Skip to content

Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.

Commits on main

fs(ggml): fill in arch prefix if necessary (#12646)
Michael Yang committed 1 year ago
d2b63c1
cuda: get driver version after props (#12707)
Daniel Hiltgen committed 1 year ago
5d22953
rocm: give it more time to bootstrap (#12681)
Daniel Hiltgen committed 1 year ago
d245dff
contiguous input per layer (#12686)
Daniel Hiltgen committed 1 year ago
bc1a818
win: more verbose load failures (#12683)
Daniel Hiltgen committed 1 year ago
ba2253d
test: harden scheduler tests (#12662)
Daniel Hiltgen committed 1 year ago
68e04c7
cuda: tidy up CC settings (#12668)
Daniel Hiltgen committed 1 year ago
2706799
65fb3ff
Grace/qwen3 thinking (#12647)
Grace committed 1 year ago
e2a0b24
cuda: bring back CC 5.2 (#12666)
Daniel Hiltgen committed 1 year ago
1813ff8
test: add a few missing embedding models (#12661)
Daniel Hiltgen committed 1 year ago
b531777
vulkan: Get FilterID from Backend for Vulkan (#12655)
Thomas Stocker committed 1 year ago
c744134
fs/ggml: fix function name in comment (#12630)
zhetaicheleba committed 1 year ago
de67057
201d937
openai: make tool call conversion fns public
Devon Rifkin committed 1 year ago
160cecc
CI: Set up temporary opt-out Vulkan support (#12614)
Daniel Hiltgen committed 1 year ago
8b6e5ba
perf: backport cuda iGPU sched spin (#12641)
Daniel Hiltgen committed 1 year ago
75d17fc
8fafc8a
llm: Enable flash attention by default for gemma3
Jesse Gross committed 1 year ago
c3c85aa
types: send index for tool calls (#12625)
Parth Sareen committed 1 year ago
c4c5a4a
53a969d
qwen3-coder: support anyOf when parsing tool calls
Devon Rifkin committed 1 year ago
08fbb60
logs: fix bogus "0 MiB free" log line (#12590)
Daniel Hiltgen committed 1 year ago
850da84
Vulkan based on #9650 (#11835)
Thomas Stocker committed 1 year ago
2aba569
fd8aa94