24
tests/e2e/models/configs/Minitron-8B-Base.yaml
Normal file
24
tests/e2e/models/configs/Minitron-8B-Base.yaml
Normal file
@@ -0,0 +1,24 @@
|
||||
model_name: "nv-community/Minitron-8B-Base"
|
||||
model_type: "vllm"
|
||||
hardware: "Atlas A2 Series"
|
||||
|
||||
serve:
|
||||
tensor_parallel_size: 1
|
||||
dtype: auto
|
||||
max_model_len: 4096
|
||||
gpu_memory_utilization: 0.9
|
||||
enforce_eager: true
|
||||
trust_remote_code: true
|
||||
|
||||
tasks:
|
||||
- name: "gsm8k"
|
||||
metrics:
|
||||
- name: "exact_match,strict-match"
|
||||
value: 0.5436
|
||||
- name: "exact_match,flexible-extract"
|
||||
value: 0.5451
|
||||
|
||||
limit: 1000
|
||||
apply_chat_template: false
|
||||
fewshot_as_multiturn: false
|
||||
batch_size: "auto"
|
||||
Reference in New Issue
Block a user