Commits

Breadcrumbs

History for

ondevelop

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 26, 2024

• [Inference] use fp8 cuda core gemm kernel when M<=4 (PaddlePaddle#9423)

  zhink

  authored
  0b4b810

  Copy full SHA for 0b4b810

Commits on Nov 25, 2024

• [LLM INFER] Support speculative decoding (llama) (PaddlePaddle#9180)

  Show description for d68a385
  Wanglongzhi2001

  authored
  d68a385

  Copy full SHA for d68a385

Commits on Nov 13, 2024

• fix hip paddle_nlp_ops bug (PaddlePaddle#9418)

  Show description for 09d2c14
  

  TBD1

  and

  root

  authored
  09d2c14

  Copy full SHA for 09d2c14

Commits on Nov 6, 2024

• [LLM INFER] Optimize fuse some kernels in postprocess (PaddlePaddle#9201)

  Show description for 0977858
  gzy19990617

  authored
  0977858

  Copy full SHA for 0977858

Commits on Nov 4, 2024

• [Inference] Append attn FP8 quant (PaddlePaddle#9328)

  Show description for 5217a3b
  ckl117

  authored
  5217a3b

  Copy full SHA for 5217a3b

Commits on Oct 23, 2024

• [LLM INFER] Append attn (PaddlePaddle#9244)

  Show description for 31c6b9a
  

  yuanlehome

  and

  lizhenyun01

  authored
  31c6b9a

  Copy full SHA for 31c6b9a

Commits on Oct 11, 2024

• [INFER] update tune_cublaslt_gemm op and fix some bugs (PaddlePaddle#9222)

  yuanlehome

  authored
  156182e

  Copy full SHA for 156182e

• [LLM INFER] top_p_sampling_reject support top_p=0 and custom seed (PaddlePaddle#9202)

  Show description for f9eb62e
  gzy19990617

  authored
  f9eb62e

  Copy full SHA for f9eb62e

Commits on Sep 20, 2024

• [Inference] FP8 dual gemm auto-tune and support compile parallelization (PaddlePaddle#9151)

  Show description for c4f7acf
  ckl117

  authored
  c4f7acf

  Copy full SHA for c4f7acf

• 【Inference】Optimize top_p kernel performance (PaddlePaddle#9132)

  Show description for 907ad20
  gzy19990617

  authored
  907ad20

  Copy full SHA for 907ad20

Commits on Sep 11, 2024

• [Inference] FP8 gemm auto-tune (PaddlePaddle#9094)

  Show description for 3675ea2
  ckl117

  authored
  3675ea2

  Copy full SHA for 3675ea2

• 【Inference】fix step kernel (PaddlePaddle#9122)

  Show description for ba9c345
  gzy19990617

  authored
  ba9c345

  Copy full SHA for ba9c345

Commits on Sep 4, 2024

• Fix the memory overflow bug of the tune_cublaslt_gemm operator (PaddlePaddle#9076)

  Show description for fbbc0a2
  Hanyonggong

  authored
  fbbc0a2

  Copy full SHA for fbbc0a2

Commits on Sep 2, 2024

• [Inference] support llama3 a8w8c8_fp8 inference and cutlass_fp8_gemm (PaddlePaddle#8953)

  Show description for a275ab7
  

  ckl117

  and

  ckl117

  authored
  a275ab7

  Copy full SHA for a275ab7

Commits on Aug 29, 2024

• [Inference] Add cutlass gemm dequant op (PaddlePaddle#8909)

  Show description for c28caf7
  gzy19990617

  authored
  c28caf7

  Copy full SHA for c28caf7

Loading