All pull requests

Pull requests

• feat(i18n): add Italian (it) translation for the Gradio UI

  #1365·fabriziosalmi opened Oct 23, 2026·

  ·
  3

• fix(mlx-vae): cap MLX buffer cache during native VAE decode to avoid swap on Apple Silicon

  #1362·Carlknowsnothing opened Oct 18, 2026·

  ·
  12

• Add Colab notebook for Side-Step LoRA/LoKR training on ACE-Step v1.5 …

  #1361·walidb27 opened Oct 17, 2026·

  ·
  20

• fix(api): honor GPU tier defaults for DiT offloading

  #1359·oodadoudou opened Oct 15, 2026·

  ·
  1

  1

• fix(inference): preserve resolved seeds and effective batch size in metadata

  #1357·laulauraluo opened Oct 14, 2026·

  ·
  1

  1

• fix(nano-vllm): keep the 5 Hz LM inside float16 range on GPUs without bfloat16

  #1353·christopherrobertbrooks-tech opened Oct 8, 2026·

  ·
  4

• fix(cuda): compute the lyric encoder's last projection in float32 on float16 GPUs

  #1352·christopherrobertbrooks-tech opened Oct 8, 2026·

  ·
  1

• fix: release nano-vLLM runtime before reload and PMI scoring

  #1349·nii-101-km opened Oct 3, 2026·

  ·
  10

• fix(mlx): honor cover_noise_strength in cover generation

  #1348·ajshooting opened Oct 3, 2026·

  ·
  1

  1

• fix: train API LoRA on MPS in bfloat16

  #1344·mchosc opened Sep 30, 2026·

  ·
  4

• fix: decode LoRA audio with ffmpeg when torchcodec cannot

  #1343·mchosc opened Sep 30, 2026·

  ·
  6

• fix: keep the automatic MLX VAE decode window at 512 frames

  #1342·mchosc opened Sep 30, 2026·

  ·
  3

• fix(training): read UTF-8 BOM CSV metadata

  #1335·YaoxinHuang opened Sep 23, 2026·

  ·
  1

• fix(lrc): cache restored lyrics for the selected batch sample

  #1334·YaoxinHuang opened Sep 23, 2026·

  ·
  1

• fix(flow-edit): keep conditioning model on CUDA

  #1331·ballerburg9005 opened Sep 17, 2026·

  ·
  3

• fix(pipeline): import loguru logger used by the quantization-probe handler

  #1321·Anai-Guo opened Sep 8, 2026·

  ·
  1

• fix[lowvram]: Optimizing CPU offloading when using a LoRA in a low vram environment

  #1319·slice27 opened Sep 2, 2026·

  ·
  1

• feat: add optional Atlas thumbnail provider

  #1316·binyangzhu000-sudo opened Aug 30, 2026·5 tasks done·

  ·
  4

• Separate generation success from export-format failure

  #1313·atlas-ocm opened Aug 29, 2026·

  ·
  12

• Implement ACESTEP_DTYPE CUDA dtype override advertised by NaN error hint

  #1312·atlas-ocm opened Aug 29, 2026·

  ·
  4

• fix(llm): Fix LM engine memory leak on web UI service reinit

  #1311·dd178 opened Aug 29, 2026Contributor·

  ·
  11

• fix(flow-edit): Fix device mismatch when DiT offload meets flow-edit morph

  #1309·dd178 opened Aug 29, 2026Contributor·

  ·
  3

• fix(inference): keep requested vocal_language over CoT detection

  #1302·badaldesai opened Aug 25, 2026·

  ·
  7

• fix(gpu_config): size the LM's DiT reserve for real, duration-aware demand

  #1301·FlexOr2 opened Aug 24, 2026Contributor·

  ·

• fix(api): failed job cache payload carries the recorded error

  #1300·FlexOr2 opened Aug 24, 2026Contributor·

  ·
  5