fix format type

[jiqing-feng]

The commit #415 breaks the llm-compressor save_pretrain
To reproduce it, run the llm-compressor example
python llama3_example.py
error:

Traceback (most recent call last):
  File "/home/sdp/jiqing/llm-compressor/examples/awq/llama_example.py", line 74, in <module>
    model.save_pretrained(SAVE_DIR, save_compressed=True)
  File "/home/sdp/jiqing/llm-compressor/src/llmcompressor/transformers/sparsification/compressed_tensors_utils.py", line 87, in save_pretrained_wrapper
    compressor = get_model_compressor(
                 ^^^^^^^^^^^^^^^^^^^^^
  File "/home/sdp/jiqing/llm-compressor/src/llmcompressor/transformers/sparsification/compressed_tensors_utils.py", line 240, in get_model_compressor
    return ModelCompressor.from_pretrained_model(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/venv/lib/python3.12/site-packages/compressed_tensors/compressors/model_compressors/model_compressor.py", line 201, in from_pretrained_model
    return cls(
           ^^^^
  File "/opt/venv/lib/python3.12/site-packages/compressed_tensors/compressors/model_compressors/model_compressor.py", line 317, in __init__
    for format in self.compression_formats:
TypeError: 'CompressionFormat' object is not iterable

The reason is that the llm-compressor pass quantization_format as a CompressionFormat instead of str here.

The BaseCompressor.load_from_registry can only accept str type, so we need to convert it.

After this PR, the save issue can be fixed.

[jiqing-feng]

Hi @dsikka . Would you please review this PR? Thanks!

[dsikka]

Hi @jiqing-feng this is a bug in llmcompressor. It is resolved here: vllm-project/llm-compressor#1713