "year": 2025,
      "free": true,
      "url": "https://rdi.berkeley.edu/adv-llm-agents/sp25",
      "topic": "agents",
      "blurb": "Spring 2025 follow-on covering advanced agent reasoning, math/theorem-proving, code generation, and safety."
    },
    {
      "title": "CS231n: Deep Learning for Computer Vision",
      "org": "Stanford",
      "instructor": "Fei-Fei Li, Ehsan Adeli et al.",
      "year": 2024,
      "free": true,
      "url": "https://cs231n.stanford.edu/",
      "topic": "vision",
      "blurb": "The canonical intro to CNNs and visual recognition; slides, notes, and assignments are public."
    },
    {
      "title": "CS231A: Computer Vision, From 3D Reconstruction to Recognition",
      "org": "Stanford",
      "instructor": "Silvio Savarese / Jeannette Bohg (staff)",
      "year": 2025,
      "free": true,
      "url": "https://web.stanford.edu/class/cs231a/",
      "topic": "vision",
      "blurb": "Geometric CV: camera models, epipolar/stereo geometry, depth and scene flow, 6D pose and tracking, with public notes."
    },
    {
      "title": "EECS 498/598: Deep Learning for Computer Vision",
      "org": "University of Michigan",
      "instructor": "Justin Johnson",
      "year": 2020,
      "free": true,
      "url": "https://web.eecs.umich.edu/~justincj/teaching/eecs498/",
      "topic": "vision",
      "blurb": "From-scratch deep learning for vision (CNNs, attention, detection, segmentation) with the full lecture set on YouTube."
    },
    {
      "title": "11-777: Multimodal Machine Learning",
      "org": "Carnegie Mellon (LTI)",
      "instructor": "Louis-Philippe Morency, Paul Liang",
      "year": 2023,
      "free": true,
      "url": "https://cmu-mmml.github.io/",
      "topic": "multimodal",
      "blurb": "Organizes multimodal ML around six challenges (representation, alignment, reasoning, generation, transference, quantification); lectures on YouTube."
    },
    {
      "title": "Community Computer Vision Course",
      "org": "Hugging Face",
      "instructor": "HF community",
      "year": 2024,
      "free": true,
      "url": "https://huggingface.co/learn/computer-vision-course/unit0/welcome/welcome",
      "topic": "vision",
      "blurb": "Free hands-on course from classical CV through ViTs, multimodal and generative vision, with runnable notebooks."
    },
    {
      "title": "CS25: Transformers United",
      "org": "Stanford",
      "instructor": "Student-led seminar",
      "year": 2025,
      "free": true,
      "url": "https://web.stanford.edu/class/cs25/",
      "topic": "vlm",
      "blurb": "Guest-lecture seminar including multimodal / vision-language and world-modeling sessions; open to audit, posted on YouTube."
    },
    {
      "title": "DeepRob: Deep Learning for Robot Perception",
      "org": "University of Michigan",
      "instructor": "Chad Jenkins, Anthony Opipari, Xiaoxiao Du",
      "year": 2024,
      "free": true,
      "url": "https://deeprob.org/w24/",
      "topic": "vision",
      "blurb": "Deep-learning-for-vision adapted to robot perception and manipulation, then reproducing recent perception papers; public slides."
    },
    {
      "title": "Practical Deep Learning for Coders",
      "org": "fast.ai",
      "instructor": "Jeremy Howard",
      "year": 2022,
      "free": true,
      "url": "https://course.fast.ai/",
      "topic": "vision",
      "blurb": "Code-first, top-down deep learning across vision, NLP, and fine-tuning in PyTorch/fastai; free lessons plus the free book."
    },
    {
      "title": "Getting Started with FiftyOne (Visual AI)",
      "org": "Voxel51",
      "instructor": "Voxel51",
      "year": 2025,
      "free": true,
      "url": "https://voxel51.com/get-started",
      "topic": "video",
      "blurb": "Free tutorials on curating, visualizing, and debugging image/video/3D datasets and model outputs — directly relevant to AD/robotics data ops."
    },
    {
      "title": "Bayesian Optimization (the book)",
      "instructor": "Roman Garnett",
      "org": "Washington University in St. Louis",
      "year": 2023,
      "free": true,
      "url": "https://bayesoptbook.com/",
      "topic": "bayesopt",
      "blurb": "The most complete modern treatment of BO theory and practice — full PDF free online."
    },
    {
      "title": "Gaussian Process Summer School (GPSS)",
      "org": "University of Sheffield (community)",
      "free": true,
      "url": "https://gpss.cc/",
      "topic": "bayesopt",
      "blurb": "Annual lectures plus notebooks on GPs, deep GPs, and Bayesian optimization; slides and videos archived per year."
    },
    {
      "title": "Gaussian Processes for Machine Learning (GPML)",
      "instructor": "Rasmussen & Williams",
      "org": "MIT Press",
      "year": 2006,
      "free": true,
      "url": "http://gaussianprocess.org/gpml/",
      "topic": "bayesopt",
      "blurb": "The definitive GP textbook (regression, classification, covariance functions, model selection) — free full PDF."
    },
    {
      "title": "BoTorch Tutorials + Ax Docs",
      "org": "Meta",
      "free": true,
      "url": "https://botorch.org/docs/tutorials/",
      "topic": "bayesopt",
      "blurb": "Runnable walkthroughs of qEI/qNEI/qEHVI/qNEHVI, constrained and multi-objective BO — closest to the day-to-day toolkit for this role."
    },
    {
      "title": "Probabilistic Machine Learning (Books 1 & 2)",
      "instructor": "Kevin Murphy",
      "org": "Google",
      "free": true,
      "url": "https://probml.github.io/pml-book/",
      "topic": "probabilistic",
      "blurb": "Free draft PDFs plus Colab notebooks — strong coverage of probabilistic modeling, GPs, and uncertainty quantification."
    },
    {
      "title": "DeepChem Tutorials",
      "org": "DeepChem",
      "free": true,
      "url": "https://deepchem.io/tutorials/",
      "topic": "molecular",
      "blurb": "Hands-on notebooks for ML on molecules — featurization, property prediction, and generative design."
    },
    {
      "title": "Neuromatch Academy — Computational Neuroscience",
      "org": "Neuromatch",
      "free": true,
      "url": "https://compneuro.neuromatch.io/",
      "topic": "neuroscience",
      "blurb": "Free, hands-on computational-neuroscience curriculum (modeling, dynamics, decoding) — the on-ramp to the neuroscience nice-to-have."
    }
  ],
  "papers": [
    {
      "title": "SmolLM2: When Smol Goes Big — Data-Centric Training of a Small Language Model",
      "authors": "Ben Allal et al.",
      "org": "Hugging Face",
      "year": 2025,
      "venue": "arXiv:2502.02737",
      "url": "https://arxiv.org/abs/2502.02737",
      "topic": "slm",
      "blurb": "A 1.7B model trained via careful data curation and overtraining, with the dataset recipe documented openly."
    },
    {
      "title": "A Survey of Small Language Models",
      "authors": "Van Nguyen et al.",
      "org": "multi-institution",
      "year": 2024,
      "venue": "arXiv:2410.20011",
      "url": "https://arxiv.org/abs/2410.20011",
      "topic": "slm",
      "blurb": "Taxonomy of SLM architectures, training, and compression methods, plus benchmark datasets and evaluation metrics."
    },
    {
      "title": "Small Language Models: Survey, Measurements, and Insights",
      "authors": "Lu et al.",
      "org": "multi-institution",
      "year": 2024,
      "venue": "arXiv:2409.15790",
      "url": "https://arxiv.org/abs/2409.15790",
      "topic": "slm",
      "blurb": "Empirical survey measuring capabilities and on-device runtime cost across a large set of released SLMs."
    },
    {
      "title": "Textbooks Are All You Need (phi-1)",
      "authors": "Gunasekar et al.",
      "org": "Microsoft",
      "year": 2023,
      "venue": "arXiv:2306.11644",
      "url": "https://arxiv.org/abs/2306.11644",
      "topic": "slm",
      "blurb": "phi-1 (1.3B), showing high-quality synthetic 'textbook' data lets small models match far larger ones on HumanEval."
    },
    {
      "title": "Textbooks Are All You Need II: phi-1.5 Technical Report",
      "authors": "Li et al.",
      "org": "Microsoft",
      "year": 2023,
      "venue": "arXiv:2309.05463",
      "url": "https://arxiv.org/abs/2309.05463",
      "topic": "slm",
      "blurb": "Extends the textbook-data approach to a 1.3B general reasoning model competitive with models 5x its size."
    },
    {
      "title": "Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone",
      "authors": "Abdin et al.",
      "org": "Microsoft",
      "year": 2024,
      "venue": "arXiv:2404.14219",
      "url": "https://arxiv.org/abs/2404.14219",
      "topic": "slm",
      "blurb": "phi-3-mini (3.8B) rivals Mixtral 8x7B and GPT-3.5 on benchmarks while being small enough to run on a phone."
    },
    {
      "title": "Gemma: Open Models Based on Gemini Research and Technology",
      "authors": "Mesnard et al.",
      "org": "Google DeepMind",
      "year": 2024,
      "venue": "arXiv:2403.08295",
      "url": "https://arxiv.org/abs/2403.08295",
      "topic": "slm",
      "blurb": "Open 2B and 7B models derived from Gemini research, released with pretrained and instruction-tuned checkpoints."
    },
    {
      "title": "Gemma 2: Improving Open Language Models at a Practical Size",
      "authors": "Gemma Team",
      "org": "Google DeepMind",
      "year": 2024,
      "venue": "arXiv:2408.00118",
      "url": "https://arxiv.org/abs/2408.00118",
      "topic": "slm",
      "blurb": "Trains the 2B/9B models with knowledge distillation over next-token prediction for strong quality at small size."
    },
    {
      "title": "MobileLLM: Optimizing Sub-billion Parameter Language Models for On-Device Use Cases",
      "authors": "Liu et al.",
      "org": "Meta",
      "year": 2024,
      "venue": "arXiv:2402.14905 (ICML 2024)",
      "url": "https://arxiv.org/abs/2402.14905",
      "topic": "slm",
      "blurb": "Shows deep-and-thin architecture, embedding sharing, and grouped-query attention matter most below 1B parameters."
    },
    {
      "title": "TinyLlama: An Open-Source Small Language Model",
      "authors": "Zhang et al.",
      "org": "SUTD",
      "year": 2024,
      "venue": "arXiv:2401.02385",
      "url": "https://arxiv.org/abs/2401.02385",
      "topic": "slm",
      "blurb": "A 1.1B Llama-architecture model pretrained on ~3T tokens with a fully open training pipeline."
