Vaya's read on this project
Problem, audience, market, and the verdict — sign in to see it.
Updates
July 2026
- clean up fragile code
- add inference benchmarking
- add a few tests
- unify the two optimizer implementations into one and prevent bugs lik…
- harden and shorten the execution sandbox
- fix deprecated commit on torchao
- delete datasets dependency bye
- nuke huggingface and its giant dependency footprint from orbit. datas…
- delete spurious 'cute' bloat that doesn't matter to a serious researc…
- remove a ton of UI bloat, we need to make nanochat small and sexy and…
- Validate checkpoint file format in find_last_step
- Improve error handling for batch size alignment
- Remove inherited parameters from scripts
- Merge pull request #540 from svlandeg/fix/kernel
- Small doc fixes for consistency
- add a comment clarifying that RoPE here rotates by -theta (transpose …
- fix token_bytes calculation to use raw token bytes instead of round-t…
- docs: note that MPS handles bf16 fine on recent macOS (opt in via NAN…
- allocate the inference KV cache in COMPUTE_DTYPE instead of hardcodin…
- fix mixed-dtype ops in the optimizer so that bf16 params (wte, value_…
May 2026
- tried and failed at DyT
April 2026
- Merge pull request #706 from svlandeg/fix/cpu
- Merge pull request #686 from marcinbogdanski/fix/init-smear-backout-l…
- add comment
- use nn.init and initialize smear gate's weight as well
- update uv lock
- add setuptools for CPU run
- fix: initialize smear and backout lambdas in init_weights
March 2026
- create a group for dev dependencies, there is no need to install all …
- delete non-essential deps from legacy use