Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit history

Commits on May 20, 2025

• Set GLM4 blk.*.attn_output.weight, kqv_out-* matmul to GGML_PREC_F32 to fix infinity values in output (#13639)

  0cc4m

  authored
  c9c64de

  Copy full SHA for c9c64de

• kv-cache : add SWA support (#13194)

  Show description for e298d2f
  ggerganov

  authored
  e298d2f

  Copy full SHA for e298d2f

Commits on May 12, 2025

• llama/ggml: add LLM training support (#10544)

  Show description for 10d2af0
  JohannesGaessler

  authored
  10d2af0

  Copy full SHA for 10d2af0

Commits on May 9, 2025

• CUDA: FA support for Deepseek (Ampere or newer) (#13306)

  Show description for 0cf6725
  JohannesGaessler

  authored
  0cf6725

  Copy full SHA for 0cf6725

Commits on May 6, 2025

• llama : fix build_ffn without gate (#13336)

  Show description for 2f54e34
  ngxson

  authored
  2f54e34

  Copy full SHA for 2f54e34

Commits on May 2, 2025

• kv-cache : separate recurrent vs non-recurrent impl (#12799)

  Show description for c642bc0
  ggerganov

  authored
  c642bc0

  Copy full SHA for c642bc0

Commits on Apr 29, 2025

• llama-graph : fix text position for mrope (#13159)

  Show description for b6ce743
  ngxson

  authored
  b6ce743

  Copy full SHA for b6ce743

Commits on Apr 28, 2025

• model : Nomic Embed Text V2 with Mixture-of-Experts (MoE) architecture (#12466)

  Show description for 5f5e39e
  

  manyoso

  and

  cebtenzzre

  authored
  5f5e39e

  Copy full SHA for 5f5e39e

• llama : (mrope) allow using normal 1D position for text token (#13138)

  Show description for d2b2031
  ngxson

  authored
  d2b2031

  Copy full SHA for d2b2031

Commits on Apr 25, 2025

• Force FP32 compute in GLM4 FFN Down (#13101)

  Show description for 558a764
  

  city96

  and

  JohannesGaessler

  authored
  558a764

  Copy full SHA for 558a764

Commits on Apr 17, 2025

• graph : make FA compatible with MLA + add initial Metal kernels (#12953)

  Show description for 2f74c35
  ggerganov

  authored
  2f74c35

  Copy full SHA for 2f74c35

Commits on Apr 15, 2025

• llama : DeepSeek V2/V3 MLA implementation (#12801)

  Show description for daa4228
  jukofyork

  authored
  daa4228

  Copy full SHA for daa4228

Commits on Apr 8, 2025

• llama : fix FA when KV cache is not used (i.e. embeddings) (#12825)

  Show description for a19b5ce
  ggerganov

  authored
  a19b5ce

  Copy full SHA for a19b5ce

Commits on Apr 7, 2025

• llama : Support llama 4 text-only (#12791)

  Show description for 1466621
  ngxson

  authored
  1466621

  Copy full SHA for 1466621

Commits on Mar 29, 2025

• llama : fix non-causal mask for gemma 3 (#12615)

  ngxson

  authored
  af6ae1e

  Copy full SHA for af6ae1e

Commits on Mar 18, 2025

• graph : normalize Q, K, V shapes + sync cross attention (#12449)

  Show description for 75422e8
  ggerganov

  authored
  75422e8

  Copy full SHA for 75422e8

Commits on Mar 14, 2025

• Load all MoE experts during warmup (#11571)

  Show description for 8fcb563
  

  fairydreaming

  and

  sszymczy

  authored
  8fcb563

  Copy full SHA for 8fcb563

• graph : simplify attn input build for unified KV cache (#12381)

  Show description for c522ce4
  ggerganov

  authored
  c522ce4

  Copy full SHA for c522ce4

• hparams : add SWA rope parameters (#12374)

  Show description for 081bee8
  ggerganov

  authored
  081bee8

  Copy full SHA for 081bee8

Commits on Mar 13, 2025

• llama : fix Gemma3 SWA KV cache shift (#12373)

  Show description for 84d5475
  ggerganov

  authored
  84d5475

  Copy full SHA for 84d5475

• llama : refactor llama_context, llama_kv_cache, llm_build_context (#12181)

  Show description for e0dbec0
  ggerganov

  authored
  e0dbec0

  Copy full SHA for e0dbec0

Loading