    },
    {
      "title": "TinyStories: How Small Can Language Models Be and Still Speak Coherent English?",
      "authors": "Eldan & Li",
      "org": "Microsoft Research",
      "year": 2023,
      "venue": "arXiv:2305.07759",
      "url": "https://arxiv.org/abs/2305.07759",
      "topic": "slm",
      "blurb": "Sub-10M-parameter models trained on a constrained synthetic story corpus can generate coherent, consistent English."
    },
    {
      "title": "Distilling Step-by-Step! Outperforming Larger LMs with Less Data and Smaller Sizes",
      "authors": "Hsieh et al.",
      "org": "Google",
      "year": 2023,
      "venue": "arXiv:2305.02301 (Findings of ACL 2023)",
      "url": "https://arxiv.org/abs/2305.02301",
      "topic": "slm",
      "blurb": "Uses LLM-generated rationales as extra supervision so small models beat much larger ones with less data."
    },
    {
      "title": "The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale",
      "authors": "Penedo et al.",
      "org": "Hugging Face",
      "year": 2024,
      "venue": "arXiv:2406.17557 (NeurIPS 2024)",
      "url": "https://arxiv.org/abs/2406.17557",
      "topic": "pretraining",
      "blurb": "Ablates deduplication and filtering to build a 15T-token open web corpus plus the FineWeb-Edu subset."
    },
    {
      "title": "The Llama 3 Herd of Models",
      "authors": "Dubey et al.",
      "org": "Meta",
      "year": 2024,
      "venue": "arXiv:2407.21783",
      "url": "https://arxiv.org/abs/2407.21783",
      "topic": "pretraining",
      "blurb": "Documents the pretraining, scaling, and post-training of the Llama 3 family, including the 405B dense flagship."
    },
    {
      "title": "GPT-4 Technical Report",
      "authors": "OpenAI",
      "org": "OpenAI",
      "year": 2023,
      "venue": "arXiv:2303.08774",
      "url": "https://arxiv.org/abs/2303.08774",
      "topic": "pretraining",
      "blurb": "A multimodal transformer with human-level exam performance and predictable scaling from small proxy models."
    },
    {
      "title": "DeepSeek-V3 Technical Report",
      "authors": "DeepSeek-AI",
      "org": "DeepSeek",
      "year": 2024,
      "venue": "arXiv:2412.19437",
      "url": "https://arxiv.org/abs/2412.19437",
      "topic": "pretraining",
      "blurb": "A 671B MoE (37B active) with MLA, auxiliary-loss-free load balancing, and multi-token prediction, trained efficiently."
    },
    {
      "title": "Scaling Laws for Neural Language Models",
      "authors": "Kaplan et al.",
      "org": "OpenAI",
      "year": 2020,
      "venue": "arXiv:2001.08361",
      "url": "https://arxiv.org/abs/2001.08361",
      "topic": "scaling",
      "blurb": "Establishes power-law relationships between loss and model size, data, and compute across many orders of magnitude."
    },
    {
      "title": "Training Compute-Optimal Large Language Models (Chinchilla)",
      "authors": "Hoffmann et al.",
      "org": "DeepMind",
      "year": 2022,
      "venue": "arXiv:2203.15556 (NeurIPS 2022)",
      "url": "https://arxiv.org/abs/2203.15556",
      "topic": "scaling",
      "blurb": "The Chinchilla result: model size and training tokens should scale equally; most large models are undertrained."
    },
    {
      "title": "Training Language Models to Follow Instructions with Human Feedback (InstructGPT)",
      "authors": "Ouyang et al.",
      "org": "OpenAI",
      "year": 2022,
      "venue": "arXiv:2203.02155 (NeurIPS 2022)",
      "url": "https://arxiv.org/abs/2203.02155",
      "topic": "posttraining",
      "blurb": "Introduces the SFT + reward model + RLHF recipe; a 1.3B tuned model was preferred over 175B GPT-3."
    },
    {
      "title": "Direct Preference Optimization: Your Language Model Is Secretly a Reward Model",
      "authors": "Rafailov et al.",
      "org": "Stanford",
      "year": 2023,
      "venue": "arXiv:2305.18290 (NeurIPS 2023)",
      "url": "https://arxiv.org/abs/2305.18290",
      "topic": "posttraining",
      "blurb": "Replaces the RLHF reward model and PPO loop with a single classification loss on preference pairs."
    },
    {
      "title": "Constitutional AI: Harmlessness from AI Feedback",
      "authors": "Bai et al.",
      "org": "Anthropic",
      "year": 2022,
      "venue": "arXiv:2212.08073",
      "url": "https://arxiv.org/abs/2212.08073",
      "topic": "posttraining",
      "blurb": "Trains a harmless assistant using AI-generated critiques and preferences guided by written principles (RLAIF)."
    },
    {
      "title": "RLAIF: Scaling RLHF with AI Feedback",
      "authors": "Lee et al.",
      "org": "Google",
      "year": 2023,
      "venue": "arXiv:2309.00267 (ICML 2024)",
      "url": "https://arxiv.org/abs/2309.00267",
      "topic": "posttraining",
      "blurb": "Shows LLM-generated preference labels can match human-labeled RLHF across summarization and dialogue tasks."
    },
    {
      "title": "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models",
      "authors": "Wei et al.",
      "org": "Google",
      "year": 2022,
      "venue": "arXiv:2201.11903 (NeurIPS 2022)",
      "url": "https://arxiv.org/abs/2201.11903",
      "topic": "posttraining",
      "blurb": "Intermediate reasoning steps in prompts unlock arithmetic, commonsense, and symbolic reasoning at scale."
    },
    {
      "title": "Training a Helpful and Harmless Assistant with RLHF",
      "authors": "Bai et al.",
      "org": "Anthropic",
      "year": 2022,
      "venue": "arXiv:2204.05862",
      "url": "https://arxiv.org/abs/2204.05862",
      "topic": "rl",
      "blurb": "Applies preference modeling and iterated online RLHF to align an assistant, analyzing the reward/KL trade-off."
    },
    {
      "title": "DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models",
      "authors": "Shao et al.",
      "org": "DeepSeek",
      "year": 2024,
      "venue": "arXiv:2402.03300",
      "url": "https://arxiv.org/abs/2402.03300",
      "topic": "rl",
      "blurb": "Introduces GRPO, a critic-free RL algorithm using group-relative advantages, later central to DeepSeek-R1."
    },
    {
      "title": "DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning",
      "authors": "DeepSeek-AI",
      "org": "DeepSeek",
      "year": 2025,
      "venue": "arXiv:2501.12948 (Nature, 2025)",
      "url": "https://arxiv.org/abs/2501.12948",
      "topic": "rl",
      "blurb": "Elicits reasoning purely via RL (R1-Zero) and distills it into smaller dense models — a key SLM-reasoning recipe."
    },
    {
      "title": "LoRA: Low-Rank Adaptation of Large Language Models",
      "authors": "Hu et al.",
      "org": "Microsoft",
      "year": 2021,
      "venue": "arXiv:2106.09685 (ICLR 2022)",
      "url": "https://arxiv.org/abs/2106.09685",
      "topic": "peft",
      "blurb": "Freezes base weights and trains injected low-rank matrices, cutting trainable parameters by orders of magnitude."
    },
    {
      "title": "QLoRA: Efficient Finetuning of Quantized LLMs",
      "authors": "Dettmers et al.",
      "org": "University of Washington",
      "year": 2023,
      "venue": "arXiv:2305.14314 (NeurIPS 2023)",
      "url": "https://arxiv.org/abs/2305.14314",
      "topic": "peft",
      "blurb": "Backpropagates through a frozen 4-bit (NF4) model into LoRA adapters, finetuning a 65B model on one 48GB GPU."
    },
    {
      "title": "Distilling the Knowledge in a Neural Network",
      "authors": "Hinton et al.",
      "org": "Google",
      "year": 2015,
      "venue": "arXiv:1503.02531",
      "url": "https://arxiv.org/abs/1503.02531",
      "topic": "distillation",
      "blurb": "Foundational knowledge distillation: training a small student on the soft targets of a larger teacher."
    },
    {
      "title": "Sequence-Level Knowledge Distillation",
      "authors": "Kim & Rush",
      "org": "Harvard",
      "year": 2016,
      "venue": "arXiv:1606.07947 (EMNLP 2016)",
      "url": "https://arxiv.org/abs/1606.07947",
      "topic": "distillation",
      "blurb": "Extends distillation from token-level to sequence-level, yielding student NMT models ~10x faster with little loss."
    },
    {
      "title": "DistilBERT: Smaller, Faster, Cheaper and Lighter",
      "authors": "Sanh et al.",
      "org": "Hugging Face",
      "year": 2019,
      "venue": "arXiv:1910.01108",
      "url": "https://arxiv.org/abs/1910.01108",
      "topic": "distillation",
      "blurb": "Distills BERT during pretraining to a 40% smaller model that retains ~97% of language understanding."
    },
    {
      "title": "GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers",
      "authors": "Frantar et al.",
      "org": "IST Austria",
      "year": 2022,
      "venue": "arXiv:2210.17323 (ICLR 2023)",
      "url": "https://arxiv.org/abs/2210.17323",
      "topic": "compression",
      "blurb": "One-shot 3-4 bit weight quantization using approximate second-order information, no retraining required."
    },
    {
      "title": "AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration",
      "authors": "Lin et al.",
      "org": "MIT",
      "year": 2023,
      "venue": "arXiv:2306.00978 (MLSys 2024)",
      "url": "https://arxiv.org/abs/2306.00978",
      "topic": "compression",
      "blurb": "Protects a small fraction of salient weights via activation-aware scaling for accurate low-bit quantization."
    },
    {
      "title": "LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale",
      "authors": "Dettmers et al.",
      "org": "University of Washington",
      "year": 2022,
      "venue": "arXiv:2208.07339 (NeurIPS 2022)",
      "url": "https://arxiv.org/abs/2208.07339",
      "topic": "compression",
      "blurb": "Int8 inference for billion-scale transformers with no accuracy loss by isolating emergent outlier features."
    },
    {
      "title": "SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shot",
      "authors": "Frantar & Alistarh",
      "org": "IST Austria",
      "year": 2023,
      "venue": "arXiv:2301.00774 (ICML 2023)",
      "url": "https://arxiv.org/abs/2301.00774",
      "topic": "compression",
      "blurb": "Prunes GPT-scale models to 50-60% sparsity in one shot without retraining and minimal perplexity increase."
