-
Notifications
You must be signed in to change notification settings - Fork 7.9k
Pull requests: karpathy/nanochat
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
SFT: Fix incorrect progress accounting and LR scaling.
#816
opened Jul 31, 2026 by
marcinbogdanski
Contributor
Loading…
fix: count Float8Linear in num_matmul_params (--fp8 understates FLOPs/token and MFU ~12x)
potential_bug
Needs investigation/confirmation whether or not it's a bug
#815
opened Jul 30, 2026 by
Xiaomu-Lin
Loading…
read-arxiv-paper skill: handle the URL and archive shapes arxiv actually serves
docs
Improvements or additions to documentation
#814
opened Jul 29, 2026 by
Palo-Alto-AI-Research-Lab
Loading…
Small typo: removing redundant head_dim from gpt file
improvement
suggest/merge
#813
opened Jul 23, 2026 by
helloaidank
Contributor
Loading…
Fix SFT dataloader buffer clog that leads to NaN loss
code robustness
#811
opened Jul 21, 2026 by
parasol-aser
Loading…
Fix wrong worked example in short_window comment
docs
Improvements or additions to documentation
#809
opened Jul 17, 2026 by
Osamaali313
Loading…
chat_rl: support resuming an interrupted RL run
feature
New feature or request
#808
opened Jul 13, 2026 by
parasol-aser
Loading…
94 minutes - Widening embedding after first transformer block
potential_improvement
#807
opened Jul 13, 2026 by
jurca
Loading…
Add counterfactual controls + response logging to the CORE eval
feature
New feature or request
#806
opened Jul 11, 2026 by
veryfansome
Loading…
Fix SDPA fallback to honor non-causal attention
potential_bug
Needs investigation/confirmation whether or not it's a bug
#805
opened Jul 10, 2026 by
KingJing1
Loading…
Fix naive generate() crashing on single-token prompts
potential_bug
Needs investigation/confirmation whether or not it's a bug
#794
opened Jun 30, 2026 by
devYRPauli
Loading…
Make device synchronization MPS-aware (fix crash + correct Mac timing)
potential_improvement
#792
opened Jun 27, 2026 by
eaglstun
Loading…
Match calculator tool output to training format (drop trailing .0)
potential_improvement
#786
opened Jun 24, 2026 by
devYRPauli
Loading…
Add MMLU-Pro Evaluation for Improved Reasoning Measurement
feature
New feature or request
#785
opened Jun 23, 2026 by
Mithun-033
Loading…
fix: make calculator timeout work on Windows and fix signal race condition
potential_improvement
#784
opened Jun 23, 2026 by
putta-saikiran
Loading…
chat_sft: fix --num-iterations to count optimizer steps, not micro-batches
potential_bug
Needs investigation/confirmation whether or not it's a bug
#775
opened Jun 12, 2026 by
walbis
Loading…
docs: improve skill description for better trigger accuracy
meta
#760
opened Jun 3, 2026 by
yanghaod2278827
Loading…
tokenizer: clean up encode_special caching + O(n) prepend in batch encode to better nano chat :)
waiting
Waiting for user feedback/action
#753
opened May 15, 2026 by
eeea2222
Loading…
fix: sliding window truncation in chat_cli to prevent context overflow crash
code robustness
waiting
Waiting for user feedback/action
#746
opened May 5, 2026 by
arehman-0
Loading…
Docker image for runpod.io to speed up instance setup
install
#745
opened May 5, 2026 by
ddudek
Loading…
get the code for evaluation of sample into a funtion to avoid duplicate code.
refactor
#744
opened May 5, 2026 by
xrdddd
Loading…
Use requests instead of urllib for downloads, fixes SSL on macOS
potential_bug
Needs investigation/confirmation whether or not it's a bug
#740
opened Apr 30, 2026 by
modulovalue
Loading…
3 tasks done
Previous Next
ProTip!
Exclude everything labeled
bug with -label:bug.