Core
libraryGPT/optimizer/data/tokenizer/eval/inference.
Minimal end-to-end LLM training harness covering tokenizer, pretraining, SFT, RL, evaluation, inference, and scaling/speedrun scripts.
Minimal end-to-end LLM training harness covering tokenizer, pretraining, SFT, RL, evaluation, inference, and scaling/speedrun scripts.
python -m nanochat.chat_webEach card below corresponds to a component identified in the repository research. Open a component record for its path, purpose, capabilities, dependencies, risks, and relationships.
GPT/optimizer/data/tokenizer/eval/inference.
Train/eval/SFT/RL/chat/tokenizer/benchmarks.
Speedrun/scaling/miniseries/CPU scripts.
Evaluation task adapters.
These capabilities are derived from the current first-party README and supporting documentation. Names follow upstream terminology so you can search the source documentation precisely.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns time-to-gpt-2 leaderboard. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns reproduce and talk to gpt-2. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns research. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns running on cpu / mps. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns precision / dtype. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns guides. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Upstream treats this as a distinct part of karpathy/nanochat. In plain language, use this area when your task concerns cite. Begin with the documented default, test it on disposable input, and open the source section for its current options and limitations.
Start with one capability card instead of asking the repository to do everything at once. This makes permissions, inputs, and output easier to understand.
Use a sample URL, copied repository, test document, or sandbox account. Keep production credentials and irreplaceable files out of the first run.
Watch Terminal output or the host application's activity view. If the behavior differs from the README, stop with Control + C or cancel inside the host before retrying.
Check generated files, diffs, API responses, logs, or previews. Never assume “command finished” means the result is correct.
Record the working command and non-secret configuration in your project's README. Store secrets in the documented environment file or password manager.
These examples are selected from the repository's first-party documentation. Replace obvious placeholders, keep quotation marks intact, and run them only in the context indicated by the surrounding explanation.
NANOCHAT_DTYPE=float32 python -m scripts.chat_cli -p "hello" # force fp32
NANOCHAT_DTYPE=bfloat16 torchrun --nproc_per_node=8 -m scripts.base_train # force bf16None documented
A configuration file changes behavior without changing source code. Make one change at a time and keep a backup before editing JSON, TOML, YAML, or environment files.
nanochat/, scripts/, runs/, tasks/
Paths identify where the relevant implementation or generated files live. Paths beginning with ~ are inside your home folder.
None documented
Prefer temporary, least-privileged credentials. Never commit .env, tokens, cookies, private keys, or session exports.
None documented
Check pricing, data retention, rate limits, and account permissions before enabling optional integrations.
git diff before committing.127.0.0.1 unless you intentionally secure and expose them.The manual reflects first-party files checked on August 18, 2026. It explains the indexed repository rather than promising that every optional third-party integration is available or safe.