Commits

History for

Commits on Jul 20, 2026

fix super long context loading issue
zhongkaifu
committed
47e6f67
fix qwen image edit q2 model black output image
zhongkaifu
committed
6a2abae

Commits on Jul 18, 2026

improve cuda backend performance
zhongkaifu
committed
0fb0da7

Commits on Jul 15, 2026

optimize qwen model inference
zhongkaifu
committed
95068d2

Commits on Jul 13, 2026

Remove native debug/timing paths, redundant CPU A/B kill-switches, housekeeping

zhongkaifu
and
claude
committed
4175be0
Remove experimental managed GGML ops path and debug/profiling-only code paths

zhongkaifu
and
claude
committed
b4a6b72

Commits on Jul 12, 2026

fix broken qwen image edit in ggml_metal
zhongkaifu
committed
c4eefe8
support multi-images edit
zhongkaifu
committed
7c5e15d

Commits on Jul 9, 2026

show temporary edited images
zhongkaifu
committed
0789c8d

Commits on Jul 8, 2026

fix RoPE dimension mismatch in Qwen
zhongkaifu
committed
60fffce

Commits on Jul 7, 2026

support PDF format document
zhongkaifu
committed
509d1e1

Commits on Jul 6, 2026

fix broken build
zhongkaifu
committed
7beb0aa
refactor(gemma4): simplify model code, remove dead paths

Adrian
authored and
zhongkaifu
committed
c39f108
feat(qwen35): integrate fused QK-Norm+RoPE kernel into prefill path

Adrian
authored and
zhongkaifu
committed
8999052
improve vulkan performance
zhongkaifu
committed
15334b8
support vulkan backend
zhongkaifu
committed
1709d1c

Commits on Jul 2, 2026

optimize image edit
zhongkaifu
committed
e30b52b

Commits on Jul 1, 2026

prefill optimization
zhongkaifu
committed
4631a25
update
zhongkaifu
committed
65dd2dd
optimize image edit
zhongkaifu
committed
84480be
optimization image edit
zhongkaifu
committed
879b85b

Commits on Jun 29, 2026

optimize gemma 4 prefill
zhongkaifu
committed
0788009
fix parallel requests bugs in Qwen
zhongkaifu
committed
5f2f60d

Commits on Jun 28, 2026

fix KV cache reuse bug
zhongkaifu
committed
1a74206
optimization
zhongkaifu
committed
fabb92c
optimization
zhongkaifu
committed
1d37cdc
optimization
zhongkaifu
committed
80ebbaa
optimize
zhongkaifu
committed
7251e88
optimize Gemma MoE decode
zhongkaifu
committed
9ae16b9
optimize qwen image edit
zhongkaifu
committed
0733c98
support Qwen Image Edit model
zhongkaifu
committed
cf5de09
optimize qwen-vl
zhongkaifu
committed
5194c1e
optimize qwen36 prefill
zhongkaifu
committed
c1a8a40
optimize qwen3.5 decode
zhongkaifu
committed
0ef1c4a
improve gemma4 decode
zhongkaifu
committed
451aaf8