Commit History
19 days ago
-
Card: single-request latency and the two settings behind it
Verified signature5ceb59c -
Server: CUDA graphs for prompt-sized steps (70 -> 37 ms), --gdn-backend, Server-Timing
Verified signature89d3a1b
20 days ago
-
vllm_plugin: host-side lm_head under TP, int32 id fix, --flashinfer-autotune (auto off for TP>1)
Verified signature27798c6
22 days ago
-
Upload clef-flash-NVFP4 (vLLM plugin + SystemOne server)
Verified signature771ba1d -
initial commit
Verified signature827aa8b