codreview updates

Signed-off-by: Brian Dellabetta <[email protected]>

src/llmcompressor/transformers/tracing/debug.py

@@ -12,6 +12,10 @@
 from llmcompressor.transformers import TextGenerationDataset
 from llmcompressor.args import DatasetArguments
 
+__all__ = [
+    "get_model_class"
+]
+
 
 def parse_args():
     parser = argparse.ArgumentParser(description="Trace a model into subgraphs")

tests/e2e/e2e_utils.py

@@ -18,13 +18,9 @@ def _load_model_and_processor(
 ):
     pretrained_model_class = get_model_class(model_class)
     loaded_model = pretrained_model_class.from_pretrained(
-        model,
-        device_map=device,
-        torch_dtype="auto",
-        trust_remote_code=True,
-        _attn_implementation="eager",
+        model, device_map=device, torch_dtype="auto"
     )
-    processor = AutoProcessor.from_pretrained(model, trust_remote_code=True)
+    processor = AutoProcessor.from_pretrained(model)
     return loaded_model, processor
 
 

tests/lmeval/configs/vl_w4a16_actorder_weight.yaml

@@ -15,5 +15,5 @@ lmeval:
   num_fewshot: 0
   limit: 1000
   metrics:
-    acc,none: 0.333
+    acc,none: 0.4
   batch_size: 4