Fix for issue 83

[horheynm]

SUMMARY:
Fixing issue 83. #83

Issue is that when a Phi-3-medium-128k-instruct model is generated, and lm_eval is run, lm_eval gives error bc some files that are expected are not included. These are python files that can be found in the model cache folder. lm_eval will fail if not provided.

Solution was to add the logic in Trainer, a common place for oneshot, etc. - a shared pathway.
So far only Phi-3-medium-128k-instruct has python files in its cache folder, so if they exist, copy them to the output directory.

TEST PLAN:
Ran

import torch
from transformers import AutoTokenizer

from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot
from llmcompressor.transformers.compression.helpers import (  # noqa
    calculate_offload_device_map,
    custom_offload_device_map,
)

# define a llmcompressor recipe for FP8 quantization
# this recipe requires no calibration data since inputs are dynamically quantized
recipe = """
quant_stage:
    quant_modifiers:
        QuantizationModifier:
            ignore: ["lm_head"]
            config_groups:
                group_0:
                    weights:
                        num_bits: 8
                        type: float
                        strategy: channel
                        dynamic: false
                        symmetric: true
                    input_activations:
                        num_bits: 8
                        type: float
                        strategy: token
                        dynamic: true
                        symmetric: true
                    targets: ["Linear"]
"""

model_stub = "meta-llama/Meta-Llama-3-70B-Instruct"

# determine which layers to offload to cpu based on available resources
device_map = calculate_offload_device_map(
    model_stub, reserve_for_hessians=False, num_gpus=1, torch_dtype=torch.float16
)

# alternatively, specify the maximum memory to allocate per GPU directly
# device_map = custom_offload_device_map(
#    model_stub, max_memory_per_gpu="10GB", num_gpus=2, torch_dtype=torch.float16
# )

model = SparseAutoModelForCausalLM.from_pretrained(
    model_stub, torch_dtype=torch.float16, device_map=device_map
)

output_dir = "./test_output_llama3b_70b_fp8"


oneshot(
    model=model,
    recipe=recipe,
    output_dir=output_dir,
    save_compressed=True,
    tokenizer=AutoTokenizer.from_pretrained(model_stub),
)

with and without the commit. python should be included and flow should be the same.

[Satrat]

I don't love saving the python files on init rather than save. This seems like in should belong in src/llmcompressor/transformers/finetune/session_mixin.py:save_model instead, where we save the model itself and tokenizer. Or I suppose it could also go in src/llmcompressor/transformers/sparsification/compressed_tensors_utils.py:modify_save_pretrained

[horheynm]

I don't love saving the python files on init rather than save. This seems like in should belong in src/llmcompressor/transformers/finetune/session_mixin.py:save_model instead, where we save the model itself and tokenizer

I dont mind doing that, but with the current flow it doesnt go to save_model

[horheynm]

src/llmcompressor/transformers/sparsification/compressed_tensors_utils.py

Ok it does go to modify_save_pretrained, can save it there