fea(): WA for run_clm config imports

[imangohari1]

What does this PR do?

There is bug in run_clm.py that instead of loading the local OH configuration, it is loading the upstream HF here:

MODEL_CONFIG_CLASSES = list(MODEL_FOR_CAUSAL_LM_MAPPING.keys())
MODEL_TYPES = tuple(conf.model_type for conf in MODEL_CONFIG_CLASSES)

This is WA to fix it.

Reproducer

Applying this patch

diff --git a/optimum/habana/transformers/models/mistral/configuration_mistral.py b/optimum/habana/transformers/models/mistral/configuration_mistral.py
index 9af0c897..e273509b 100644
--- a/optimum/habana/transformers/models/mistral/configuration_mistral.py
+++ b/optimum/habana/transformers/models/mistral/configuration_mistral.py
@@ -56,6 +56,7 @@ class MistralConfig(MistralConfig):
             **kwargs,
         )
 
+        breakpoint()
         self.rope_scaling = rope_scaling
 
         # Validate the correctness of rotary position embeddings parameters

and run cmd

make test_instals; python3 examples/language-modeling/run_clm.py --model_name_or_path mistralai/Mistral-7B-v0.1 --gaudi_config_name Habana/gpt2 --dataset_name wikitext --do_train --output_dir /tmp/tmp3vtnetji --overwrite_output_dir --learning_rate 0.0002 --per_device_train_batch_size 4 --per_device_eval_batch_size 4 --num_train_epochs 2 --use_habana --throughput_warmup_steps 3 --save_strategy no --use_lazy_mode --do_eval --dataset_config_name wikitext-2-raw-v1 --use_hpu_graphs_for_inference

Results

• with this PR: the code load local OH mistral and breaks (expected)
• without this PR: the code loads the HF mistral and doesn't break (not expected)

Notes:

• I've checked the models where we have the configuration overwrite and cross-referenced them with HF CLM and MLM. We only have configuration overwrite for CLM models.
• This issue has not been noticed since none of the config var changes are not directly used in modeling_XX.py files.
• There might be a better, more comprehensive solution this. Therefore this PR starts with draft.

--
co-authored by Yaser Afshar @yafshar

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[imangohari1]

CC: @libinta
Hi @regisss,
can you take a look at this and suggest what is the best way to handle this? the current PR works fine, but I would love to hear your inputs. Thank you.

[regisss]

I think the current change is fine

[imangohari1]

I think the current change is fine

@regisss thanks. I made this PR ready then.

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[regisss]

LGTM