Code

llama : add reranking support#9510

Code

Merged

ggerganov merged 25 commits intomasterggerganov/llama.cpp:masterfrom gg/rerankggerganov/llama.cpp:gg/rerankCopy head branch name to clipboardSep 28, 2024

+602-56

Commits

Commits on Sep 25, 2024

• py : add XLMRobertaForSequenceClassification [no ci]

  ggerganov

  committed
  3453e62

  Copy full SHA for 3453e62

• py : fix scalar-tensor conversion [no ci]

  ggerganov

  committed
  77723ed

  Copy full SHA for 77723ed

• py : fix position embeddings chop [no ci]

  ggerganov

  committed
  49f90de

  Copy full SHA for 49f90de

• llama : read new cls tensors [no ci]

  ggerganov

  committed
  dc0cdd8

  Copy full SHA for dc0cdd8

• llama : add classigication head (wip) [no ci]

  ggerganov

  committed
  d0a7bf9

  Copy full SHA for d0a7bf9

• llama : add "rank" pooling type

  Show description for 125a067
  ggerganov

  committed
  125a067

  Copy full SHA for 125a067

• server : add rerank endpoint

  Show description for 6235c62
  ggerganov

  committed
  6235c62

  Copy full SHA for 6235c62

• llama : aboud ggml_repeat during classification

  ggerganov

  committed
  6916ed1

  Copy full SHA for 6916ed1

• rerank : cleanup + comments

  ggerganov

  committed
  62a45d1

  Copy full SHA for 62a45d1

• server : accept /rerank endpoint in addition to /v1/rerank [no ci]

  ggerganov

  committed
  7bde9a0

  Copy full SHA for 7bde9a0

• embedding : parse special tokens

  ggerganov

  committed
  c62a39d

  Copy full SHA for c62a39d

• jina : support v1 reranker

  ggerganov

  committed
  866c011

  Copy full SHA for 866c011

• vocab : minor style

  Show description for 84f56f3
  ggerganov

  committed
  84f56f3

  Copy full SHA for 84f56f3

Commits on Sep 26, 2024

• server : initiate tests for later

  Show description for 00b3376
  ggerganov

  committed
  00b3376

  Copy full SHA for 00b3376

• server : add docs

  ggerganov

  committed
  877a04c

  Copy full SHA for 877a04c

• llama : add comment [no ci]

  ggerganov

  committed
  4d45775

  Copy full SHA for 4d45775

• llama : fix uninitialized tensors

  ggerganov

  committed
  ca99a6c

  Copy full SHA for ca99a6c

• ci : add rerank tests

  Show description for f19554f
  ggerganov

  committed
  f19554f

  Copy full SHA for f19554f

• add reranking test

  ngxson

  committed
  f27dd69

  Copy full SHA for f27dd69

• change test data

  ngxson

  committed
  1ae8376

  Copy full SHA for 1ae8376

Commits on Sep 27, 2024

• Update examples/server/server.cpp

  Show description for 84b0af8
  

  ggerganov

  and

  ngxson

  authored
  84b0af8

  Copy full SHA for 84b0af8

• add --reranking argument

  ngxson

  committed
  0d6f6a7

  Copy full SHA for 0d6f6a7

• update server docs

  ngxson

  committed
  a4ac45f

  Copy full SHA for a4ac45f

Commits on Sep 28, 2024

• llama : fix comment [no ci]

  Show description for 39167b6
  ggerganov

  committed
  39167b6

  Copy full SHA for 39167b6

• Merge branch 'master' into gg/rerank

  ggerganov

  authored
  aeac876

  Copy full SHA for aeac876