add rwkv-5-3b model (#666)

* support rwkv5-3b learnboard

* update rwkv-5-3b config

* update config

* refine

* fix bug

* update config

* refine

* reduce batch size

* refine

* reduce batch size to avoid oom in special datasets

* Update huggingface.py

* Update huggingface.py
This commit is contained in:
Xiaoyu Zhang 2023-12-12 18:15:19 +08:00 committed by GitHub
parent 4780b39eda
commit 82a533a690
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 31 additions and 0 deletions

6
configs/eval_rwkv5_3b.py Normal file
View File

@ -0,0 +1,6 @@
from mmengine.config import read_base
with read_base():
from .models.rwkv.rwkv5_3b import models
from .datasets.collections.base_medium_llama import datasets
from .summarizers.leaderboard import summarizer

View File

@ -0,0 +1,25 @@
from opencompass.models import HuggingFaceCausalLM
models = [
dict(
type=HuggingFaceCausalLM,
abbr='rwkv-5-3b',
path="RWKV/rwkv-5-world-3b",
tokenizer_path='RWKV/rwkv-5-world-3b',
model_kwargs=dict(
device_map='auto',
trust_remote_code=True,
),
tokenizer_kwargs=dict(
padding_side='left',
truncation_side='left',
trust_remote_code=True,
use_fast=False,
),
max_out_len=100,
max_seq_len=2048,
batch_padding=True,
batch_size=16,
run_cfg=dict(num_gpus=1, num_procs=1),
)
]