    },
    {
      "title": "Learning Transferable Visual Models From Natural Language Supervision (CLIP)",
      "authors": "Radford et al.",
      "org": "OpenAI",
      "year": 2021,
      "venue": "ICML 2021 · arXiv:2103.00020",
      "url": "https://arxiv.org/abs/2103.00020",
      "topic": "vlm",
      "blurb": "Contrastively pretrains dual image/text encoders on 400M web pairs, enabling zero-shot classification and the embedding backbone most VLMs still build on."
    },
    {
      "title": "Flamingo: a Visual Language Model for Few-Shot Learning",
      "authors": "Alayrac et al.",
      "org": "DeepMind",
      "year": 2022,
      "venue": "NeurIPS 2022 · arXiv:2204.14198",
      "url": "https://arxiv.org/abs/2204.14198",
      "topic": "vlm",
      "blurb": "Bridges frozen vision and language backbones with gated cross-attention over interleaved image/video-text for in-context few-shot multimodal learning."
    },
    {
      "title": "BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Encoders and LLMs",
      "authors": "Li et al.",
      "org": "Salesforce Research",
      "year": 2023,
      "venue": "ICML 2023 · arXiv:2301.12597",
      "url": "https://arxiv.org/abs/2301.12597",
      "topic": "vlm",
      "blurb": "Introduces the lightweight Q-Former connecting a frozen vision encoder to a frozen LLM — a cheap recipe for adding vision to language models."
    },
    {
      "title": "Visual Instruction Tuning (LLaVA)",
      "authors": "Liu et al.",
      "org": "UW-Madison / Microsoft Research",
      "year": 2023,
      "venue": "NeurIPS 2023 · arXiv:2304.08485",
      "url": "https://arxiv.org/abs/2304.08485",
      "topic": "vlm",
      "blurb": "Uses GPT-4-generated image-instruction data to instruction-tune a CLIP-encoder-plus-LLM model, establishing the dominant open VLM training pattern."
    },
    {
      "title": "Improved Baselines with Visual Instruction Tuning (LLaVA-1.5)",
      "authors": "Liu et al.",
      "org": "UW-Madison / Microsoft Research",
      "year": 2023,
      "venue": "CVPR 2024 · arXiv:2310.03744",
      "url": "https://arxiv.org/abs/2310.03744",
      "topic": "vlm",
      "blurb": "MLP connector, higher-res CLIP, and academic VQA data hit SOTA on 11 benchmarks with ~1.2M public samples and one day on 8×A100."
    },
    {
      "title": "Qwen2-VL: Perception of the World at Any Resolution",
      "authors": "Wang et al. (Qwen Team)",
      "org": "Alibaba",
      "year": 2024,
      "venue": "arXiv:2409.12191",
      "url": "https://arxiv.org/abs/2409.12191",
      "topic": "vlm",
      "blurb": "Adds naive dynamic resolution and multimodal RoPE (M-RoPE) to handle arbitrary image sizes and video under one paradigm across 2B/8B/72B."
    },
    {
      "title": "Qwen2.5-VL Technical Report",
      "authors": "Qwen Team",
      "org": "Alibaba",
      "year": 2025,
      "venue": "arXiv:2502.13923",
      "url": "https://arxiv.org/abs/2502.13923",
      "topic": "vlm",
      "blurb": "Absolute-time encoding for hour-long video, strong bbox/point grounding and document parsing — a strong fine-tuning base for driving/robotics."
    },
    {
      "title": "InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks",
      "authors": "Chen et al.",
      "org": "Shanghai AI Lab / OpenGVLab",
      "year": 2023,
      "venue": "CVPR 2024 · arXiv:2312.14238",
      "url": "https://arxiv.org/abs/2312.14238",
      "topic": "vlm",
      "blurb": "Scales the vision encoder to 6B params and progressively aligns it to an LLM, reaching SOTA across 32 image/video perception and retrieval benchmarks."
    },
    {
      "title": "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Detection",
      "authors": "Liu et al.",
      "org": "IDEA Research",
      "year": 2023,
      "venue": "ECCV 2024 · arXiv:2303.05499",
      "url": "https://arxiv.org/abs/2303.05499",
      "topic": "multimodal",
      "blurb": "Fuses a DINO detector with language so free-text prompts localize arbitrary objects — a workhorse for open-vocabulary grounding/auto-labeling."
    },
    {
      "title": "Video-LLaMA: Instruction-tuned Audio-Visual Language Model for Video Understanding",
      "authors": "Zhang, Li, Bing",
      "org": "DAMO Academy, Alibaba",
      "year": 2023,
      "venue": "EMNLP 2023 Demo · arXiv:2306.02858",
      "url": "https://arxiv.org/abs/2306.02858",
      "topic": "video",
      "blurb": "Couples a video Q-Former and audio branch to an LLM to instruction-tune joint audio-visual temporal understanding of video."
    },
    {
      "title": "Video-LLaVA: Learning United Visual Representation by Alignment Before Projection",
      "authors": "Lin et al.",
      "org": "Peking University",
      "year": 2023,
      "venue": "EMNLP 2024 · arXiv:2311.10122",
      "url": "https://arxiv.org/abs/2311.10122",
      "topic": "video",
      "blurb": "Aligns image and video features into a shared space before the LLM projector, letting one model train jointly on images and video with mutual gains."
    },
    {
      "title": "ViViT: A Video Vision Transformer",
      "authors": "Arnab et al.",
      "org": "Google Research",
      "year": 2021,
      "venue": "ICCV 2021 · arXiv:2103.15691",
      "url": "https://arxiv.org/abs/2103.15691",
      "topic": "video",
      "blurb": "Pure-transformer video classifier with factorized spatial/temporal attention variants to keep long token sequences tractable."
    },
    {
      "title": "VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training",
      "authors": "Tong et al.",
      "org": "Nanjing University / Tencent",
      "year": 2022,
      "venue": "NeurIPS 2022 · arXiv:2203.12602",
      "url": "https://arxiv.org/abs/2203.12602",
      "topic": "video",
      "blurb": "90-95% tube masking makes MAE-style self-supervised video pretraining work even on a few thousand clips — a practical video backbone recipe."
    },
    {
      "title": "Is Space-Time Attention All You Need for Video Understanding? (TimeSformer)",
      "authors": "Bertasius, Wang, Torresani",
      "org": "Facebook AI / Dartmouth",
      "year": 2021,
      "venue": "ICML 2021 · arXiv:2102.05095",
      "url": "https://arxiv.org/abs/2102.05095",
      "topic": "video",
      "blurb": "Convolution-free video model; divided space-then-time attention gives the best accuracy/cost trade-off for action recognition."
    },
    {
      "title": "InternVideo: General Video Foundation Models via Generative and Discriminative Learning",
      "authors": "Wang et al.",
      "org": "Shanghai AI Lab / OpenGVLab",
      "year": 2022,
      "venue": "arXiv:2212.03191",
      "url": "https://arxiv.org/abs/2212.03191",
      "topic": "video",
      "blurb": "Combines masked video modeling with video-language contrastive learning into one foundation model hitting SOTA on 39 video datasets."
    },
    {
      "title": "Scaling Up Visual and Vision-Language Representation Learning With Noisy Text Supervision (ALIGN)",
      "authors": "Jia et al.",
      "org": "Google",
      "year": 2021,
      "venue": "ICML 2021 · arXiv:2102.05918",
      "url": "https://arxiv.org/abs/2102.05918",
      "topic": "retrieval",
      "blurb": "A simple dual-encoder trained on 1B+ noisy alt-text pairs (no cleaning) sets SOTA image-text retrieval, validating scale over curation."
    },
    {
      "title": "Sigmoid Loss for Language Image Pre-Training (SigLIP)",
      "authors": "Zhai et al.",
      "org": "Google DeepMind",
      "year": 2023,
      "venue": "ICCV 2023 · arXiv:2303.15343",
      "url": "https://arxiv.org/abs/2303.15343",
      "topic": "retrieval",
      "blurb": "A pairwise sigmoid loss needing no global normalization improves small-batch training and large-scale embedding learning."
    },
    {
      "title": "LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Models",
      "authors": "Zhang, Li, Liu et al.",
      "org": "NTU (LMMs-Lab)",
      "year": 2024,
      "venue": "arXiv:2407.12772",
      "url": "https://arxiv.org/abs/2407.12772",
      "topic": "eval",
      "blurb": "Unified 50+-task evaluation framework (plus a Lite variant and contamination-resistant LiveBench) — the harness to standardize your own VLM/video evals."
    },
    {
      "title": "Video-MME: Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis",
      "authors": "Fu et al.",
      "org": "multi-institution",
      "year": 2024,
      "venue": "CVPR 2025 · arXiv:2405.21075",
      "url": "https://arxiv.org/abs/2405.21075",
      "topic": "eval",
      "blurb": "Full-spectrum video benchmark spanning 11s-to-1h clips across 6 domains with subtitle/audio modalities, the standard for long-video VLM eval."
    },
    {
      "title": "MMBench: Is Your Multi-modal Model an All-around Player?",
      "authors": "Liu et al.",
      "org": "Shanghai AI Lab / CUHK / NTU",
      "year": 2023,
      "venue": "ECCV 2024 · arXiv:2307.06281",
      "url": "https://arxiv.org/abs/2307.06281",
      "topic": "eval",
      "blurb": "Fine-grained bilingual VLM benchmark with a CircularEval protocol that shuffles answer order to reduce position bias in scoring."
    },
    {
      "title": "MMMU: Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark",
      "authors": "Yue et al.",
      "org": "multi-institution",
      "year": 2023,
      "venue": "CVPR 2024 · arXiv:2311.16502",
      "url": "https://arxiv.org/abs/2311.16502",
      "topic": "eval",
      "blurb": "11.5K college-level questions over 30 subjects and heterogeneous image types, probing knowledge-heavy expert reasoning where even strong VLMs score low."
    },
    {
      "title": "MLAgentBench: Evaluating Language Agents on Machine Learning Experimentation",
      "authors": "Huang et al.",
      "org": "Stanford",
      "year": 2023,
      "venue": "arXiv:2310.03302",
      "url": "https://arxiv.org/abs/2310.03302",
      "topic": "eval",
      "blurb": "13 ML experimentation tasks and metrics for measuring agents that iteratively edit code to improve model performance."
