{"full_name":"karpathy/nanochat","owner":"karpathy","name":"nanochat","url":"https://github.com/karpathy/nanochat","description":"The best ChatGPT that $100 can buy.","homepage":"","topics":[],"language":"Python","license":"MIT","avatar":"https://avatars.githubusercontent.com/u/241138?u=05376db54475c3d23b3a409f4c47d14c4855dc28&v=4","created_at":"2025-10-13T13:46:35Z","pushed_at":"2026-09-07T20:06:43Z","metrics":{"stars":60412,"forks":8471,"watchers":380,"issues":146,"prs":564,"releases":0,"issues_open":null,"issues_closed":null,"prs_open":null,"prs_closed":null,"prs_merged":null,"discussions":161},"growth":{"day":{"stars":21,"forks":2,"base_day":"2026-11-09"}},"history":{"dates":["2026-11-09","2026-11-10"],"stars":[60391,60412],"forks":[8469,8471],"watchers":[380,380],"issues_open":[null,null],"prs_open":[null,null]},"boards":{},"same_owner":[{"full_name":"karpathy/autoresearch","name":"autoresearch","boards":{"total":145}},{"full_name":"karpathy/LLM101n","name":"LLM101n","boards":{}},{"full_name":"karpathy/arxiv-sanity-preserver","name":"arxiv-sanity-preserver","boards":{}},{"full_name":"karpathy/build-nanogpt","name":"build-nanogpt","boards":{}},{"full_name":"karpathy/llama2.c","name":"llama2.c","boards":{}},{"full_name":"karpathy/micrograd","name":"micrograd","boards":{}},{"full_name":"karpathy/nanoGPT","name":"nanoGPT","boards":{}},{"full_name":"karpathy/neuraltalk2","name":"neuraltalk2","boards":{}},{"full_name":"karpathy/nn-zero-to-hero","name":"nn-zero-to-hero","boards":{}}],"issues":[{"number":737,"title":"Use Official PyTorch FA3 Builds","comments":5,"created_at":"2026-04-29T18:25:02Z"},{"number":590,"title":"[Bug] scripts/chat_sft.py produces loss: nan from step 00001 on small device-batch-size (≤8) due to fully-masked micro-batches","comments":4,"created_at":"2026-03-05T16:32:03Z"},{"number":858,"title":"Could nanochat be a testbed for recurrent/loop Transformers?","comments":3,"created_at":"2026-09-22T01:33:27Z"},{"number":810,"title":"chat_rl: the sampled assistant_end token is masked out, so stopping is never reinforced","comments":3,"created_at":"2026-07-20T16:12:53Z"},{"number":698,"title":"RTX 3050TI Not detecting m.get_device()","comments":3,"created_at":"2026-04-12T06:35:53Z"},{"number":542,"title":"Add --model-tag to scripts.chat_cli call in runcpu.sh/speedrun.sh example","comments":2,"created_at":"2026-02-19T00:11:39Z"},{"number":514,"title":"[BUG] AssertionError in gpt.py: Evaluation prompts exceed static RoPE cache (seq_len=256)","comments":2,"created_at":"2026-02-08T11:42:56Z"},{"number":73,"title":"Question: Are there plans to implement TorchScript/TorchTrace/ONNX inference?","comments":2,"created_at":"2025-10-15T21:18:44Z"}],"discussions":[{"number":889,"title":"MeG","comments":1,"category":"Show and tell","created_at":"2026-11-04T11:34:43Z"},{"number":868,"title":"Polynomial kernel attention of odd degree","comments":0,"category":"Show and tell","created_at":"2026-10-29T07:08:48Z"},{"number":867,"title":"VLLM plugin for nanchat 2026 #92d63d4","comments":0,"category":"Show and tell","created_at":"2026-10-27T00:30:48Z"},{"number":863,"title":"airgpt: a nanochat-style pipeline for Apple Silicon in MLX","comments":0,"category":"Show and tell","created_at":"2026-10-12T23:14:46Z"},{"number":862,"title":"8x RTX Pro 6000 S GPUs pre-trained in ~2hr 15m","comments":0,"category":"General","created_at":"2026-10-12T16:56:13Z"},{"number":856,"title":"Stuck at the optimizer after figuring out backward()","comments":1,"category":"General","created_at":"2026-09-17T11:25:34Z"},{"number":846,"title":"Verify run 6 baseline metrics","comments":3,"category":"General","created_at":"2026-09-01T15:38:58Z"},{"number":819,"title":"GPT-2 grade on a single RTX 5090 in 41 hours — recipe, Blackwell gotcha, and an open question about sliding windows","comments":1,"category":"Show and tell","created_at":"2026-08-02T10:10:58Z"}],"latest_release":null,"readme":true,"readme_zh":false,"news":[],"stale":false,"snapshot_at":"2026-11-10T09:35:25Z","history_start":"2026-09-06"}