Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
Commits on main
create: add server-side MLX imports and drop GGUF conversion (#14969)
98acec4 Daniel Hiltgen committed 2 months ago
mlxrunner: release the buffers weight loading leaves in the MLX pool
b68b112 Jesse Gross committed 2 months ago
mlx: scope array lifetimes instead of pinning and sweeping
13037ec Jesse Gross committed 2 months ago
llm: raise token repeat limit to 100 and return error instead of silently closing (#18374)
4512d2b frob committed 2 months ago
mlxrunner: wait for a killed runner to exit before the scheduler loads the next model
f09d55d Jesse Gross committed 2 months ago
mlxrunner: bound MLX loads by system free memory while other models are loaded
1548f78 Jesse Gross committed 2 months ago
mlxrunner: evict the active conversation's own checkpoints under the budget
6137793 Jesse Gross committed 2 months ago
mlxrunner: keep the reused head of a cached edge safe from eviction
b859a94 Jesse Gross committed 2 months ago
mlxrunner: keep cache boundaries out of non-causal media items
45a0280 Jesse Gross committed 2 months ago
mlxrunner: capture whole-state at close for nodes split out of an edge
8d66f08 Jesse Gross committed 2 months ago
server: extract GGUF metadata and unify capabilities (#17858)
ea8d650 Daniel Hiltgen committed 2 months ago
openai: support standalone named function outputs (#18348)
d8ab4b4 Parth Sareen committed 2 months ago
proxy: normalize namespaced commands in Full Access (#18331)
3738935 Parth Sareen committed 2 months ago
openai: accept plaintext-labeled Codex agent messages (#18329)
86f7292 Parth Sareen committed 2 months ago
openai: finalize responses at the web search limit (#18328)
cd1c5a1 Parth Sareen committed 2 months ago