Code

Add Bamba Model#10909

Code

Draft

fabianlim wants to merge 20 commits intovllm-project:mainvllm-project/vllm:mainfrom fabianlim:bamba-prfabianlim/vllm:bamba-prCopy head branch name to clipboard

+3,271-9

Commits

Commits on Dec 5, 2024

• initial pr without tp fix

  Show description for 62181d5
  fabianlim

  committed
  62181d5

• fix casting in rms norm gated

  Show description for 51bc78c
  fabianlim

  committed
  51bc78c

Commits on Dec 12, 2024

• TP fix

  Show description for 81b93b4
  fabianlim

  committed
  81b93b4

• fix mamba scan invalid address

  Show description for 0f93e4a
  fabianlim

  committed
  0f93e4a

• some fixes and remove unused kernels

  Show description for 742ae79
  fabianlim

  committed
  742ae79

• fmt + lint

  Show description for b2dc5ca
  fabianlim

  committed
  b2dc5ca

• more comments

  Show description for 9ad9e20
  fabianlim

  committed
  9ad9e20

• initial fix for chunked prefill (incomplete)

  Show description for 25bf381
  fabianlim

  committed
  25bf381

• improve comments

  Show description for 43ce07c
  fabianlim

  committed
  43ce07c

• do not attach seq_idx to attn_metadata

  Show description for 80f14b5
  fabianlim

  committed
  80f14b5

• activate initial states for chunked prefill

  Show description for 6b8ac49
  fabianlim

  committed
  6b8ac49

Commits on Dec 13, 2024

• reuse softplus and remove triton2 remark

  Show description for d788db6
  fabianlim

  committed
  d788db6

• add comment on weight loader and format

  Show description for 400db27
  fabianlim

  committed
  400db27

• rename test_jamba to test_hybrid and got rid of test_bamba

  Show description for bda8ea7
  fabianlim

  committed
  bda8ea7

Commits on Dec 16, 2024

• Merge remote-tracking branch 'upstream/main' into bamba-pr

  fabianlim

  committed
  66078d6

• update bamba to ishybrid and support pp

  Show description for a74de9f
  fabianlim

  committed
  a74de9f

• lint

  Show description for b44caa7
  fabianlim

  committed
  b44caa7

• add unit test for mamba ssd

  Show description for 8cf3644
  fabianlim

  committed
  8cf3644

• fix lint

  Show description for e375b40
  fabianlim

  committed
  e375b40

Commits on Dec 21, 2024

• full chunked-prefill fix (sans unit tests)

  Show description for dcbae7b
  fabianlim

  committed
  dcbae7b