Luodian / Otter

🦦 Otter, a multi-modal model based on OpenFlamingo (open-sourced version of DeepMind's Flamingo), trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability.
https://otter-ntu.github.io/
MIT License
3.54k stars 242 forks source link

[FlamingoForConditionalGeneration] Loading huggingface models #251

Open llnns opened 1 year ago

llnns commented 1 year ago

The following code to load "luodian/openflamingo-9b-hf" for ConditionalGeneration:

from flamingo import FlamingoForConditionalGeneration
model = FlamingoForConditionalGeneration.from_pretrained("luodian/openflamingo-9b-hf")

Resulted in:

Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/lib/python3.9/dist-packages/transformers/modeling_utils.py", line 2700, in from_pretrained
    model = cls(config, *model_args, **model_kwargs)
  File "/Otter/flamingo/modeling_flamingo.py", line 713, in __init__
    if config.text_config.architectures[0] == "MPTForCausalLM":
TypeError: 'NoneType' object is not subscriptable

Based on the FlamingoModel Class, I noticed the following comment line, when I uncomment it, the model loads correctly.

Luodian commented 1 year ago

ok I see it's a bug from our side. we overlooked some config files dont have this attribute.

Could you start with luodian/OTTER-MPT1B-RPJama-Init?

Luodian commented 1 year ago

This one should be fine since it's our latest used model.