Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Feb 14, 2025

• tfacc支持mergemoe操作

  

  黄宇扬

  committed
  fb00385

  Copy full SHA for fb00385

Commits on Feb 11, 2025

• 将权重合并放到读模型时

  

  黄宇扬

  committed
  d7c61fd

  Copy full SHA for d7c61fd

Commits on Feb 10, 2025

• 实现简单的1.58bit量化计算

  

  黄宇扬

  committed
  bdba2b2

  Copy full SHA for bdba2b2

Commits on Feb 8, 2025

• Build改成V4

  

  黄宇扬

  committed
  a6ad757

  Copy full SHA for a6ad757

• 支持deepseekV3

  

  黄宇扬

  committed
  0a34ce7

  Copy full SHA for 0a34ce7

• 修改一下tokenizer的调用

  

  黄宇扬

  committed
  0b49648

  Copy full SHA for 0b49648

Commits on Jan 24, 2025

• fix pytools_t2s for cpu

  

  黄宇扬

  committed
  ecb057d

  Copy full SHA for ecb057d

• 暂时取消一下build.yml里面的android测试

  

  黄宇扬

  committed
  ccafd18

  Copy full SHA for ccafd18

• fix

  

  黄宇扬

  committed
  a46a197

  Copy full SHA for a46a197

• 优化TP推理

  

  黄宇扬

  committed
  c17a8d3

  Copy full SHA for c17a8d3

• fp16 * int4计算换kernel

  

  黄宇扬

  committed
  85b0453

  Copy full SHA for 85b0453

• 优化fastllm cuda cache的逻辑

  

  黄宇扬

  committed
  ad2a89a

  Copy full SHA for ad2a89a

• 部分GemmKernel处理不能整除情况下的bug

  

  黄宇扬

  committed
  d697052

  Copy full SHA for d697052

Commits on Jan 20, 2025

• Merge branch 'master' of https://www.github.com/ztxz16/fastllm

  

  黄宇扬

  committed
  f23bbe3

  Copy full SHA for f23bbe3

• 试验性加入多卡MLP合并

  

  黄宇扬

  committed
  0755a3c

  Copy full SHA for 0755a3c

• 小优化

  

  黄宇扬

  committed
  03ce4f7

  Copy full SHA for 03ce4f7

• getspan增加精度

  

  黄宇扬

  committed
  ec1a067

  Copy full SHA for ec1a067

• 略加速int4 * float16

  

  黄宇扬

  committed
  9d00991

  Copy full SHA for 9d00991

Commits on Jan 13, 2025

• 增加对t2s_model的辅助计算

  huangsheng-tf

  committed
  45ce866

  Copy full SHA for 45ce866

• 支持Relu

  huangsheng-tf

  committed
  f05c92f

  Copy full SHA for f05c92f

Commits on Dec 23, 2024

• 增加QwQ的CMMLU测试

  huangsheng-tf

  committed
  df71273

  Copy full SHA for df71273

Commits on Dec 17, 2024

• Merge pull request #500 from TylunasLi/doc

  Show description for b2e8ff1
  ztxz16

  authored
  b2e8ff1

  Copy full SHA for b2e8ff1

Commits on Dec 15, 2024

• 更改Jinja elif逻辑，修复Qwen2.5多轮；支持Jinja负数下标，支持Qwen1.5最初版本

  

  cgli

  committed
  603911b

  Copy full SHA for 603911b

• 修复Qwen一代推理错误

  

  cgli

  committed
  ee2b970

  Copy full SHA for ee2b970

Commits on Dec 13, 2024

• 修改一下调度

  

  黄宇扬

  committed
  2ed9993

  Copy full SHA for 2ed9993

• 增加支持TeleAI的telechat2-3B, 7B

  

  黄宇扬

  committed
  24d73d8

  Copy full SHA for 24d73d8

Commits on Dec 11, 2024

• 支持gemma2

  

  黄宇扬

  committed
  60871ce

  Copy full SHA for 60871ce

Commits on Dec 9, 2024

• Merge branch 'master' of https://www.github.com/ztxz16/fastllm

  

  黄宇扬

  committed
  a51dea3

  Copy full SHA for a51dea3

• 完成minicpm3的graph形式

  

  黄宇扬

  committed
  5e27fb7

  Copy full SHA for 5e27fb7

• Merge pull request #497 from TylunasLi/doc

  Show description for d22dd1d
  ztxz16

  authored
  d22dd1d

  Copy full SHA for d22dd1d

Commits on Dec 5, 2024

• Jinja2模板支持取模运算和数组切片，支持直接加载codellama-7b-instruct的HF模型

  

  cgli

  committed
  1650218

  Copy full SHA for 1650218

• 修复qwen批量推理错误

  

  cgli

  committed
  3b2220d

  Copy full SHA for 3b2220d

Commits on Dec 2, 2024

• 加速graph llm，达到和llama基本相同的速度

  

  黄宇扬

  committed
  f711b32

  Copy full SHA for f711b32

Commits on Dec 1, 2024

• 支持直接加载Qwen（一代）的HF模型

  

  cgli

  committed
  8c7eb27

  Copy full SHA for 8c7eb27

• 支持MiniCPM3-4B导出并使用C++ Tokenizer

  

  cgli

  committed
  e6d4e56

  Copy full SHA for e6d4e56

Pagination

Previous

Next