SmoothQuant Mappings Only Work When Defined in a Recipe String

[Satrat]

Describe the bug
When a oneshot recipe is defined programmatically, the SmoothQuant mappings are not correctly parsed. This bug does not occur when the recipe is specified as a YAML string or file

Expected behavior
SmoothQuantModifier is correctly initialized, and oneshot runs to completion.

Environment
Include all relevant environment information:

1. OS [e.g. Ubuntu 18.04]: Ubuntu
2. Python version [e.g. 3.7]: 3.10.12
3. LLM Compressor version or commit hash [e.g. 0.1.0, f7245c8]: main, 07c1fd7
4. ML framework version(s) [e.g. torch 1.7.1]: torch 2.3.1, transformers 4.42.4
5. Other Python package versions [e.g. SparseZoo, DeepSparse, numpy, ONNX]: n/a
6. Other relevant environment information [e.g. hardware, CUDA version]: CUDA 12.3

To Reproduce
Example script:

from llmcompressor.modifiers.smoothquant.base import DEFAULT_SMOOTHQUANT_MAPPINGS
from llmcompressor.modifiers.smoothquant import SmoothQuantModifier
from llmcompressor.modifiers.quantization.gptq import GPTQModifier

from llmcompressor.transformers import SparseAutoModelForCausalLM, oneshot

recipe = [
    SmoothQuantModifier(smoothing_strength=0.8, mappings=DEFAULT_SMOOTHQUANT_MAPPINGS),
    GPTQModifier(targets="Linear", scheme="W8A8", ignore=["lm_head"], sequential_update=False),
]

model_stub = "TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T"
model = SparseAutoModelForCausalLM.from_pretrained(model_stub, device_map="auto", torch_dtype="auto")

dataset = "ultrachat-200k"
output_dir = "./test_output"
splits = {"calibration": "train_gen[:5%]"}
max_seq_length = 2048
pad_to_max_length = False
num_calibration_samples = 8

oneshot(
    model=model,
    dataset=dataset,
    recipe=recipe,
    output_dir=output_dir,
    splits=splits,
    max_seq_length=max_seq_length,
    pad_to_max_length=pad_to_max_length,
    num_calibration_samples=num_calibration_samples,
    save_compressed=True
)

Errors
oneshot fails with the following error:

Could not parse recipe from string DEFAULT_stage:
  DEFAULT_modifiers:
    SmoothQuantModifier:
      index: null
      group: null
      start: -1
      end: -1
      update: null
      initialized_structure_: false
      initialized_: false
      finalized_: false
      started_: false
      ended_: false
      smoothing_strength: 0.8
      mappings:
      - !!python/tuple
        - - re:.*q_proj
          - re:.*k_proj
          - re:.*v_proj
        - re:.*input_layernorm
      - !!python/tuple
        - - re:.*gate_proj
          - re:.*up_proj
        - re:.*post_attention_layernorm

Additional context
Updating the recipe to the equivalent yaml string fixes the issue:

recipe = """
DEFAULT_stage:
  DEFAULT_modifiers:
    SmoothQuantModifier:
      smoothing_strength: 0.8
      mappings:
      - - ['re:.*q_proj', 're:.*k_proj', 're:.*v_proj']
        - re:.*input_layernorm
      - - ['re:.*gate_proj', 're:.*up_proj']
        - re:.*post_attention_layernorm
    GPTQModifier:
      sequential_update: false
      targets: Linear
      scheme: W8A8
"""

[rahul-tuli]

Resolved by PR #48