Repository navigation

chore(deps): bump vite-plus to v0.2.2 - #11

Draft

fengmk2 wants to merge 8 commits into

mainfrom

Draft

Commits on Jun 25, 2026
- feat(convert): quantize gemma4_unified (12B) — recognize + keep vision_embedder bf16 (#76)
  
  Brooooooklyn
  and
  claude
  authored
  4be27cf
- feat(gemma4): load tied lm_head as packed-quantized (8-bit affine) (#77)
  
  Brooooooklyn
  and
  claude
  authored
  4a474a9
Commits on Jun 26, 2026
- fix(convert): apply imatrix AWQ on VLM checkpoints + compensate GDN in_proj_a/b (#78)
  
  Brooooooklyn
  and
  claude
  authored
  409a7b5
- test(losses): de-flake test_cross_entropy_qwen3_vocab (batch 2 → 64) (#79)
  
  Brooooooklyn
  and
  claude
  authored
  ed672d1
Commits on Jun 27, 2026
- fix(qwen3.5-VL): correct image inference (interleaved M-RoPE, patch-embed, compressed-position decode + delta lifetime) (#80)
  
  Brooooooklyn
  and
  claude
  authored
  855b91c
Commits on Jun 30, 2026
- feat(server): honor stop_sequences in /v1/messages + loosen Qwen repetition cutoff (#81)
  
  Brooooooklyn
  and
  claude
  authored
  73f4a89
- fix: align repetition handling with vLLM (cutoff off by default) (#82)
  
  Brooooooklyn
  and
  claude
  authored
  f65104a
Commits on Jul 2, 2026
- chore(deps): bump vite-plus to v0.2.2
  fengmk2
  committed
  1bdb9f6