karpathy/autoresearch

AI agents running research on single-GPU nanochat training automatically

Stars: 96.1k
Forks: 13.5k
Open beginner issues: 0
Indexed issues: 52

Dominant language: Python
License: No license data
Last GitHub push: Mar 26, 2026
Latest indexed: Sep 18, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed
PR merge metrics: No merged PRs in 30d

Open All

52 open issues indexed

MFU calculation is hardcoded to H100 peak FLOPS Open

karpathy/autoresearch#5 · 0 comments · 2 reactions · 0 assignees · Indexed Sep 18, 2026
Rotary embeddings pre-allocate 10x more memory than needed Open

karpathy/autoresearch#6 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Training results lost if evaluation crashes (no pre-eval checkpoint) Open

karpathy/autoresearch#7 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Low creativity Open

karpathy/autoresearch#22 · 1 comment · 7 reactions · 0 assignees · Indexed Sep 18, 2026
should it support blackwell gpus? Open

karpathy/autoresearch#29 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 18, 2026
Why not fixed FLOPs bugdet? Open

karpathy/autoresearch#35 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Harden cache artifact trust boundary in prepare.py Open

karpathy/autoresearch#41 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Proposal: explicit "Ralph Wiggum loop" for autonomous experimentation Open

karpathy/autoresearch#42 · 3 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026
improvements to novelty Open

karpathy/autoresearch#47 · 8 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Codex doesn't seem to work? Open

karpathy/autoresearch#57 · 41 comments · 36 reactions · 0 assignees · Indexed Sep 18, 2026
Your experiment logs predict a structural ceiling — and how to break past it Open

karpathy/autoresearch#89 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026
Proposal: Long-term Semantic Memory Bank via Vector Embeddings (Preventing Idea Regression) Open

karpathy/autoresearch#100 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Feat: Expose PyTorch Profiler traces as LLM-readable summaries for MFU optimization Open

karpathy/autoresearch#118 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Feat: Progressive sequence-length Curriculum Learning in the Dataloader Open

karpathy/autoresearch#121 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Use Taguchi Orthogonal Arrays as an optional search criteria (PR attached) Open

karpathy/autoresearch#135 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Test GLM-5 as CC backend on single RTX PRO 6000 with flash-attn-4. 50 experiments Open

karpathy/autoresearch#140 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Feat: Structured experiment orchestration via Worker/Function/Trigger primitives Open

karpathy/autoresearch#169 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Proposal: Introduce a Project-Level Long-Term Memory File and a Guidance Agent Mechanism Open

karpathy/autoresearch#179 · 2 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026
Is there a plan to reduce hallucinations? Open

karpathy/autoresearch#200 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Only does depth first search Open

karpathy/autoresearch#206 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Request: add Colab/Kaggle T4 fork to notable forks Open

karpathy/autoresearch#208 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
New folk: autoresearch-glm (autonomous feature discovery for tabular GLMs) Open

karpathy/autoresearch#209 · 0 comments · 4 reactions · 0 assignees · Indexed Sep 18, 2026
Harden downloaded dataset shard cache in prepare.py Open

karpathy/autoresearch#215 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Suggested Experiment - Diffusion Models like Mercury 2 Open

karpathy/autoresearch#220 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Principlex Open

karpathy/autoresearch#226 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Proposal: add a guidance agent before the working agent Open

karpathy/autoresearch#239 · 1 comment · 3 reactions · 0 assignees · Indexed Sep 18, 2026
Merge discussion-43 validated config to master to avoid re-discovery Open

karpathy/autoresearch#243 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
GNAP: git-native coordination layer for the SETI@home-style agent swarm vision Open

karpathy/autoresearch#249 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
AMD / RDNA 4 fork — FA3 replaced with PyTorch SDPA Open

karpathy/autoresearch#250 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026
Proposal: Research Integration Process Open

karpathy/autoresearch#261 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Decision needed: bot-driven automated PR labeling using agreed criteria for hooks, MCP, ACP Open

karpathy/autoresearch#267 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Add UCB1 Dimension-Aware Search + Experiment Memory to program.md — no code changes required Open

karpathy/autoresearch#284 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Primary Agent Context window Open

karpathy/autoresearch#298 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Allow iterative refinement of `program.md` based on experiment results alongside `train.py` Open

karpathy/autoresearch#314 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Plan for Research Constraints/Linting Open

karpathy/autoresearch#335 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
[Bug] Tokenizer.encode: prepend_id scope issue causes NameError when prepend is string Open

karpathy/autoresearch#348 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Proposal: data-centric autoresearch Open

karpathy/autoresearch#349 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 18, 2026
autoresearch-automl: extended / benchmarked autoresearch with LLM-based & classical HPO Open

karpathy/autoresearch#383 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Bug: BPB metric inflated by UTF-8 replacement characters in token byte count Open

karpathy/autoresearch#384 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026
Editable install fails due to multiple top-level modules (setuptools discovery issue) Open

karpathy/autoresearch#387 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
--dangerously-skip-permissions blocked on RunPod (root-only environments) Open

karpathy/autoresearch#396 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
🎉 Featured in awesome-claude-code-workflows Open

karpathy/autoresearch#397 · 1 comment · 3 reactions · 0 assignees · Indexed Sep 18, 2026
Request to add V100 fork to Notable forks Open

karpathy/autoresearch#437 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
is it possible to support hybrid inference (using both RAM and VRAM) on Windows and Ubuntu in the future? Open

karpathy/autoresearch#445 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Copilot code review cleanup job fails deleting artifacts with 403 Open

karpathy/autoresearch#460 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 18, 2026
feat: CLI analysis tool for experiment results — structured feedback for the autonomous agent Open

karpathy/autoresearch#476 · 0 comments · 3 reactions · 0 assignees · Indexed Sep 18, 2026
Trust verification for autonomous research agents? Open

karpathy/autoresearch#505 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
Request to include Ascend-compatible fork in **Notable forks** Open

karpathy/autoresearch#519 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 18, 2026
RTX 3060 fork — 27 experiments, 25% improvement (1.604 → 1.201 val_bpb) Open

karpathy/autoresearch#596 · 2 comments · 2 reactions · 0 assignees · Indexed Sep 18, 2026
Agent can lower val_bpb without improving the model — add an optional experiment integrity log? Open

karpathy/autoresearch#599 · 4 comments · 1 reaction · 0 assignees · Indexed Sep 18, 2026

A short digest of beginner-friendly GitHub issues.