All pull requests

Pull requests

• docs: deploy LitGPT on Nebius Serverless Endpoints

  #2342·SalikovAlex opened Sep 28, 2026·

  ·

• Keep a comma that sits inside a JSON string

  #2340·SashaMIT opened Sep 28, 2026·

  ·
  1

  1

• fix(tokenizer): keep trailing eos when encode(eos=True) input already ends with it

  #2338·MohammadHijjawi97 opened Sep 27, 2026·

  ·

• Add configurable per-layer LoRA ranks (Related to Issue #2335)

  #2336·kor80 opened Sep 25, 2026·

  ·
  1

• Enable Flash Attention for safe KV cache decoding

  #2334·mukeshr06 opened Sep 23, 2026·

  ·
  1

• Fix pretraining max steps and checkpoint saving

  #2333·mukeshr06 opened Sep 23, 2026·

  ·
  1

• docs: clarify tokenizer considerations for new-language finetuning

  #2332·mukeshr06 opened Sep 22, 2026·

  ·

• fix: prevent deadlock in pretrain validation when rank dataloader is empty

  #2331·mayuriphad opened Sep 21, 2026·

  ·
  1

  1

• fix(chat): make --compile work

  #2330·manduinca opened Sep 15, 2026·

  ·
  1

• build(deps): bump astral-sh/setup-uv from 10.0.1 to 10.1.0 in the gha-updates group

  CI / actions
  #2329·dependabot[bot] opened Sep 14, 2026Bot·

  ·

• docs: add Simplified Chinese translation (README.zh.md)

  #2328·JasonYeYuhe opened Sep 13, 2026·

  ·
  1

• Fix adapter KV cache reuse during training

  #2327·Excelius-Wang opened Sep 9, 2026·

  ·
  1

• fix(tokenizer): decode_stream drops leading space per-word on the huggingface backend

  #2326·999thBaam opened Sep 9, 2026·

  ·
  1

• fix(generate): preserve sampling support for CPU float16 probabilities

  #2325·Excelius-Wang opened Sep 9, 2026·

  ·
  1

• fix(merge_lora): use CPU-safe precision to avoid false AMP warning

  #2322·ShamikOfficial opened Sep 7, 2026·

  ·
  1

• fix: keep pretrain validate lockstep on empty ranks (#2319)

  #2321·ShamikOfficial opened Sep 4, 2026·

  ·

• Handle alternate CLI entry points when saving checkpoints

  #2318·Marchematics opened Aug 31, 2026·

  ·
  1

  1

• feat: add MultiturnJSON datamodule + full multi-turn SFT support

  #2317·ysjprojects opened Aug 31, 2026Collaborator·3/3·

  ·

• feat: MultiturnSFTDataset

  #2315·ysjprojects opened Aug 30, 2026Collaborator·2/3·

  ·

• feat: multiturn support for ChatML

  #2314·ysjprojects opened Aug 30, 2026Collaborator·1/3·

  ·

• Preserve pretraining epoch across resume

  #2313·Alphaxiaoteng opened Aug 30, 2026·

  ·
  1

• feat: make cross_entropy_chunk_size tunable, with an auto memory budget (#2190) [3/3]

  #2312·OnePunchMonk opened Aug 28, 2026Contributor·

  ·
  1

• profile: measure the chunked_cross_entropy memory spike and find its source (#2190) [2/3]

  #2311·OnePunchMonk opened Aug 28, 2026Contributor·

  ·
  1

• fix: raise a clear error on stale KV cache, avoid double allocation in chat (#2190) [1/3]

  #2310·OnePunchMonk opened Aug 28, 2026Contributor·

  ·

• fix(generate): reject top_k below 1 instead of masking every logit

  #2309·Anai-Guo opened Aug 27, 2026Contributor·

  ·
  2