MAD/scripts/vllm/configs/unittest.yaml at develop · ROCm/MAD · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
- benchmark: latency
  model: meta-llama/Llama-3.1-8B-Instruct amd/Llama-3.1-8B-Instruct-FP8-KV
  tp: 1
  inp:  512
  out: 128
  dtype: auto
  bs: 1 16
  env:
    VLLM_ROCM_USE_AITER: 1
  extra_args:
    --load-format: dummy
  arch_overrides:
    gfx942:
      dtype: float16

- benchmark: throughput
  model: meta-llama/Llama-3.1-8B-Instruct amd/Llama-3.1-8B-Instruct-FP8-KV
  tp: 1
  inp: 128
  out: 128
  dtype: auto
  num_prompts: 10
  env:
    VLLM_ROCM_USE_AITER: 1
  extra_args:
    --load-format: dummy
    --kv-cache-dtype: fp8
  arch_overrides:
    gfx942:
      dtype: float16

- benchmark: serving
  model: meta-llama/Llama-3.1-8B-Instruct amd/Llama-3.1-8B-Instruct-FP8-KV
  tp: 1
  inp: 128
  out: 128
  dtype: auto
  num_prompts: 252
  max_concurrency: 128
  env:
    VLLM_ROCM_USE_AITER: 1
  extra_args:
    --load-format: dummy
  bench_args:
    --random-range-ratio: 0.8
    --lmeval_apply_chat_template: True
  arch_overrides:
    gfx942:
      dtype: float16