Code

enable LoRA for embedding models#821

Code

Open

skaulintel wants to merge 32 commits intohabana_mainHabanaAI/vllm-fork:habana_mainfrom dev/skaul_enable_lora_embedHabanaAI/vllm-fork:dev/skaul_enable_lora_embedCopy head branch name to clipboard

+88-70

Commits

Commits on Jan 24, 2025

• Initial draft to enable embedding task.

  libinta

  committed
  56b42c3

  Copy full SHA for 56b42c3

• remove ENCODER_ONLY

  libinta

  committed
  b62d611

  Copy full SHA for b62d611

Commits on Jan 27, 2025

• Added support for embedding model with self attention without causal such as bert,roberta, bart.

  libinta

  committed
  a647baa

  Copy full SHA for a647baa

• Change set_attn_bias padding element from -math.inf to -3e38 as -math.inf caused nan for softmax calculation.

  libinta

  committed
  46e1aad

  Copy full SHA for 46e1aad

• rewrite is_causal and add dbg msg

  libinta

  committed
  2f74e6b

  Copy full SHA for 2f74e6b

• update maskoff value

  libinta

  committed
  99947c8

  Copy full SHA for 99947c8

Commits on Jan 28, 2025

• fix wrong base mask

  libinta

  committed
  094294c

  Copy full SHA for 094294c

Commits on Jan 29, 2025

• cleanup code

  libinta

  committed
  1c7416f

  Copy full SHA for 1c7416f

• cleanup code

  libinta

  committed
  c6cdae1

  Copy full SHA for c6cdae1

• cleanup code

  libinta

  committed
  8ac281b

  Copy full SHA for 8ac281b

Commits on Jan 30, 2025

• Add pooler support for padded batch inputs for hpu with CLSPoll, Lastpool

  libinta

  committed
  e72c2f0

  Copy full SHA for e72c2f0

• add meanpool for padded input

  libinta

  committed
  7c1c74b

  Copy full SHA for 7c1c74b

• revert bert change

  libinta

  committed
  5c49ca1

  Copy full SHA for 5c49ca1

• modify meanpool for padded input

  libinta

  committed
  ae6fbe0

  Copy full SHA for ae6fbe0

Commits on Jan 31, 2025

• write is_pooler function

  libinta

  committed
  d65340a

  Copy full SHA for d65340a

• fix is_causal logic

  libinta

  committed
  0c28519

  Copy full SHA for 0c28519

Commits on Feb 1, 2025

• Set is_causal based on attn_type

  libinta

  committed
  1fe398f

  Copy full SHA for 1fe398f

• Set is_causal based on attn_type

  libinta

  committed
  c3a92f3

  Copy full SHA for c3a92f3

Commits on Feb 3, 2025

• enable lora embedding models on hpu

  skaulintel

  committed
  afe8bb3

  Copy full SHA for afe8bb3

Commits on Feb 4, 2025

• fix with warmup issue

  libinta

  committed
  55ae676

  Copy full SHA for 55ae676

Commits on Feb 5, 2025

• fix cpu test issue and format

  libinta

  committed
  787700b

  Copy full SHA for 787700b

• fix code format

  libinta

  committed
  6f02b86

  Copy full SHA for 6f02b86

• Merge branch 'habana_main' into dev/enable_embedding_ace

  libinta

  authored
  b97f7c6

  Copy full SHA for b97f7c6

• fix hpu attn coding issue

  libinta

  committed
  593ded0

  Copy full SHA for 593ded0

• fix hpu_pooling_model_runner.py code format and add requirement-hpu with vllm-hpu-extension change

  libinta

  committed
  30f43b5

  Copy full SHA for 30f43b5

Commits on Feb 6, 2025

• Merge branch 'dev/enable_embedding_ace' into dev/skaul_enable_lora_embed

  skaulintel

  committed
  7dc5239

  Copy full SHA for 7dc5239

• move create lora mask

  skaulintel

  committed
  c636da7

  Copy full SHA for c636da7

Commits on Feb 7, 2025

• add support for batch padding

  libinta

  committed
  1185c2e

  Copy full SHA for 1185c2e

• Merge branch 'dev/enable_embedding_ace' into dev/skaul_enable_lora_embed

  skaulintel

  committed
  82a6e70

  Copy full SHA for 82a6e70

Commits on Feb 12, 2025

• Merge branch 'habana_main' into dev/enable_embedding_ace

  kzawora-intel

  authored
  53f94e0

  Copy full SHA for 53f94e0

• Merge branch 'dev/enable_embedding_ace' into dev/skaul_enable_lora_embed

  skaulintel

  committed
  05ecf57

  Copy full SHA for 05ecf57

Commits on Feb 19, 2025

• Merge branch 'habana_main' into dev/skaul_enable_lora_embed

  skaulintel

  authored
  8d8f1b2

  Copy full SHA for 8d8f1b2