Code

IQ1_XS FTYPE quant strategy#6310

Code

Closed

Nexesenex wants to merge 21 commits intoggerganov:masterggerganov/llama.cpp:masterfrom Nexesenex:Nexesenex-IQ1_XS-IQ1_S-quant-strategiesNexesenex/croco.cpp:Nexesenex-IQ1_XS-IQ1_S-quant-strategiesCopy head branch name to clipboard

+52-5

Commits

Commits on Mar 25, 2024

• b2532

  Show description for f4949bc
  Nexesenex

  authored
  f4949bc

• Update quantize.cpp - Quant option IQ1_XS

  Nexesenex

  authored
  8f7a7ee

• Update llama.h - Enum IQ1_XS

  Nexesenex

  authored
  3d88431

• Update llama.cpp - Case IQ1_XS

  Nexesenex

  authored
  8eff402

• Update llama.cpp - Fix possible typo

  Show description for 51ff04e
  Nexesenex

  authored
  51ff04e

• Update llama.cpp - Embeddings and output tensors strategy.

  Nexesenex

  authored
  1c4da5d

• Update llama.cpp - Non-FFN layer-tensors strategy

  Nexesenex

  authored
  ddc7701

• Update llama.h - change IQ1_XS enum number

  Show description for b355333
  Nexesenex

  authored
  b355333

• Update llama.cpp - adjustements non-FFN layer tensors

  Nexesenex

  authored
  066efbb

• Update llama.cpp - correction wrong case declaration

  Nexesenex

  authored
  3031c01

Commits on Mar 26, 2024

• Update quantize.cpp - mix label

  Nexesenex

  authored
  9c27b0e

• Update llama.cpp - correction embd.weight GQA-4 & qkv.weight to K-Quants

  Show description for f162b2e
  Nexesenex

  authored
  f162b2e

• Update llama.cpp typo

  Nexesenex

  authored
  62c1f5b

• Update llama.cpp - remove trailing space

  Nexesenex

  authored
  d183936

• Update llama.cpp - exception for the IQ2_S token embedding error

  Nexesenex

  authored
  eaf9571

• Update llama.cpp - switch from IQ4_XS to Q4_K in related cases.

  Show description for 599a4b2
  Nexesenex

  authored
  599a4b2

• Merge branch 'master' into Nexesenex-IQ1_XS-IQ1_S-quant-strategies

  Nexesenex

  authored
  915be46

Commits on Mar 29, 2024

• Update llama.cpp

  Show description for ed4be6b
  Nexesenex

  authored
  ed4be6b

Commits on Apr 1, 2024

• Update llama.cpp - adjustements

  Show description for dce3e27
  Nexesenex

  authored
  dce3e27

Commits on May 10, 2024

• Merge branch 'master' into Nexesenex-IQ1_XS-IQ1_S-quant-strategies

  mofosyne

  authored
  3a83878

• Update llama.h respect current numerology

  Nexesenex

  authored
  e4ac8ae