Code

llama : fix embeddings#5796

Code

Merged

ggerganov merged 9 commits intomasterggml-org/llama.cpp:masterfrom gg/fix-embeddingsggml-org/llama.cpp:gg/fix-embeddingsCopy head branch name to clipboardMar 4, 2024

+358-133

Commits

Commits on Mar 4, 2024

• llama : fix embeddings

  Show description for d034784
  ggerganov

  committed
  d034784

  Copy full SHA for d034784

• llama : do not use KV cache for non-causal models

  Show description for eb42596
  ggerganov

  committed
  eb42596

  Copy full SHA for eb42596

• embeddings : fix llama_batch_init arg

  ggerganov

  committed
  9bbeb0f

  Copy full SHA for 9bbeb0f

• llama : add pooling switch

  ggerganov

  committed
  e66da35

  Copy full SHA for e66da35

• llama : distinguish token vs sequence embeddings

  Show description for 79e4eed
  ggerganov

  committed
  79e4eed

  Copy full SHA for 79e4eed

• llama : assert pooling tensor

  ggerganov

  committed
  fc9af15

  Copy full SHA for fc9af15

• llama : simplify causal mask condition

  Show description for c23c554
  ggerganov

  committed
  c23c554

  Copy full SHA for c23c554

• llama : assert input batch with pooling enabled

  ggerganov

  committed
  1af2d06

  Copy full SHA for 1af2d06

• readme : update API changes list

  ggerganov

  committed
  7cafaa4

  Copy full SHA for 7cafaa4