    },
    {
      "title": "The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery",
      "authors": "Lu et al.",
      "org": "Sakana AI",
      "year": 2024,
      "venue": "arXiv:2408.06292",
      "url": "https://arxiv.org/abs/2408.06292",
      "topic": "research",
      "blurb": "A pipeline where an LLM generates ideas, runs code experiments, and writes and reviews papers within a fixed compute budget."
    },
    {
      "title": "The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search",
      "authors": "Yamada et al.",
      "org": "Sakana AI",
      "year": 2025,
      "venue": "arXiv:2504.08066",
      "url": "https://arxiv.org/abs/2504.08066",
      "topic": "research",
      "blurb": "Agentic tree search plus VLM figure review; one generated manuscript passed workshop peer review above the human acceptance bar."
    },
    {
      "title": "Towards an AI Co-Scientist",
      "authors": "Gottweis et al.",
      "org": "Google / DeepMind",
      "year": 2025,
      "venue": "arXiv:2502.18864",
      "url": "https://arxiv.org/abs/2502.18864",
      "topic": "research",
      "blurb": "Multi-agent system on Gemini that generates and debates hypotheses, validated on drug repurposing and antimicrobial-resistance tasks."
    },
    {
      "title": "ResearchAgent: Iterative Research Idea Generation over Scientific Literature",
      "authors": "Baek et al.",
      "org": "KAIST / Microsoft Research",
      "year": 2024,
      "venue": "NAACL 2025 · arXiv:2404.07738",
      "url": "https://arxiv.org/abs/2404.07738",
      "topic": "research",
      "blurb": "Generates problems, methods, and experiment designs from literature, refined by reviewing agents with human-aligned feedback."
    },
    {
      "title": "Autonomous chemical research with large language models (Coscientist)",
      "authors": "Boiko et al.",
      "org": "Carnegie Mellon University",
      "year": 2023,
      "venue": "Nature 624, 570-578",
      "url": "https://www.nature.com/articles/s41586-023-06792-0",
      "topic": "research",
      "blurb": "A GPT-4 system with search/code/lab-automation tools that designs and physically executes chemistry experiments."
    },
    {
      "title": "Automated Design of Agentic Systems (ADAS)",
      "authors": "Hu, Lu, Clune",
      "org": "UBC / Vector Institute",
      "year": 2024,
      "venue": "ICLR 2025 · arXiv:2408.08435",
      "url": "https://arxiv.org/abs/2408.08435",
      "topic": "research",
      "blurb": "Frames agent design as a search problem where a meta-agent writes code to invent new agents, outperforming hand-designed baselines."
    },
    {
      "title": "Assisting in Writing Wikipedia-like Articles From Scratch with LLMs (STORM)",
      "authors": "Shao et al.",
      "org": "Stanford OVAL",
      "year": 2024,
      "venue": "NAACL 2024 · arXiv:2402.14207",
      "url": "https://arxiv.org/abs/2402.14207",
      "topic": "research",
      "blurb": "Simulate multi-perspective question-asking to build an outline, then retrieve sources and write a grounded, cited article."
    },
    {
      "title": "A Survey on Large Language Model based Autonomous Agents",
      "authors": "Wang et al.",
      "org": "Renmin University of China",
      "year": 2023,
      "venue": "arXiv:2308.11432",
      "url": "https://arxiv.org/abs/2308.11432",
      "topic": "research",
      "blurb": "Systematic survey of LLM-agent construction, application domains, and evaluation — context for research-loop agents."
    },
    {
      "title": "Gödel Agent: A Self-Referential Agent Framework for Recursive Self-Improvement",
      "authors": "Yin et al.",
      "org": "Peking University / UC Santa Barbara",
      "year": 2025,
      "venue": "ACL 2025 · arXiv:2410.04444",
      "url": "https://arxiv.org/abs/2410.04444",
      "topic": "research",
      "blurb": "An agent that reads and rewrites its own logic at runtime to recursively self-improve without a fixed optimization routine."
    },
    {
      "title": "Agent Laboratory: Using LLM Agents as Research Assistants",
      "authors": "Schmidgall et al.",
      "org": "AMD / Johns Hopkins University",
      "year": 2025,
      "venue": "arXiv:2501.04227",
      "url": "https://arxiv.org/abs/2501.04227",
      "topic": "research",
      "blurb": "Autonomous framework covering literature review, experimentation, and report writing, with human feedback improving output quality."
    },
    {
      "title": "Curie: Toward Rigorous and Automated Scientific Experimentation with AI Agents",
      "authors": "Kon et al.",
      "org": "University of Michigan",
      "year": 2025,
      "venue": "arXiv:2502.16069",
      "url": "https://arxiv.org/abs/2502.16069",
      "topic": "research",
      "blurb": "Adds rigor components — controlled procedure, reproducibility, result interpretation — to agent-run experimentation."
    },
    {
      "title": "AI-Researcher: Autonomous Scientific Innovation",
      "authors": "Tang et al.",
      "org": "University of Hong Kong",
      "year": 2025,
      "venue": "arXiv:2505.18705",
      "url": "https://arxiv.org/abs/2505.18705",
      "topic": "research",
      "blurb": "A fully autonomous system spanning idea generation, algorithm implementation, and manuscript writing, released with an open codebase."
    },
    {
      "title": "Robin: A Multi-Agent System for Automating Scientific Discovery",
      "authors": "Ghareeb et al.",
      "org": "FutureHouse",
      "year": 2025,
      "venue": "arXiv:2505.13400",
      "url": "https://arxiv.org/abs/2505.13400",
      "topic": "research",
      "blurb": "Couples literature-search and data-analysis agents to run a discovery loop; identified ripasudil as a candidate for dry AMD."
    },
    {
      "title": "MORPHEUS: A Persistent Enterprise Simulation for Continual Reinforcement Learning",
      "authors": "Skyfall AI (+ Toronto / Georgia Tech / Alberta)",
      "org": "Skyfall AI",
      "year": 2026,
      "venue": "OpenReview 2026",
      "url": "https://morpheus.skyfall.ai/",
      "topic": "rl",
      "blurb": "Big-world CRL benchmark with structured non-stationarity; PPO/HER/EWC/LCM all show a ~1.0 settled-state gap — no family durably learns without resets."
    },
    {
      "title": "The Landscape of Agentic Reinforcement Learning for LLMs: A Survey",
      "authors": "multi-institution",
      "org": "multi-institution",
      "year": 2025,
      "venue": "arXiv:2509.02547",
      "url": "https://arxiv.org/abs/2509.02547",
      "topic": "rl",
      "blurb": "Map of agentic RL for LLMs — environments, verifiable rewards, multi-turn credit assignment, and the open frameworks implementing them."
    },
    {
      "title": "Reinforcement Learning Foundations for Deep Research Systems: A Survey",
      "authors": "multi-institution",
      "org": "multi-institution",
      "year": 2025,
      "venue": "arXiv:2509.06733",
      "url": "https://arxiv.org/abs/2509.06733",
      "topic": "rl",
      "blurb": "How RL post-training underpins long-horizon research agents: reward design, tool-use credit assignment, and training infrastructure."
    },
    {
      "title": "A Tutorial on Bayesian Optimization",
      "authors": "Peter Frazier",
      "org": "Cornell",
      "year": 2018,
      "venue": "arXiv:1807.02811",
      "url": "https://arxiv.org/abs/1807.02811",
      "topic": "bayesopt",
      "blurb": "The single best self-contained intro — GP regression plus EI, entropy search, and knowledge gradient. Read this first."
    },
    {
      "title": "Efficient Global Optimization of Expensive Black-Box Functions (EGO)",
      "authors": "Jones, Schonlau & Welch",
      "org": "—",
      "year": 1998,
      "venue": "J. Global Optimization 13:455",
      "url": "https://doi.org/10.1023/A:1008306431147",
      "topic": "bayesopt",
      "blurb": "The paper that launched surrogate-based BO: a Kriging/GP surrogate plus Expected Improvement to optimize costly black boxes."
    },
    {
      "title": "Practical Bayesian Optimization of Machine Learning Algorithms",
      "authors": "Snoek, Larochelle & Adams",
      "org": "Toronto / Harvard",
      "year": 2012,
      "venue": "NeurIPS · arXiv:1206.2944",
      "url": "https://arxiv.org/abs/1206.2944",
      "topic": "bayesopt",
      "blurb": "The modern reboot of BO for ML (the 'Spearmint' paper): GP priors, integrated acquisition, and cost-aware EI."
    },
    {
      "title": "Taking the Human Out of the Loop: A Review of Bayesian Optimization",
      "authors": "Shahriari, Swersky, Wang, Adams & de Freitas",
      "org": "multi-institution",
      "year": 2016,
      "venue": "Proc. IEEE 104(1)",
      "url": "https://doi.org/10.1109/JPROC.2015.2494218",
      "topic": "bayesopt",
      "blurb": "The canonical survey tying together surrogates, acquisition functions, and applications of BO."
    },
    {
      "title": "Gaussian Process Optimization in the Bandit Setting (GP-UCB)",
      "authors": "Srinivas, Krause, Kakade & Seeger",
      "org": "Caltech / ETH",
      "year": 2010,
      "venue": "ICML · arXiv:0912.3995",
      "url": "https://arxiv.org/abs/0912.3995",
      "topic": "bayesopt",
      "blurb": "Introduces GP-UCB with the first sublinear cumulative-regret bounds — the theoretical backbone of exploration in BO."
    },
    {
      "title": "Max-value Entropy Search for Efficient Bayesian Optimization (MES)",
      "authors": "Wang & Jegelka",
      "org": "MIT",
      "year": 2017,
      "venue": "ICML · arXiv:1703.01968",
      "url": "https://arxiv.org/abs/1703.01968",
      "topic": "bayesopt",
      "blurb": "A cheaper, more robust information-theoretic acquisition that targets info about the optimal value rather than the arg-max."
    },
    {
      "title": "BoTorch: A Framework for Efficient Monte-Carlo Bayesian Optimization",
      "authors": "Balandat, Karrer, Jiang, Daulton, Letham, Wilson & Bakshy",
      "org": "Meta",
      "year": 2020,
      "venue": "NeurIPS · arXiv:1910.06403",
      "url": "https://arxiv.org/abs/1910.06403",
      "topic": "bayesopt",
      "blurb": "The reference open-source BO library — Monte-Carlo acquisition (qEI/qNEI) with autodiff and sample-average approximation."
