Commits

Branch selector

main

User selector

All users

All time

Commit history

Commits on Feb 18, 2026

• fix(qwen2): load attention bias in quantized model loader

  Show description for cd22d9d
  

  BH3GEI

  and

  claude

  committed
  cd22d9d

  Copy full SHA for cd22d9d

Commits on Feb 14, 2026

• docs: Add model management API endpoints to main README

  Show description for fc386a9
  

  ymote

  and

  claude

  committed
  fc386a9

  Copy full SHA for fc386a9

• fix: Strip punctuation in AISHELL CER benchmark for fair comparison

  Show description for 10edc82
  

  ymote

  and

  claude

  committed
  10edc82

  Copy full SHA for 10edc82

• feat: Add model management API (download, delete, config) to SALA server

  Show description for 357c6ff
  

  ymote

  and

  claude

  committed
  357c6ff

  Copy full SHA for 357c6ff

• docs: Add Quick Start, examples, and API documentation to SALA README

  Show description for a0592b2
  

  ymote

  and

  claude

  committed
  a0592b2

  Copy full SHA for a0592b2

• docs: Add MiniCPM-SALA to main README with model path and API endpoints

  Show description for d828020
  

  ymote

  and

  claude

  committed
  d828020

  Copy full SHA for d828020

• chore: Clean up crates, update deps, and refactor funasr-qwen4b model

  Show description for 2b3f490
  

  ymote

  and

  claude

  committed
  2b3f490

  Copy full SHA for 2b3f490

• feat: Add benchmarks, NIAH test, code quality fixes, and Qwen3 comparison

  Show description for 20a8137
  

  ymote

  and

  claude

  committed
  20a8137

  Copy full SHA for 20a8137

• feat: Add OpenAI-compatible API server for MiniCPM-SALA

  Show description for 9e01d55
  

  ymote

  and

  claude

  committed
  9e01d55

  Copy full SHA for 9e01d55

• feat: Add chat template, prompt caching, and self-speculative decoding

  Show description for a3d71ca
  

  ymote

  and

  claude

  committed
  a3d71ca

  Copy full SHA for a3d71ca

• feat: Add batched inference, custom Metal kernels, 8-bit quantization, and RoPE batch fix

  Show description for b7a32df
  

  ymote

  and

  claude

  committed
  b7a32df

  Copy full SHA for b7a32df

• feat: Add MiniCPM-SALA 9B hybrid attention inference crate

  Show description for 8ec2a46
  

  ymote

  and

  claude

  committed
  8ec2a46

  Copy full SHA for 8ec2a46

Commits on Feb 13, 2026

• Added dark mode image

  hzeng412

  authored
  85eed8a

  Copy full SHA for 85eed8a

Commits on Feb 12, 2026

• fix: Fix float16 precision overflow and improve GPT-SoVITS voice clone quality

  Show description for 127017a
  

  Server

  and

  claude

  committed
  127017a

  Copy full SHA for 127017a

Commits on Feb 10, 2026

• Enhance README with image and project details

  Show description for 739aa97
  hzeng412

  authored
  739aa97

  Copy full SHA for 739aa97

Commits on Feb 9, 2026

• Update README to remove Contributing section

  Show description for 66d8a80
  hzeng412

  authored
  66d8a80

  Copy full SHA for 66d8a80

Commits on Feb 7, 2026

• feat: Add Moxin-7B VLM inference crate with 8-bit quantization

  Show description for 88ab9a2
  

  ymote

  and

  claude

  committed
  88ab9a2

  Copy full SHA for 88ab9a2

• fix: default-init norms not present in diffusers weights

  Show description for 79618b0
  

  ymote

  and

  claude

  committed
  79618b0

  Copy full SHA for 79618b0

Commits on Feb 6, 2026

• feat: Add pre-quantized 8-bit FLUX loading support

  Show description for e383ace
  

  ymote

  and

  claude

  committed
  e383ace

  Copy full SHA for e383ace

• fix: Fix ASR preprocessing and generation truncation, add 8-bit support

  Show description for 0e9b273
  

  ymote

  and

  claude

  committed
  0e9b273

  Copy full SHA for 0e9b273

• feat: Auto-download pre-built MLX when Xcode is unavailable

  Show description for 7540f68
  

  ymote

  and

  claude

  committed
  7540f68

  Copy full SHA for 7540f68

• feat: Add GLM-4.7-Flash (MoE + MLA) inference crate

  Show description for 7427ef5
  

  ymote

  and

  claude

  committed
  7427ef5

  Copy full SHA for 7427ef5

Commits on Feb 2, 2026

• feat: Add source code, examples, and documentation across all MLX crates

  Show description for cba4621
  

  ymote

  and

  claude

  committed
  cba4621

  Copy full SHA for cba4621

Commits on Feb 1, 2026

• feat: Add speed factor support for TTS synthesis

  Show description for ac9b5b5
  

  ymote

  and

  claude

  committed
  ac9b5b5

  Copy full SHA for ac9b5b5

• Add preprocessing module and refactor voice_clone config

  Show description for 97cd591
  

  ymote

  and

  claude

  committed
  97cd591

  Copy full SHA for 97cd591

Commits on Jan 30, 2026

• Add FP32 accumulation before norm_out for numerical stability

  Show description for b71e73a
  

  ymote

  and

  claude

  committed
  b71e73a

  Copy full SHA for b71e73a

• docs: Deprioritize INT8 quantization in performance plan

  Show description for 74a8ef6
  

  ymote

  and

  claude

  committed
  74a8ef6

  Copy full SHA for 74a8ef6

• docs: Add performance enhancement plan

  Show description for 6d82049
  

  ymote

  and

  claude

  committed
  6d82049

  Copy full SHA for 6d82049

• Add full precision example, DORA model path support, and README

  Show description for 36c02d8
  

  ymote

  and

  claude

  committed
  36c02d8

  Copy full SHA for 36c02d8

• Add gradient-based discriminator training for VITS

  Show description for e62dce2
  

  ymote

  and

  claude

  committed
  e62dce2

  Copy full SHA for e62dce2

• feat(gpt-sovits-mlx): Add VITS training CLI and checkpoint save/load

  Show description for 10539ac
  

  ymote

  and

  claude

  committed
  10539ac

  Copy full SHA for 10539ac

• feat(gpt-sovits-mlx): Add PosteriorEncoder and forward_train for VITS training

  Show description for 0f759a5
  

  ymote

  and

  claude

  committed
  0f759a5

  Copy full SHA for 0f759a5

• feat(gpt-sovits-mlx): Add VITS training infrastructure

  Show description for cf7f3f8
  

  ymote

  and

  claude

  committed
  cf7f3f8

  Copy full SHA for cf7f3f8

• refactor(gpt-sovits-mlx): Move model paths to ~/.dora/models/primespeech

  Show description for c6d7e4a
  

  ymote

  and

  claude

  committed
  c6d7e4a

  Copy full SHA for c6d7e4a

• feat(gpt-sovits-mlx): Add T2S training support with MLX acceleration

  Show description for b38d090
  

  ymote

  and

  claude

  committed
  b38d090

  Copy full SHA for b38d090