All pull requests

Pull requests

• fix: keep the caller's speed for short text outside streaming

  #1332·li-lizhe opened Oct 22, 2026Contributor·

  ·
  1

• fix(infer): support Arabic and Urdu punctuation in chunk_text

  #1329·xuilen opened Oct 17, 2026·

  ·

• fix: skip reference clips over 12 s in the finetune Gradio test sample

  #1328·Cassian433 opened Oct 15, 2026·

  ·
  1

• fix: split Hindi text on the danda in chunk_text

  #1325·ANDROIDHASSAN opened Oct 14, 2026·

  ·

• Keep the last character of a vocab line that has no newline

  #1321·SashaMIT opened Sep 28, 2026·1 task done·

  ·

• fix(dataset): call datasets.load_dataset for the HFDataset path

  #1316·Anai-Guo opened Aug 29, 2026·

  ·

• Fix grammar in Triton/TRT-LLM setup note

  #1314·GlowLED opened Aug 21, 2026·

  ·

• Fix OS command injection in finetune Gradio start_training()

  #1313·GlowLED opened Aug 21, 2026·

  ·
  1

• Multi gpu

  #1307·webadmin2021 opened Jul 14, 2026·

  ·
  1

• K8s deployment argocd

  #1302·kalyanikumar-shunmugasundaram opened Jun 29, 2026·

  ·

• docs: clarify EMA weights checkpoint loading (#1292)

  #1301·Vidit-lab opened Jun 17, 2026·

  ·
  1

• Add cache=False mode to CFM.sample for gradient-based applications

  #1299·atomsai opened May 27, 2026·

  ·

• feat: opt-in perf — selective GC, CFG-Zero*, torch.compile

  #1297·medyas opened May 17, 2026·

  ·

• feat(train): PEFT (LoRA / LoHa) adapter support for finetune CLI

  #1296·medyas opened May 17, 2026·

  ·
  1

• Fix: align input lengths and disable forced duration extension

  #1255·mikhail2013ru opened Jan 20, 2026·

  ·

• Adding community models as options in custom model dropdown

  #1251·raivisdejus opened Jan 17, 2026Contributor·

  ·

• Move torchcodec to optional dependencies

  #1244·onuryuruten opened Dec 27, 2025·

  ·

• Add compatibility check for torch.xpu before accessing

  #1243·onuryuruten opened Dec 27, 2025·

  ·