    },
    {
      "title": "Constrained Bayesian Optimization with Noisy Experiments (qNEI)",
      "authors": "Letham, Karrer, Ottoni & Bakshy",
      "org": "Meta",
      "year": 2019,
      "venue": "Bayesian Analysis 14(2) · arXiv:1706.07094",
      "url": "https://arxiv.org/abs/1706.07094",
      "topic": "bayesopt",
      "blurb": "Noisy, constrained expected improvement via quasi-Monte-Carlo integration — the basis of Meta's Ax and directly on-point for noisy DBTL assays."
    },
    {
      "title": "Differentiable Expected Hypervolume Improvement for Multi-Objective BO (qEHVI)",
      "authors": "Daulton, Balandat & Bakshy",
      "org": "Meta",
      "year": 2020,
      "venue": "NeurIPS · arXiv:2006.05078",
      "url": "https://arxiv.org/abs/2006.05078",
      "topic": "bayesopt",
      "blurb": "Exact-gradient, parallelizable hypervolume-improvement acquisition — state of the art for multi-objective BO."
    },
    {
      "title": "Parallel BO of Multiple Noisy Objectives with EHVI (qNEHVI)",
      "authors": "Daulton, Balandat & Bakshy",
      "org": "Meta",
      "year": 2021,
      "venue": "NeurIPS · arXiv:2105.08195",
      "url": "https://arxiv.org/abs/2105.08195",
      "topic": "bayesopt",
      "blurb": "Extends qEHVI to noisy observations with polynomial batch scaling — the current default for noisy multi-objective DBTL problems."
    },
    {
      "title": "Scalable Global Optimization via Local Bayesian Optimization (TuRBO)",
      "authors": "Eriksson, Pearce, Gardner, Turner & Poloczek",
      "org": "Uber AI",
      "year": 2019,
      "venue": "NeurIPS · arXiv:1910.01739",
      "url": "https://arxiv.org/abs/1910.01739",
      "topic": "bayesopt",
      "blurb": "Trust-region local GPs with bandit sample allocation — the workhorse for high-dimensional, high-budget BO."
    },
    {
      "title": "High-Dimensional BO with Sparse Axis-Aligned Subspaces (SAASBO)",
      "authors": "Eriksson & Jankowiak",
      "org": "Meta",
      "year": 2021,
      "venue": "UAI · arXiv:2103.00349",
      "url": "https://arxiv.org/abs/2103.00349",
      "topic": "bayesopt",
      "blurb": "Strong sparsity priors plus HMC to find the few relevant dimensions — excellent in the low-data, high-dim regime typical of experiments."
    },
    {
      "title": "Active Learning Literature Survey",
      "authors": "Burr Settles",
      "org": "UW–Madison",
      "year": 2009,
      "venue": "CS Tech Report 1648",
      "url": "https://minds.wisconsin.edu/handle/1793/60660",
      "topic": "active-learning",
      "blurb": "The canonical survey of query strategies — uncertainty sampling, query-by-committee, expected error reduction."
    },
    {
      "title": "Bayesian Active Learning for Classification and Preference Learning (BALD)",
      "authors": "Houlsby, Huszár, Ghahramani & Lengyel",
      "org": "Cambridge",
      "year": 2011,
      "venue": "arXiv:1112.5745",
      "url": "https://arxiv.org/abs/1112.5745",
      "topic": "active-learning",
      "blurb": "Introduces BALD (maximize mutual information / model disagreement) — foundational for modern deep and Bayesian active learning."
    },
    {
      "title": "A Tutorial on Thompson Sampling",
      "authors": "Russo, Van Roy, Kazerouni, Osband & Wen",
      "org": "Stanford / DeepMind",
      "year": 2018,
      "venue": "FnT ML 11(1) · arXiv:1707.02038",
      "url": "https://arxiv.org/abs/1707.02038",
      "topic": "active-learning",
      "blurb": "The definitive reference for posterior-sampling exploration, widely used as a scalable BO acquisition strategy."
    },
    {
      "title": "Deep Kernel Learning",
      "authors": "Wilson, Hu, Salakhutdinov & Xing",
      "org": "CMU",
      "year": 2016,
      "venue": "AISTATS · arXiv:1511.02222",
      "url": "https://arxiv.org/abs/1511.02222",
      "topic": "representation",
      "blurb": "Combines neural-net feature learning with GP kernels — representation learning that preserves calibrated GP uncertainty."
    },
    {
      "title": "Predictive Entropy Search for Efficient Global Optimization (PES)",
      "authors": "Hernández-Lobato, Hoffman & Ghahramani",
      "org": "Cambridge / Harvard",
      "year": 2014,
      "venue": "NeurIPS · arXiv:1406.2541",
      "url": "https://arxiv.org/abs/1406.2541",
      "topic": "uncertainty",
      "blurb": "Information-theoretic acquisition that picks points maximizing information gain about the location of the optimum."
    },
    {
      "title": "Automatic Chemical Design Using a Data-Driven Continuous Representation of Molecules",
      "authors": "Gómez-Bombarelli et al.",
      "org": "Harvard / Toronto",
      "year": 2018,
      "venue": "ACS Central Science · arXiv:1610.02415",
      "url": "https://arxiv.org/abs/1610.02415",
      "topic": "molecular",
      "blurb": "The chemical VAE: a continuous latent space of molecules enabling gradient-based and Bayesian optimization of properties."
    },
    {
      "title": "GAUCHE: A Library for Gaussian Processes in Chemistry",
      "authors": "Griffiths et al.",
      "org": "multi-institution",
      "year": 2023,
      "venue": "NeurIPS · arXiv:2212.04450",
      "url": "https://arxiv.org/abs/2212.04450",
      "topic": "molecular",
      "blurb": "GPs and BO over molecular graphs, strings, fingerprints, reactions, and proteins — interfaces directly with GPyTorch/BoTorch."
    },
    {
      "title": "Olympus: A Benchmarking Framework for Noisy Optimization and Experiment Planning",
      "authors": "Häse, Aldeghi & Aspuru-Guzik",
      "org": "Toronto",
      "year": 2021,
      "venue": "Mach. Learn.: Sci. Technol. 2(3) · arXiv:2010.04153",
      "url": "https://arxiv.org/abs/2010.04153",
      "topic": "molecular",
      "blurb": "Realistic noisy experiment-planning benchmarks for autonomous / self-driving labs — a proving ground for closed-loop optimizers."
    },
    {
      "title": "Machine-Learning-Guided Directed Evolution for Protein Engineering",
      "authors": "Yang, Wu & Arnold",
      "org": "Caltech",
      "year": 2019,
      "venue": "Nature Methods 16:687 · arXiv:1811.10775",
      "url": "https://arxiv.org/abs/1811.10775",
      "topic": "molecular",
      "blurb": "The reference framework for ML-in-the-loop directed evolution — sequence–function models guiding the DBTL loop for proteins."
    },
    {
      "title": "Efficiently Modeling Long Sequences with Structured State Spaces (S4)",
      "authors": "Gu, Goel & Ré",
      "org": "Stanford",
      "year": 2022,
      "venue": "ICLR · arXiv:2111.00396",
      "url": "https://arxiv.org/abs/2111.00396",
      "topic": "ssm",
      "blurb": "A structured SSM parameterization that handles 10k+ step dependencies efficiently — relevant to neural time-series."
    },
    {
      "title": "Mamba: Linear-Time Sequence Modeling with Selective State Spaces",
      "authors": "Gu & Dao",
      "org": "CMU / Princeton",
      "year": 2023,
      "venue": "arXiv:2312.00752",
      "url": "https://arxiv.org/abs/2312.00752",
      "topic": "ssm",
      "blurb": "Selective (input-dependent) SSM matching Transformer quality with linear-time scaling — the modern state-space model to know."
    },
    {
      "title": "Inferring Single-Trial Neural Population Dynamics Using Sequential Auto-Encoders (LFADS)",
      "authors": "Pandarinath et al.",
      "org": "Stanford / Emory",
      "year": 2018,
      "venue": "Nature Methods 15:805 · arXiv:1608.06315",
      "url": "https://arxiv.org/abs/1608.06315",
      "topic": "neuroscience",
      "blurb": "Latent dynamical-systems models of neural population activity — foundational for decoding brain signals and a bridge from state-space models to neuroscience."
    },
    {
      "title": "BARE: Leveraging Base Language Models for Few-Shot Synthetic Data Generation",
      "authors": "Zhu, Asawa, Davis, Chen, Hanin, Stoica, Gonzalez, Zaharia",
      "org": "UC Berkeley / Stanford / Princeton",
      "year": 2025,
      "venue": "arXiv:2502.01697",
      "url": "https://arxiv.org/abs/2502.01697",
      "topic": "vlm",
      "blurb": "Base models for diversity, instruction-tuned models for correctness — the two-stage recipe the syndata case study adapts to vision-language data; operationalized in skills/syndata-bare."
    },
    {
      "title": "MemGPT: Towards LLMs as Operating Systems",
      "authors": "Packer, Wooders, Lin, Fang, Patil, Stoica, Gonzalez",
      "org": "UC Berkeley",
      "year": 2023,
      "venue": "arXiv:2310.08560",
      "url": "https://arxiv.org/abs/2310.08560",
      "topic": "personalization",
      "blurb": "Virtual context management — OS-style hierarchical memory for LLMs; the canonical per-user persistent-memory design behind personal models. Operationalized in skills/personalization-loop."
    }
  ],
  "jobs": [
    {
      "name": "Anthropic Careers",
      "type": "company",
      "url": "https://www.anthropic.com/careers",
      "focus": "Frontier lab; alignment, pretraining, RL, and fine-tuning research and engineering roles."
    },
    {
      "name": "OpenAI Careers",
      "type": "company",
      "url": "https://openai.com/careers/",
      "focus": "Frontier lab; research, post-training/RLHF, and applied ML roles."
    },
    {
      "name": "Google DeepMind Careers",
      "type": "company",
      "url": "https://deepmind.google/careers/",
      "focus": "Frontier lab; research scientist and engineering roles across pretraining, RL, and alignment."
    },
    {
      "name": "Mistral AI Careers",
      "type": "company",
      "url": "https://mistral.ai/careers/",
      "focus": "European frontier lab; open-weight model pretraining, fine-tuning, and inference roles."
    },
    {
      "name": "Hugging Face Careers",
      "type": "company",
      "url": "https://apply.workable.com/huggingface/",
      "focus": "Open-source ML platform; ML research/engineering on training, datasets, and model deployment."
