[Bugfix] Remove tracing imports from tests (#1498)

## Purpose ##
* We no longer use traceable definitions as of #1411, so we should no
longer be importing these definitions in the tests

## Changes ##
* Remove traceable model definition imports from test configs
* Remove `get_model_class` utility, which is no longer necessary since
we only import model classes from transformers now

## Testing ##
```bash
grep -r -i 'trace' src tests examples
```

---------

Signed-off-by: Kyle Sayers <kylesayrs@gmail.com>
Co-authored-by: Dipika Sikka <dipikasikka1@gmail.com>

Author: kylesayrs

src/llmcompressor/transformers/tracing/__init__.py

@@ -1,3 +1,3 @@
-from .debug import get_model_class
+from .debug import trace
 
-__all__ = ["get_model_class"]
+__all__ = ["trace"]

src/llmcompressor/transformers/tracing/debug.py

@@ -7,15 +7,14 @@
 import transformers
 from transformers import AutoProcessor, PreTrainedModel
 
-from llmcompressor.transformers import tracing
 from llmcompressor.utils.pytorch.module import get_no_split_params
 from llmcompressor.pipelines.sequential.helpers import trace_subgraphs, Subgraph
 from llmcompressor.transformers import TextGenerationDataset
 from llmcompressor.args import DatasetArguments
 
 from llmcompressor.utils.dev import skip_weights_download
 
-__all__ = ["get_model_class"]
+__all__ = ["trace"]
 
 
 def parse_args():
@@ -111,14 +110,6 @@ def trace(
     return model, subgraphs, sample
 
 
-def get_model_class(model_class: str) -> Type[PreTrainedModel]:
-    model_cls = getattr(tracing, model_class, getattr(transformers, model_class, None))
-    if model_cls is None:
-        raise ValueError(f"Could not import model class {model_class}")
-
-    return model_cls
-
-
 def get_dataset_kwargs(modality: str, ignore: List[str]) -> Dict[str, str]:
     dataset_kwargs = {
         "text": {
@@ -167,7 +158,7 @@ def main():
 
     trace(
         model_id=args.model_id,
-        model_class=get_model_class(args.model_class),
+        model_class=getattr(transformers, args.model_class),
         sequential_targets=args.sequential_targets,
         ignore=args.ignore,
         modality=args.modality,

tests/e2e/e2e_utils.py

@@ -1,11 +1,11 @@
 import torch
+import transformers
 from datasets import load_dataset
 from loguru import logger
 from transformers import AutoProcessor
 
 from llmcompressor import oneshot
 from llmcompressor.modifiers.quantization import GPTQModifier, QuantizationModifier
-from llmcompressor.transformers.tracing import get_model_class
 from tests.test_timer.timer_utils import log_time
 from tests.testing_utils import process_dataset
 
@@ -16,7 +16,7 @@ def _load_model_and_processor(
     model_class: str,
     device: str,
 ):
-    pretrained_model_class = get_model_class(model_class)
+    pretrained_model_class = getattr(transformers, model_class)
     loaded_model = pretrained_model_class.from_pretrained(
         model, device_map=device, torch_dtype="auto"
     )

tests/lmeval/configs/vl_fp8_dynamic_per_token.yaml

@@ -1,6 +1,6 @@
 cadence: weekly
 model: Qwen/Qwen2.5-VL-7B-Instruct
-model_class: TraceableQwen2_5_VLForConditionalGeneration
+model_class: Qwen2_5_VLForConditionalGeneration
 scheme: FP8_DYNAMIC
 lmeval:
   model: "hf-multimodal"

tests/lmeval/configs/vl_int8_w8a8_dynamic_per_token.yaml

@@ -1,6 +1,6 @@
 cadence: "weekly"
 model: Qwen/Qwen2.5-VL-7B-Instruct
-model_class: TraceableQwen2_5_VLForConditionalGeneration
+model_class: Qwen2_5_VLForConditionalGeneration
 scheme: INT8_dyn_per_token
 recipe: tests/e2e/vLLM/recipes/INT8/recipe_int8_channel_weight_dynamic_per_token.yaml
 dataset_id: lmms-lab/flickr30k

tests/lmeval/configs/vl_w4a16_actorder_weight.yaml

@@ -1,6 +1,6 @@
 cadence: "weekly"
 model: Qwen/Qwen2.5-VL-7B-Instruct
-model_class: TraceableQwen2_5_VLForConditionalGeneration
+model_class: Qwen2_5_VLForConditionalGeneration
 scheme: W4A16_actorder_weight
 recipe: tests/e2e/vLLM/recipes/actorder/recipe_w4a16_actorder_weight.yaml
 dataset_id: lmms-lab/flickr30k