OpenCompass/configs/models/hf_llama/lmdeploy_llama_7b.py

16 lines
440 B
Python

from opencompass.models import TurboMindModel
models = [
dict(
type=TurboMindModel,
abbr='llama-7b-turbomind',
path='huggyllama/llama-7b',
engine_config=dict(session_len=2048, max_batch_size=16, tp=1),
gen_config=dict(top_k=1, temperature=1e-6, top_p=0.9, max_new_tokens=1024),
max_seq_len=2048,
max_out_len=1024,
batch_size=16,
run_cfg=dict(num_gpus=1),
)
]