    },
    {
      "name": "DeepSeek Talent",
      "type": "company",
      "url": "https://talent.deepseek.com",
      "focus": "AGI-focused lab; pretraining, RL-for-reasoning, and infrastructure roles (official talent portal)."
    },
    {
      "name": "ai-jobs.net",
      "type": "board",
      "url": "https://aijobs.net/",
      "focus": "Large dedicated AI/ML/data-science job board with research and engineering filters."
    },
    {
      "name": "80,000 Hours Job Board",
      "type": "aggregator",
      "url": "https://jobs.80000hours.org/",
      "focus": "Curated board emphasizing frontier-lab and AI-safety roles across pretraining, RL, and alignment."
    },
    {
      "name": "Import AI",
      "type": "newsletter",
      "url": "https://jack-clark.net/",
      "focus": "Weekly research-analysis newsletter (Jack Clark); tracks frontier labs and hiring signals."
    },
    {
      "name": "Ahead of AI",
      "type": "newsletter",
      "url": "https://magazine.sebastianraschka.com/",
      "focus": "Sebastian Raschka's newsletter; deep technical coverage of LLM training and fine-tuning methods."
    },
    {
      "name": "The Batch (DeepLearning.AI)",
      "type": "newsletter",
      "url": "https://www.deeplearning.ai/the-batch",
      "focus": "Weekly AI news/insights from Andrew Ng's team; useful for tracking labs and the talent market."
    },
    {
      "name": "Merge Labs Careers",
      "type": "company",
      "url": "https://jobs.ashbyhq.com/Merge%20Labs",
      "focus": "Frontier BCI lab (bio × AI); hiring ML scientists for Bayesian optimization, RL, and closed-loop molecular/neuro engineering. See the FM-os Merge dossier."
    }
  ],
  "labs": [
    {
      "name": "Thinking Machines Lab",
      "category": "frontier-lab",
      "url": "https://thinkingmachines.ai",
      "github": "https://github.com/thinking-machines-lab",
      "focus": "Frontier lab (Mira Murati); makes Tinker (LoRA fine-tuning API) and the open-weights Inkling multimodal MoE."
    },
    {
      "name": "Unsloth AI",
      "category": "finetuning",
      "url": "https://unsloth.ai",
      "github": "https://github.com/unslothai",
      "focus": "2x-faster, ~70%-less-VRAM LoRA/QLoRA fine-tuning for small models on a single GPU."
    },
    {
      "name": "Together AI",
      "category": "inference-serving",
      "url": "https://together.ai",
      "github": "https://github.com/togethercomputer",
      "focus": "Training + high-throughput inference cloud for open models; an Inkling deployment partner."
    },
    {
      "name": "Fireworks AI",
      "category": "inference-serving",
      "url": "https://fireworks.ai",
      "github": "https://github.com/fw-ai",
      "focus": "Fast, low-cost inference + fine-tuning for open models; an Inkling deployment partner."
    },
    {
      "name": "Baseten",
      "category": "inference-serving",
      "url": "https://baseten.co",
      "github": "https://github.com/basetenlabs",
      "focus": "ML infrastructure for developers (Truss packaging + autoscaling inference); an Inkling deployment partner."
    },
    {
      "name": "Inferact",
      "category": "inference-serving",
      "url": "https://inferact.ai",
      "github": "https://github.com/Inferact",
      "focus": "Commercializing vLLM as a universal inference layer ($150M seed); mission to make inference cheaper and faster."
    },
    {
      "name": "RadixArk",
      "category": "inference-serving",
      "url": "https://github.com/radixark",
      "github": "https://github.com/radixark",
      "focus": "Commercializing SGLang (RadixAttention) as an optimization service; the SGLang counterpart to Inferact's vLLM."
    },
    {
      "name": "Modal",
      "category": "infra",
      "url": "https://modal.com",
      "github": "https://github.com/modal-labs",
      "focus": "Serverless GPU cloud — run training/inference code in the cloud with no infra management; an Inkling deployment partner."
    },
    {
      "name": "Databricks",
      "category": "data-platform",
      "url": "https://databricks.com",
      "github": "https://github.com/databricks",
      "focus": "Data + AI lakehouse (Mosaic AI training, DBRX open model, MLflow); an Inkling deployment partner."
    },
    {
      "name": "LightSeek",
      "category": "open-research",
      "url": "https://lightseek.org",
      "github": "https://github.com/lightseekorg",
      "focus": "Foundation accelerating open research and open-source innovation for next-generation AI systems."
    },
    {
      "name": "Skyfall AI",
      "category": "open-research",
      "url": "https://skyfall.ai",
      "github": "https://github.com/Skyfall-Research",
      "focus": "Enterprise RL company (ex-Maluuba founders); ships MORPHEUS, a persistent, non-resetting enterprise simulation that argues today's LLMs don't durably learn without continual RL."
    },
    {
      "name": "Adaptive ML",
      "category": "finetuning",
      "url": "https://adaptive-ml.com",
      "github": "https://github.com/adaptive-ml",
      "focus": "RLOps platform (Adaptive Engine) for reinforcement-learning post-training + evaluation of open models on enterprise tasks."
    },
    {
      "name": "HUD",
      "category": "data-platform",
      "url": "https://www.hud.ai",
      "focus": "RL-environment platform for enterprise workflows — standardized, reproducible, closed-loop-training-ready environments and benchmarks."
    },
    {
      "name": "CoreWeave",
      "category": "infra",
      "url": "https://www.coreweave.com",
      "github": "https://github.com/coreweave",
      "focus": "GPU cloud with a managed RL-as-a-service offering (serverless infra, frameworks, APIs) for post-training agentic models."
    },
    {
      "name": "Merge Labs",
      "category": "bio-ai",
      "url": "https://merge.io",
      "focus": "Frontier BCI lab bridging biological and artificial intelligence; building a closed-loop DBTL optimization backbone for molecular + neuroengineering discovery (the hiring target)."
    },
    {
      "name": "Meta — Adaptive Experimentation (BoTorch / Ax)",
      "category": "bio-ai",
      "url": "https://botorch.org",
      "github": "https://github.com/meta-pytorch/botorch",
      "focus": "Builds and maintains the core PyTorch Bayesian-optimization + adaptive-experimentation stack this role uses."
    },
    {
      "name": "Matter Lab (Aspuru-Guzik)",
      "category": "bio-ai",
      "url": "https://www.matter.toronto.edu",
      "github": "https://github.com/aspuru-guzik-group",
      "focus": "Self-driving laboratories — robotics + ML + BO for autonomous molecular and materials discovery (SELFIES, Olympus, Atlas)."
    },
    {
      "name": "Acceleration Consortium",
      "category": "bio-ai",
      "url": "https://acceleration.utoronto.ca",
      "focus": "Global consortium standardizing self-driving labs and materials-acceleration platforms for closed-loop discovery."
    },
    {
      "name": "Coley Research Group (MIT)",
      "category": "bio-ai",
      "url": "https://coley.mit.edu",
      "github": "https://github.com/coleygroup",
      "focus": "ML for synthesis planning, molecular design, and autonomous experimentation / DBTL workflows."
    },
    {
      "name": "Recursion",
      "category": "bio-ai",
      "url": "https://www.recursion.com",
      "focus": "Industrial-scale closed-loop drug discovery combining automated wet-lab data generation with ML."
    },
    {
      "name": "Isomorphic Labs",
      "category": "bio-ai",
      "url": "https://www.isomorphiclabs.com",
      "focus": "Alphabet/DeepMind spinout applying AI (AlphaFold lineage) to drug design."
    },
    {
      "name": "NVIDIA BioNeMo",
      "category": "bio-ai",
      "url": "https://www.nvidia.com/en-us/clara/bionemo/",
      "github": "https://github.com/NVIDIA/bionemo-framework",
      "focus": "Foundation-model platform plus generative models (MolMIM) for molecular and protein design at scale."
