During our last submission, we successfully go throught the docker building process. And there’s an error during the inference stage when we are attempting to load the large model:
File “/src/models/dummy_model.py”, line 86, in init
File “/src/models/dummy_model.py”, line 86, in init
self.m = LlamaForCausalLM.from_pretrained(model, device_map="balanced", self.m = LlamaForCausalLM.from_pretrained(model, device_map="balanced",File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 3531, in from_pretrained
File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 3531, in from_pretrained
) = cls._load_pretrained_model( ) = cls._load_pretrained_model(File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 3938, in _load_pretrained_model
File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 3938, in _load_pretrained_model
state_dict = load_state_dict(shard_file, is_quantized=is_quantized) state_dict = load_state_dict(shard_file, is_quantized=is_quantized)File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 542, in load_state_dict
File “/home/aicrowd/.conda/lib/python3.8/site-packages/transformers/modeling_utils.py”, line 542, in load_state_dict
raise OSError( raise OSError(OSError: You seem to have cloned a repository without having git-lfs installed. Please install git-lfs and run
git lfs installfollowed bygit lfs pullin the folder you cloned.OSError: You seem to have cloned a repository without having git-lfs installed. Please install git-lfs and run
git lfs installfollowed bygit lfs pullin the folder you cloned.
In the website page of our repository, the model files are correctly uploaded with an LFS tag on the left hand. The log message suggest installing git-lfs. So we made another try.
We add “git-lfs” in the original “apt.txt”, but this time, it doesn’t go through the docker buiding stage.
How can I fix this issue?
Another question is that: what’s the submission number limitations now. Is it 6 times a week, and does that include a failure submission?