Code

Hp/split/load model (test CI)#4

Code

Open

ngxson wants to merge 11 commits intomasterngxson/llama.cpp:masterfrom

hp/split/load-modelngxson/llama.cpp:hp/split/load-modelCopy head branch name to clipboard

Commits

Commits on Mar 20, 2024

• split: support in llama_model_loader

  phymbert

  committed
  7c64fef

  Copy full SHA for 7c64fef

Commits on Mar 21, 2024

• Avoir copying the entire vector

  Show description for b8feff4
  

  phymbert

  and

  slaren

  authored
  b8feff4

  Copy full SHA for b8feff4

• split: move llama_tensor_offset to llama_model_loader

  phymbert

  committed
  18ff6ca

  Copy full SHA for 18ff6ca

• Merge branch 'master' into hp/split/load-model

  phymbert

  committed
  60a87ae

  Copy full SHA for 60a87ae

• llama_model_loader: PR feedbacks:

  Show description for 1892ae7
  phymbert

  committed
  1892ae7

  Copy full SHA for 1892ae7

• avoid copying the entire vector

  phymbert

  committed
  00381b0

  Copy full SHA for 00381b0

• Simplify this by making these optional, switch some layer creation tensor optional

  Show description for c34a5de
  

  phymbert

  and

  ggerganov

  authored
  c34a5de

  Copy full SHA for c34a5de

• Handle optional tensors

  Show description for 1c931f3
  

  phymbert

  and

  ggerganov

  authored
  1c931f3

  Copy full SHA for 1c931f3

• llama_model_loader: fail if backend cannot allocate buffer

  phymbert

  committed
  d8b567d

  Copy full SHA for d8b567d

• fix mmap buffer management

  slaren

  committed
  02020b0

  Copy full SHA for 02020b0

• test new llama_split_prefix

  ngxson

  committed
  9306936

  Copy full SHA for 9306936