    }
  ],
  "models": [
    {
      "name": "SmolLM2-1.7B",
      "org": "Hugging Face",
      "params": "1.7B",
      "license": "Apache-2.0",
      "context": "8K",
      "ondevice": true,
      "url": "https://huggingface.co/HuggingFaceTB/SmolLM2-1.7B"
    },
    {
      "name": "Qwen2.5-1.5B",
      "org": "Alibaba (Qwen)",
      "params": "1.5B",
      "license": "Apache-2.0",
      "context": "32K",
      "ondevice": true,
      "url": "https://huggingface.co/Qwen/Qwen2.5-1.5B"
    },
    {
      "name": "Gemma-2-2B",
      "org": "Google",
      "params": "2B",
      "license": "Gemma",
      "context": "8K",
      "ondevice": true,
      "url": "https://huggingface.co/google/gemma-2-2b"
    },
    {
      "name": "Phi-3-mini (3.8B)",
      "org": "Microsoft",
      "params": "3.8B",
      "license": "MIT",
      "context": "4K",
      "ondevice": true,
      "url": "https://huggingface.co/microsoft/Phi-3-mini-4k-instruct"
    },
    {
      "name": "Llama-3.2-1B",
      "org": "Meta",
      "params": "1B",
      "license": "Llama 3.2 Community",
      "context": "128K",
      "ondevice": true,
      "url": "https://huggingface.co/meta-llama/Llama-3.2-1B"
    },
    {
      "name": "Llama-3.2-3B",
      "org": "Meta",
      "params": "3B",
      "license": "Llama 3.2 Community",
      "context": "128K",
      "ondevice": true,
      "url": "https://huggingface.co/meta-llama/Llama-3.2-3B"
    },
    {
      "name": "OLMo-2-1B",
      "org": "Allen Institute for AI",
      "params": "1B",
      "license": "Apache-2.0",
      "context": "4K",
      "ondevice": true,
      "url": "https://huggingface.co/allenai/OLMo-2-0425-1B"
    },
    {
      "name": "H2O-Danube3-4B",
      "org": "H2O.ai",
      "params": "4B",
      "license": "Apache-2.0",
      "context": "8K",
      "ondevice": true,
      "url": "https://huggingface.co/h2oai/h2o-danube3-4b-base"
    },
    {
      "name": "Falcon3-1B",
      "org": "TII",
      "params": "1B",
      "license": "TII Falcon-LLM 2.0",
      "context": "4K",
      "ondevice": true,
      "url": "https://huggingface.co/tiiuae/Falcon3-1B-Base"
    },
    {
      "name": "MiniCPM3-4B",
      "org": "OpenBMB",
      "params": "4B",
      "license": "Apache-2.0 (weights: registration)",
      "context": "32K",
      "ondevice": true,
      "url": "https://huggingface.co/openbmb/MiniCPM3-4B"
    },
    {
      "name": "TinyLlama-1.1B",
      "org": "TinyLlama (community)",
      "params": "1.1B",
      "license": "Apache-2.0",
      "context": "2K",
      "ondevice": true,
      "url": "https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v1.0"
    },
    {
      "name": "MobileLLM-1B",
      "org": "Meta",
      "params": "1B",
      "license": "FAIR Noncommercial Research",
      "nc": true,
      "context": "2K",
      "ondevice": true,
      "url": "https://huggingface.co/facebook/MobileLLM-1B"
    },
    {
      "name": "StableLM-Zephyr-3B",
      "org": "Stability AI",
      "params": "3B",
      "license": "Stability AI Community",
      "nc": true,
      "context": "4K",
      "ondevice": true,
      "url": "https://huggingface.co/stabilityai/stablelm-zephyr-3b"
    }
  ],
  "registry": [
    {
      "name": "slm-quickstart",
      "kind": "skill",
      "path": "skills/slm-quickstart",
      "tags": [
        "finetune",
        "rl"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/slm-quickstart"
    },
    {
      "name": "vlm-quickstart",
      "kind": "skill",
      "path": "skills/vlm-quickstart",
      "tags": [
        "vlm",
        "finetune"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vlm-quickstart"
    },
    {
      "name": "agentic-eval",
      "kind": "skill",
      "path": "skills/agentic-eval",
      "tags": [
        "eval",
        "benchmark"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/agentic-eval"
    },
    {
      "name": "curation-loop",
      "kind": "skill",
      "path": "skills/curation-loop",
      "tags": [
        "curation"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/curation-loop"
    },
    {
      "name": "vector-rag",
      "kind": "skill",
      "path": "skills/vector-rag",
      "tags": [
        "retrieval"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vector-rag"
    },
    {
      "name": "research-loop",
      "kind": "skill",
      "path": "skills/research-loop",
      "tags": [
        "research"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/research-loop"
    },
    {
      "name": "bayesopt-loop",
      "kind": "skill",
      "path": "skills/bayesopt-loop",
      "tags": [
        "bayesopt",
        "research"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/bayesopt-loop"
    },
    {
      "name": "continual-rl-eval",
      "kind": "skill",
      "path": "skills/continual-rl-eval",
      "tags": [
        "rl",
        "eval"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/continual-rl-eval"
    },
    {
      "name": "fm-os",
      "kind": "skill",
      "path": "skills/fm-os",
      "tags": [
        "hub",
        "router"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/fm-os"
    },
    {
      "name": "vlm-failure-probe",
      "kind": "skill",
      "path": "skills/vlm-failure-probe",
      "tags": [
        "vlm",
        "eval",
        "benchmark"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vlm-failure-probe"
    },
    {
      "name": "syndata-bare",
      "kind": "skill",
      "path": "skills/syndata-bare",
      "tags": [
        "vlm",
        "data",
        "eval"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/syndata-bare"
    },
    {
      "name": "product-rl-loop",
      "kind": "skill",
      "path": "skills/product-rl-loop",
      "tags": [
        "rl",
        "product-rl",
        "eval"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/product-rl-loop"
    },
    {
      "name": "personalization-loop",
      "kind": "skill",
      "path": "skills/personalization-loop",
      "tags": [
        "personalization",
        "eval",
        "retrieval"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/personalization-loop"
    },
    {
      "name": "tinker-loop",
      "kind": "skill",
      "path": "skills/tinker-loop",
      "tags": [
        "dx",
        "cookbook",
        "oss",
        "posttrain-api",
        "finetune",
        "rl"
      ],
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/tinker-loop"
    },
    {
      "name": "fm-os-sync",
      "kind": "workflow",
      "path": "scripts",
      "author": "wjlgatech",
      "source": "https://github.com/wjlgatech/FM-os/tree/main/scripts"
    },
    {
      "name": "eval-llm",
      "kind": "skill",
      "author": "community",
      "note": "LLM output-quality eval — highly SLM-relevant; awaiting self-cert."
    },
    {
      "name": "continual-learning-research",
      "kind": "skill",
      "author": "community",
      "note": "Continual learning / catastrophic forgetting for SLMs; awaiting self-cert."
    }
  ],
  "certifications": {
    "agentic-eval": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 94,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/agentic-eval",
      "tier": "certified"
    },
    "bayesopt-loop": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/bayesopt-loop",
      "tier": "certified"
    },
    "continual-rl-eval": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/continual-rl-eval",
      "tier": "certified"
    },
    "curation-loop": {
      "gaps": [
        "correctness (55)",
        "eval (30)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 91,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/curation-loop",
      "tier": "certified"
    },
    "fm-os": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 94,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/fm-os",
      "tier": "certified"
    },
    "fm-os-sync": {
      "gaps": [
        "docs (0)"
      ],
      "kind": "workflow",
      "on_mission": true,
      "score": 83,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/scripts",
      "tier": "certified"
    },
    "personalization-loop": {
      "gaps": [
        "relevance (54)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 91,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/personalization-loop",
      "tier": "certified"
    },
    "product-rl-loop": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/product-rl-loop",
      "tier": "certified"
    },
    "research-loop": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 92,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/research-loop",
      "tier": "certified"
    },
    "slm-quickstart": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 94,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/slm-quickstart",
      "tier": "certified"
    },
    "syndata-bare": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/syndata-bare",
      "tier": "certified"
    },
    "tinker-loop": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/tinker-loop",
      "tier": "certified"
    },
    "vector-rag": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 94,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vector-rag",
      "tier": "certified"
    },
    "vlm-failure-probe": {
      "gaps": [],
      "kind": "skill",
      "on_mission": true,
      "score": 98,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vlm-failure-probe",
      "tier": "certified"
    },
    "vlm-quickstart": {
      "gaps": [
        "correctness (55)"
      ],
      "kind": "skill",
      "on_mission": true,
      "score": 94,
      "security_ok": true,
      "source": "https://github.com/wjlgatech/FM-os/tree/main/skills/vlm-quickstart",
      "tier": "certified"
    }
  },
  "jd_taxonomy": [
    {
      "id": "python_pytorch",
      "label": "Python / PyTorch / large-scale ML workflows",
      "keywords": [
        "python",
        "pytorch",
        "large-scale ml",
        "ml workflow"
      ],
      "repo_categories": [
        "frameworks",
        "finetuning"
      ],
      "topics": [
        "foundations",
        "pretraining"
      ],
      "kind": "knowledge"
    },
    {
      "id": "vlm_multimodal",
      "label": "Vision-Language / multi-modal models",
      "keywords": [
        "vision-language",
        "vision language",
        "vlm",
        "multi-modal",
        "multimodal",
        "multi modal"
      ],
      "repo_categories": [
        "multimodal"
      ],
      "topics": [
        "vlm",
        "multimodal",
        "vision"
      ],
      "skill_tag": "vlm"
    },
    {
      "id": "video_understanding",
      "label": "Video / motion understanding, spatiotemporal",
      "keywords": [
        "video",
        "motion",
        "spatiotemporal",
        "spatio-temporal",
        "frames",
        "localization"
      ],
      "repo_categories": [
        "multimodal",
        "dataset"
      ],
      "topics": [
        "video"
      ],
      "skill_tag": "vlm"
    },
    {
      "id": "finetuning",
      "label": "Training / fine-tuning foundation models",
      "keywords": [
        "fine-tun",
        "finetun",
        "training and fine-tuning",
        "train",
        "sft",
        "lora",
        "peft"
      ],
      "repo_categories": [
        "finetuning",
        "frameworks"
      ],
      "topics": [
        "finetuning",
        "posttraining"
      ],
      "skill_tag": "finetune"
    },
    {
      "id": "rl_posttraining",
      "label": "Post-training / RL / alignment",
      "keywords": [
        "rlhf",
        "dpo",
        "grpo",
        "reinforcement learning",
        "post-train",
        "posttrain",
        "alignment"
      ],
      "repo_categories": [
        "rl"
      ],
      "topics": [
        "rl",
        "posttraining"
      ],
      "skill_tag": "rl"
    },
    {
      "id": "framework_breadth",
      "label": "Framework breadth (JAX / TensorFlow / Flax)",
      "keywords": [
        "jax",
        "tensorflow",
        "flax",
        "keras"
      ],
      "repo_categories": [
        "jax"
      ],
      "topics": [],
      "kind": "knowledge"
    },
    {
      "id": "agentic_eval",
      "label": "Agentic evaluation / benchmarking",
      "keywords": [
        "eval",
        "benchmark",
        "evaluation framework",
        "narrative consistency",
        "accuracy"
      ],
      "repo_categories": [
        "eval"
      ],
      "topics": [
        "eval"
      ],
      "skill_tag": "eval"
    },
    {
      "id": "benchmark_creation",
      "label": "Building datasets & benchmarks",
      "keywords": [
        "datasets and benchmarks",
        "build large-scale datasets",
        "benchmarks for vision",
        "build benchmark",
        "build datasets"
      ],
      "repo_categories": [
        "eval",
        "dataset"
      ],
      "topics": [
        "eval"
      ],
      "skill_tag": "benchmark"
    },
    {
