Commits

Branch selector

 main

User selector

liangan1

Datepicker

All time

Commit History

Commits on Aug 30, 2024

• Enabled flash attn varlen for chunked prefilll (#3148)

  liangan1

  authored
  1479d3d

  Copy full SHA for 1479d3d

Commits on Aug 21, 2024

• Refactor IAKV and optimize beam index process (#3209)

  Show description for 65e9663
  liangan1

  authored
  65e9663

  Copy full SHA for 65e9663

Commits on Aug 8, 2024

• Optimize GQA for paged attention (#3174)

  liangan1

  authored
  3848399

  Copy full SHA for 3848399

Commits on Jul 16, 2024

• Fix get_world_size issue (#3082)

  liangan1

  authored
  309255e

  Copy full SHA for 309255e

Commits on Jul 15, 2024

• Fix the WOQ-INT4 crash issue when the pre-allocated buffer is not enough (#3079) (#3081)

  liangan1

  authored
  eb121e0

  Copy full SHA for eb121e0

Commits on Jul 9, 2024

• Fix offset issue in MaskedMultiHeadAttentionKrnl.cpp (#3061)

  liangan1

  authored
  9c13a75

  Copy full SHA for 9c13a75

Commits on Jun 28, 2024

• Enable multiple query for the next tokens (#3018)

  Show description for d6599b5
  liangan1

  authored
  d6599b5

  Copy full SHA for d6599b5

Commits on Jun 4, 2024

• Add fast path for beam idx update in the iakv (#2950)

  Show description for 61ff58f
  liangan1

  authored
  61ff58f

  Copy full SHA for 61ff58f

Commits on May 30, 2024

• [Paged Attention]Eanble flash decoding and use a new data layout (#2929)

  Show description for cd1f0ce
  liangan1

  authored
  cd1f0ce

  Copy full SHA for cd1f0ce

Commits on May 22, 2024

• Fix iakv regression (#2900)

  Show description for 21b5030
  liangan1

  authored
  21b5030

  Copy full SHA for 21b5030

Commits on May 14, 2024

• Add fast path for greedy and bs=1 (#2878)

  liangan1

  authored
  ec863c0

  Copy full SHA for ec863c0

Commits on May 13, 2024

• Enable stride paged attention (#2875)

  liangan1

  authored
  c608a02

  Copy full SHA for c608a02

Commits on May 10, 2024

• Fix ipex.comm import issue (#2854)

  liangan1

  authored
  a178f99

  Copy full SHA for a178f99

Commits on Apr 29, 2024

• Refine batch scaling with kv block. (#2821)

  Show description for 4e6d4c0
  liangan1

  authored
  4e6d4c0

  Copy full SHA for 4e6d4c0

Commits on Mar 18, 2024

• Refine oneccl build (#2668)

  Show description for 9dcceef
  liangan1

  authored
  9dcceef

  Copy full SHA for 9dcceef

Commits on Mar 7, 2024

• IPEX Tensor Parallel (#2435)

  liangan1

  authored
  4fa6445

  Copy full SHA for 4fa6445

Commits on Jan 12, 2024

• Refine env setup script (#2462)

  Show description for f232119
  liangan1

  authored
  f232119

  Copy full SHA for f232119

Commits on Jan 11, 2024

• Reduce write cache polution to improve IAKV performance (#2457)

  liangan1

  authored
  d656e86

  Copy full SHA for d656e86

Commits on Dec 13, 2023

• Enable paged attention op (#2245)

  liangan1

  authored
  af8646c

  Copy full SHA for af8646c

Commits on Dec 11, 2023

• Refactor ROPE kernel and enable concat qkv input for ROPE (#2305)

  liangan1

  authored
  339bd25

  Copy full SHA for 339bd25

• Reduce the initilization overheads in IAKV (#2341)

  liangan1

  authored
  fc2b99a

  Copy full SHA for fc2b99a

Commits on Nov 20, 2023

• Reduce all-reduce call stack overhead. (#2277)

  liangan1

  authored
  2387926

  Copy full SHA for 2387926

Commits on Sep 8, 2023

• Change comm.all_reduce to comm.inference_all_reduce (#1985)

  Show description for 95eee49
  

  liangan1

  and

  chunyuan-w

  authored
  95eee49

  Copy full SHA for 95eee49

Commits on Aug 7, 2023

• Add ROPE kernel for GPT-J/LLama/GPT-Neox (#1660)

  Show description for 234e598
  

  liangan1

  and

  jianan-gu

  authored
  234e598

  Copy full SHA for 234e598

Commits on Aug 4, 2023

• Enable indirect access kv_cache (#1750)

  liangan1

  authored
  496c109

  Copy full SHA for 496c109

Commits on May 19, 2023

• Set CCL as the defualt backend to align with DS (#1646)

  Show description for c686eca
  liangan1

  authored
  c686eca

  Copy full SHA for c686eca

Commits on May 5, 2023

• Remove runtime reorder for linear (#1545)

  Show description for 6db7bcd
  

  3 peopleauthored
  6db7bcd

  Copy full SHA for 6db7bcd

Commits on Apr 17, 2023

• Enable all_reduce jit by defining a "deepspeed_comm" namespace in torch.ops (#1538)

  Show description for b3c847d
  liangan1

  authored
  b3c847d

  Copy full SHA for b3c847d

Commits on Mar 10, 2023

• Refine fast_bert logic (#1474)

  Show description for 9c88643
  liangan1

  authored
  9c88643

  Copy full SHA for 9c88643

Commits on Feb 21, 2023

• Refine fast_bert solution (#1412)

  Show description for 7b3e7ea
  liangan1

  authored
  7b3e7ea

  Copy full SHA for 7b3e7ea

Commits on Feb 16, 2023

• Integrate TPP code into IPEX (#1357)

  Show description for 82b6aa0
  liangan1

  authored
  82b6aa0

  Copy full SHA for 82b6aa0

Commits on Nov 21, 2022

• Enable logical core for ccl in launcher (#1258)

  liangan1

  authored
  688fd72

  Copy full SHA for 688fd72

Commits on Nov 9, 2022

• Enable merge dense embedding and ParameterList weight cast (#1198)

  Show description for abda5f0
  liangan1

  authored
  abda5f0

  Copy full SHA for abda5f0

Commits on Apr 25, 2022

• Fixed einsum_add scalar issue (#699)

  Show description for 8f56d79
  liangan1

  authored
  8f56d79

  Copy full SHA for 8f56d79

Commits on Apr 11, 2022

• Add ipex::einsum_add (#674)

  Show description for 3094f34
  liangan1

  authored
  3094f34

  Copy full SHA for 3094f34

Pagination

Previous

Next