Code

[Gaudi][Model] Qwen2.5-vl #870

Code

Open

malkomes wants to merge 34 commits intoHabanaAI:habana_mainHabanaAI/vllm-fork:habana_mainfrom malkomes:qwen2.5-vl-hpumalkomes/vllm-fork:qwen2.5-vl-hpuCopy head branch name to clipboard

+158-26

Commits

Commits on Mar 10, 2025

• Initial commit

  Show description for 938ef83
  

  ssarkar2

  authored and

  malkomes

  committed
  938ef83

  Copy full SHA for 938ef83

• Comments to trace execution diff between cpu/hpu

  

  ssarkar2

  authored and

  malkomes

  committed
  a3f884b

  Copy full SHA for a3f884b

• minor

  

  ssarkar2

  authored and

  malkomes

  committed
  c83c882

  Copy full SHA for c83c882

• _validate_and_reshape_mm_tensor looks buggy...

  Show description for c5f65f9
  

  ssarkar2

  authored and

  malkomes

  committed
  c5f65f9

  Copy full SHA for c5f65f9

• Some comments regd buggy hpu graphs

  Show description for fca160d
  

  ssarkar2

  authored and

  malkomes

  committed
  fca160d

  Copy full SHA for fca160d

• Return early to prevent mem profiling

  

  ssarkar2

  authored and

  malkomes

  committed
  095dbbd

  Copy full SHA for 095dbbd

• Initial commit for the Qwen 2.5 VL

  

  pallavijaini0525

  authored and

  malkomes

  committed
  f557d99

  Copy full SHA for f557d99

• workaround to make HPU graphs work. disable_tensor_cache set to false.

  

  ssarkar2

  authored and

  malkomes

  committed
  8c7a2b3

  Copy full SHA for 8c7a2b3

• adding qwen2.5-vl to hpu + small cleanups

  malkomes

  committed
  22bc3ef

  Copy full SHA for 22bc3ef

• removing duplicates CPU

  malkomes

  committed
  d4a721c

  Copy full SHA for d4a721c

• small changes to work with llama-3.2-vl

  malkomes

  committed
  5474d9b

  Copy full SHA for 5474d9b

• skip profile_run for now

  malkomes

  committed
  008fbb5

  Copy full SHA for 008fbb5

• reshape positions in MRotaryEmbedding for HPU

  malkomes

  committed
  f48d6fc

  Copy full SHA for f48d6fc

• input positions [3, seq_len] or [seq_len,] for Qwen2.5vl

  malkomes

  committed
  4caf383

  Copy full SHA for 4caf383

• fix the decoder

  malkomes

  committed
  998d090

  Copy full SHA for 998d090

• comment prints

  malkomes

  committed
  cd1bbe0

  Copy full SHA for cd1bbe0

• cleanup

  malkomes

  committed
  99f8e9f

  Copy full SHA for 99f8e9f

• polishing

  malkomes

  committed
  9eac068

  Copy full SHA for 9eac068

• add type ignore

  malkomes

  committed
  dcc2c6c

  Copy full SHA for dcc2c6c

• set HPU_DISABLE_TENSOR_CACHE to false for Qwen2.5vl

  malkomes

  committed
  7c5871b

  Copy full SHA for 7c5871b

• make lint happy?

  malkomes

  committed
  fc9e7ee

  Copy full SHA for fc9e7ee

• Change torch dtype to bflat16 for qwen2.5-VL test

  

  jiminha

  authored and

  malkomes

  committed
  67b696e

  Copy full SHA for 67b696e

• fea(): Added the tests requirements

  Show description for cf97bed
  

  imangohari1

  authored and

  malkomes

  committed
  cf97bed

  Copy full SHA for cf97bed

• add check_transformers to qwen2_5_VL

  malkomes

  committed
  c986f8d

  Copy full SHA for c986f8d

• improving code and comments

  malkomes

  committed
  08b35bf

  Copy full SHA for 08b35bf

• lint

  malkomes

  committed
  75eb21b

  Copy full SHA for 75eb21b

• remove Optinal

  malkomes

  committed
  70ef940

  Copy full SHA for 70ef940

• lint qwen2_5_vl

  malkomes

  committed
  15d735c

  Copy full SHA for 15d735c

• add reviewers suggestions

  malkomes

  committed
  f6b95f8

  Copy full SHA for f6b95f8

• lint

  malkomes

  committed
  175a927

  Copy full SHA for 175a927

• remove blank line

  malkomes

  committed
  5baa1ed

  Copy full SHA for 5baa1ed

• input_mrope_positions if/else simplifications

  malkomes

  committed
  7fe109a

  Copy full SHA for 7fe109a

• Enable FusedSDPA for Qwen2.5 VL

  jiminha

  committed
  264676d

  Copy full SHA for 264676d

• Lint fix

  jiminha

  committed
  cb09a4b

  Copy full SHA for cb09a4b