      "id": "retrieval_vectordb",
      "label": "Retrieval, embeddings & vector databases",
      "keywords": [
        "retrieval",
        "embedding",
        "vector database",
        "vector db",
        "rag"
      ],
      "repo_categories": [
        "retrieval"
      ],
      "topics": [
        "retrieval"
      ],
      "skill_tag": "retrieval"
    },
    {
      "id": "gpu_optimization",
      "label": "GPU optimization / efficient inference",
      "keywords": [
        "gpu optimization",
        "gpu-accelerated",
        "quantiz",
        "inference",
        "throughput",
        "cuda"
      ],
      "repo_categories": [
        "serving",
        "compression"
      ],
      "topics": [
        "compression"
      ],
      "kind": "knowledge"
    },
    {
      "id": "distributed_training",
      "label": "Distributed training & ML orchestration",
      "keywords": [
        "distributed",
        "deepspeed",
        "ray",
        "kubeflow",
        "mlflow",
        "megatron",
        "orchestration"
      ],
      "repo_categories": [
        "distributed",
        "frameworks"
      ],
      "topics": [
        "pretraining"
      ],
      "kind": "knowledge"
    },
    {
      "id": "data_curation",
      "label": "Dataset curation loops (\"AI training AI\")",
      "keywords": [
        "curation",
        "data curation",
        "generate and refine datasets",
        "ai training ai",
        "synthetic data"
      ],
      "repo_categories": [
        "dataset"
      ],
      "topics": [
        "pretraining"
      ],
      "skill_tag": "curation"
    },
    {
      "id": "av_robotics_data",
      "label": "Autonomous-driving / robotics datasets",
      "keywords": [
        "autonomous",
        "self-driving",
        "robotics",
        "sensor data",
        "nuscenes",
        "waymo",
        "driving"
      ],
      "repo_categories": [
        "dataset"
      ],
      "topics": [],
      "kind": "knowledge"
    },
    {
      "id": "research_methodology",
      "label": "Research judgment & empirical rigor (experiment loop)",
      "keywords": [
        "research judgment",
        "empirical rigor",
        "design and analyze",
        "experiment",
        "ablation",
        "reproduc",
        "end-to-end",
        "iterate quickly"
      ],
      "repo_categories": [
        "research"
      ],
      "topics": [
        "research"
      ],
      "skill_tag": "research"
    },
    {
      "id": "research_publishing",
      "label": "Publishing research (NeurIPS / CVPR)",
      "keywords": [
        "publish",
        "research",
        "neurips",
        "cvpr",
        "iclr",
        "publication"
      ],
      "repo_categories": [],
      "topics": [
        "vlm",
        "video",
        "multimodal",
        "slm",
        "rl",
        "posttraining",
        "pretraining",
        "scaling",
        "bayesopt",
        "active-learning"
      ],
      "kind": "knowledge"
    },
    {
      "id": "bayesian_optimization",
      "label": "Bayesian optimization & Gaussian-process surrogates",
      "keywords": [
        "bayesian optimization",
        "bayesian optimisation",
        "botorch",
        "gaussian process",
        "acquisition function",
        "acquisition-strategy",
        "acquisition strategy",
        "surrogate model",
        "expected improvement",
        "ax platform"
      ],
      "repo_categories": [
        "bayesopt"
      ],
      "topics": [
        "bayesopt"
      ],
      "skill_tag": "bayesopt"
    },
    {
      "id": "active_learning_closed_loop",
      "label": "Active learning & closed-loop optimization (DBTL cycles)",
      "keywords": [
        "active learning",
        "active-learning",
        "closed-loop",
        "closed loop",
        "design-build-test-learn",
        "design–build–test–learn",
        "dbtl",
        "self-driving lab",
        "experimental design",
        "iterative design",
        "sequential design"
      ],
      "repo_categories": [
        "bayesopt",
        "bio-ml"
      ],
      "topics": [
        "active-learning",
        "bayesopt"
      ],
      "skill_tag": "bayesopt"
    },
    {
      "id": "uncertainty_quantification",
      "label": "Probabilistic modeling & uncertainty quantification",
      "keywords": [
        "uncertainty quantification",
        "uncertainty-quantification",
        "probabilistic modeling",
        "probabilistic modelling",
        "probabilistic programming",
        "pyro",
        "numpyro",
        "calibration",
        "bayesian deep learning",
        "predictive uncertainty"
      ],
      "repo_categories": [
        "probabilistic",
        "bayesopt"
      ],
      "topics": [
        "uncertainty",
        "bayesopt"
      ],
      "skill_tag": "bayesopt"
    },
    {
      "id": "representation_learning",
      "label": "Representation learning (molecular / sequence embeddings)",
      "keywords": [
        "representation learning",
        "representation-learning",
        "molecular representation",
        "learned representation",
        "latent space",
        "encode domain",
        "featuriz",
        "fingerprint"
      ],
      "repo_categories": [
        "bio-ml",
        "probabilistic"
      ],
      "topics": [
        "representation",
        "molecular"
      ],
      "kind": "knowledge"
    },
    {
      "id": "molecular_bio_ml",
      "label": "ML for molecular / biomolecular design & discovery",
      "keywords": [
        "molecular",
        "molecule",
        "biomolecular",
        "molecule discovery",
        "drug discovery",
        "protein",
        "chemistry",
        "cheminformatics",
        "library design",
        "ligand",
        "therapeutic",
        "wet-lab",
        "wet lab"
      ],
      "repo_categories": [
        "bio-ml"
      ],
      "topics": [
        "molecular"
      ],
      "skill_tag": "bayesopt"
    },
    {
      "id": "multi_objective_constrained",
      "label": "Multi-objective & constrained optimization",
      "keywords": [
        "multi-objective",
        "multi objective",
        "multiobjective",
        "constrained optimization",
        "constrained optimisation",
        "pareto",
        "trade-off",
        "hypervolume",
        "constraint"
      ],
      "repo_categories": [
        "bayesopt"
      ],
      "topics": [
        "bayesopt"
      ],
      "skill_tag": "bayesopt"
    },
    {
      "id": "neuroscience_bci",
      "label": "Neuroscience / brain-computer interfaces",
      "keywords": [
        "neuroscience",
        "brain-computer",
        "brain computer",
        "bci",
        "neural decoding",
        "neural",
        "neuroengineering",
        "non-invasive imaging",
        "electrophysiology"
      ],
      "repo_categories": [
        "neuro"
      ],
      "topics": [
        "neuroscience"
      ],
      "kind": "knowledge"
    },
    {
      "id": "state_space_models",
      "label": "State-space / sequence models (Mamba / S4)",
      "keywords": [
        "state-space",
        "state space",
        "mamba",
        "s4 ",
        "ssm",
        "structured state",
        "selective state"
      ],
      "repo_categories": [
        "bayesopt",
        "frameworks"
      ],
      "topics": [
        "ssm"
      ],
      "kind": "knowledge"
    },
    {
      "id": "product_driven_research",
      "label": "Product-driven research / deployment research",
      "keywords": [
        "product-driven",
        "real world deployment",
        "deployment to millions",
        "research and product teams",
        "combining reinforcement learning and products",
        "deploy to production users",
        "a/b test"
      ],
      "repo_categories": [
        "rl",
        "eval"
      ],
      "topics": [
        "posttraining",
        "product"
      ],
      "skill_tag": "product-rl"
    },
    {
      "id": "personalization_memory",
      "label": "Personalization / memory (personal models)",
      "keywords": [
        "personal agi",
        "personalization",
        "personalize",
        "personal ai",
        "user memory",
        "optimize their own models",
        "custom models",
        "memory"
      ],
      "repo_categories": [
        "retrieval",
        "models"
      ],
      "topics": [
        "personalization"
      ],
      "skill_tag": "personalization"
    },
    {
      "id": "automated_research_agenda",
      "label": "Owning a research agenda / automated AI research",
      "keywords": [
        "research agenda",
        "own and pursue",
        "novel and highly capable models",
        "automated research",
        "ai scientist"
      ],
      "repo_categories": [
        "research"
      ],
      "topics": [
        "research"
      ],
      "kind": "knowledge"
    },
    {
      "id": "ml_codebase_engineering",
      "label": "Large ML codebase engineering & debugging",
      "keywords": [
        "large ml codebase",
        "codebase to debug",
        "research stack",
        "design, implement, test, and debug"
      ],
      "repo_categories": [
        "frameworks"
      ],
      "topics": [
        "foundations",
        "engineering"
      ],
      "kind": "knowledge"
    },
    {
      "id": "dev_experience",
      "label": "Developer experience / API ergonomics / onboarding",
      "keywords": [
        "developer experience",
        "developer ergonomics",
        "ergonomics",
        "onboarding",
        "user friction",
        "reduce friction",
        "dx "
      ],
      "repo_categories": [
        "devtools",
        "serving"
      ],
      "topics": [
        "engineering",
        "devex"
      ],
      "skill_tag": "dx"
    },
    {
      "id": "technical_writing",
      "label": "Code examples / tutorials / cookbook recipes (docs as product)",
      "keywords": [
        "cookbook",
        "tutorial",
        "recipes",
        "code examples",
        "documentation",
        "guides",
        "jupyter notebook"
      ],
      "repo_categories": [
        "devtools",
        "finetuning"
      ],
      "topics": [
        "devex",
        "courses"
      ],
      "skill_tag": "cookbook"
    },
    {
      "id": "oss_library",
      "label": "Open-source library development & community engagement",
      "keywords": [
        "open-source librar",
        "open source librar",
        "software librar",
        "community engagement",
        "open-source community",
        "open source community"
      ],
      "repo_categories": [
        "frameworks",
        "devtools"
      ],
      "topics": [
        "engineering",
        "oss"
      ],
      "skill_tag": "oss"
    },
    {
      "id": "finetune_service_api",
      "label": "Fine-tuning-service API primitives (training-client design)",
      "keywords": [
        "fine-tuning api",
        "fine tuning api",
        "forward_backward",
        "optim_step",
        "training primitives",
        "managed training",
        "tinker",
        "training client"
      ],
      "repo_categories": [
        "finetuning",
        "rl"
      ],
      "topics": [
        "posttraining",
        "devex"
      ],
      "skill_tag": "posttrain-api"
    },
    {
      "id": "tool_use_connectors",
      "label": "Tool-use / connectors / instruction following (agent harness)",
      "keywords": [
        "tool-use",
        "tool use",
        "connectors",
        "instruction following",
        "feature discovery",
        "harness improvements",
        "mcp",
        "world interaction",
        "bringing agents"
      ],
      "repo_categories": [
        "agents",
        "serving"
      ],
      "topics": [
        "agents",
        "tooluse"
      ],
      "skill_tag": "agent-harness"
    },
    {
      "id": "proactivity",
      "label": "Proactive assistant behavior (anticipate user goals)",
      "keywords": [
        "proactivity",
        "proactive",
        "further user goals",
        "anticipate",
        "collaborative assistant"
      ],
      "repo_categories": [
        "agents"
      ],
      "topics": [
        "personalization",
        "agents"
      ],
      "skill_tag": "proactivity"
    },
    {
      "id": "model_behavior_judgment",
      "label": "Model behavior judgment / human-AI interaction quality",
      "keywords": [
        "model behavior",
        "human-ai interaction",
        "human ai interaction",
        "personality",
        "factuality",
        "qualitative problems",
        "training interventions",
        "sycophancy"
      ],
      "repo_categories": [
        "eval"
      ],
      "topics": [
        "eval",
        "posttraining"
      ],
      "kind": "knowledge"
    }
  ],
  "counts": {
    "repos": 117,
    "courses": 33,
    "papers": 99,
    "jobs": 12,
    "labs": 22,
    "models": 13,
    "registry": 17,
    "jd_taxonomy": 34
  }
}