transformers - 💡(How to fix) Fix ValueError: Unrecognized model in allenai/Olmo-Hybrid-Instruct-SFT-7B. Should have a `model_type` key in its config.json. [4 comments, 3 participants]

from transformers import AutoConfig AutoConfig.from_pretrained('allenai/Olmo-Hybrid-Instruct-SFT-7B') Traceback (most recent call last): File "<stdin>", line 1, in <module> AutoConfig.from_pretrained('allenai/Olmo-Hybrid-Instruct-SFT-7B') ~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File ".../transformers/src/transformers/models/auto/configuration_auto.py", line 1471, in from_pretrained raise ValueError( ...<2 lines>... ) ValueError: Unrecognized model in allenai/Olmo-Hybrid-Instruct-SFT-7B. Should have a model_type key in its config.json.

Fix Plan

1. Update Transformers Library

Update the transformers library to the latest version (5.3.0 or later) using pip:

pip install --upgrade transformers

2. Check Model Configuration

Verify that the model configuration is correctly loaded by checking the model_type key in the config.json file:

import json

with open('config.json') as f:
    config = json.load(f)
print(config['model_type'])

This should print the expected model type.

3. Use from_pretrained with config_name and model_name

Instead of using AutoConfig.from_pretrained with the model name, use the config_name and model_name separately:

from transformers import AutoConfig, AutoModelForSequenceClassification

config_name = 'allenai/Olmo-Hybrid-Instruct-SFT-7B'
model_name = 'allenai/Olmo-Hybrid-Instruct-SFT-7B'

config = AutoConfig.from_pretrained(config_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name, config=config)

This should load the model configuration and model weights correctly.

Verification

1. Check that the model configuration is loaded correctly by printing the model_type key.
2. Verify that the model can be loaded and used without errors.

Extra Tips

• Make sure to update the transformers library to the latest version to ensure you have the latest bug fixes and features.
• When loading models from Hugging Face Hub, use the config_name and model_name separately to avoid issues with model configuration loading.
• Always check the model configuration and weights loading by printing or logging relevant information to ensure everything is loaded correctly.