Qwen3-235B-A22B-Instruct-2507-FP8.yaml · tduncan/Nvidia-Optimization-Stack-VLLM

main

yaml 17 lines 554 Bytes

Raw

1	model_name: "Qwen/Qwen3-235B-A22B-Instruct-2507-FP8"
2	required_gpu_arch:
3	- gfx942
4	- gfx950
5	tasks:
6	- name: "mmlu_pro"
7	metrics:
8	- name: "exact_match,custom-extract"
9	value: 0.82
10	limit: 250 # will run on 250 * 14 subjects = 3500 samples
11	num_fewshot: 5
12	enforce_eager: false # we use false to speed up the eval process
13	kv_cache_dtype: fp8 # we use fp8 to speed up the eval process
14	max_model_len: 40960
15	apply_chat_template: true
16	fewshot_as_multiturn: true
17	gen_kwargs: "temperature=0,top_p=1,top_k=0,max_gen_toks=5632,until=<\|ENDANSWER\|>"