fix module mapping error of baichuan model

[grimoire]

python -m lmdeploy.pytorch.chat \
    /path/to/baichuan2/Baichuan2-7B-Chat/ \
    baichuan2-7b

python -m lmdeploy.pytorch.chat \
    /path/to/baichuan2/Baichuan2-7B-Chat/ \
    baichuan2-7b

leads to different result.

[lvhan028]

This is baichuan1 or baichuan2?

[AllentDan]

Why not just remove the last char /?

[grimoire]

This is baichuan1 or baichuan2?

baichuan2

[grimoire]

Why not just remove the last char /?

Inorder to match renamed model directory.

[lvhan028]

baichuan 1 or baichuan 2

[zhulinJulia24]

命令

lmdeploy convert baichuan2-7b /nvme/qa_test_models/Baichuan2-7B-Chat --dst-path /nvme/qa_test_models/autotest_model/workspace_Baichuan2-7B-Chat

报错

/usr/local/lib/python3.8/dist-packages/fuzzywuzzy/fuzz.py:11: UserWarning: Using slow pure-python SequenceMatcher. Install python-Levenshtein to remove this warning
warnings.warn('Using slow pure-python SequenceMatcher. Install python-Levenshtein to remove this warning')
Traceback (most recent call last):
File "/usr/local/bin/lmdeploy", line 8, in
sys.exit(run())
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/cli/entrypoint.py", line 18, in run
args.run(args)
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/cli/cli.py", line 97, in convert
main(**kwargs)
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/converter.py", line 286, in main
output_model = OUTPUT_MODELS.get(output_format)(input_model=input_model,
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/target_model/fp.py", line 26, in init
super().init(input_model, cfg, to_file, out_dir)
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/target_model/base.py", line 141, in init
self.cfg = self.get_config(cfg)
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/target_model/fp.py", line 30, in get_config
final_cfg = super().get_config(cfg).dict
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/target_model/base.py", line 160, in get_config
for bin in self.input_model.bins():
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/source_model/base.py", line 174, in bins
for mgr in self.get_mgrs():
File "/usr/local/lib/python3.8/dist-packages/lmdeploy/turbomind/deploy/source_model/llama.py", line 163, in get_mgrs
ret = self.Reader(new_params, unused_params,
TypeError: init() takes 4 positional arguments but 5 were given

这个是修复baichuan convert报错问题的么？
我这边尝试了更新了环境内的conda文件，问题还未解决

[grimoire]

baichuan 1 or baichuan 2

baichuan 2

[grimoire]

这个是修复baichuan convert报错问题的么？ 我这边尝试了更新了环境内的conda文件，问题还未解决

不是，这个是修 pytorch 那边的问题