32
tests/e2e/models/configs/Mixtral-8x7B-Instruct-v0.1.yaml
Normal file
32
tests/e2e/models/configs/Mixtral-8x7B-Instruct-v0.1.yaml
Normal file
@@ -0,0 +1,32 @@
|
||||
model_name: "mistralai/Mixtral-8x7B-Instruct-v0.1"
|
||||
hardware: "Atlas A2 Series"
|
||||
|
||||
serve:
|
||||
tensor_parallel_size: 4
|
||||
dtype: bfloat16
|
||||
max_model_len: 4096
|
||||
gpu_memory_utilization: 0.7
|
||||
trust_remote_code: true
|
||||
enforce_eager: true
|
||||
block_size: 128
|
||||
|
||||
envs:
|
||||
HCCL_OP_EXPANSION_MODE: "AIV"
|
||||
OMP_PROC_BIND: "false"
|
||||
OMP_NUM_THREADS: "10"
|
||||
VLLM_USE_V1: "1"
|
||||
HCCL_BUFFSIZE: "200"
|
||||
VLLM_ASCEND_ENABLE_MLAPO: "1"
|
||||
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
||||
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
||||
|
||||
tasks:
|
||||
- name: "ceval-valid"
|
||||
metrics:
|
||||
- name: "acc,none"
|
||||
value: 0.45
|
||||
|
||||
num_fewshot: 5
|
||||
apply_chat_template: true
|
||||
fewshot_as_multiturn: false
|
||||
batch_size: 32
|
||||
Reference in New Issue
Block a user