|
1
|
model_name: "Qwen/Qwen3-4B" |
|
2
|
accuracy_threshold: 0.78 |
|
3
|
num_questions: 1319 |
|
4
|
num_fewshot: 5 |
|
5
|
# Triton serves the unquantized boundary layers with TurboQuant-compatible layout. |
|
6
|
server_args: "--kv-cache-dtype turboquant_k3v4_nc --max-model-len 4096 --attention-backend TRITON_ATTN" |