testing

Author: HDCharles

log-fp4.log

@@ -7,8 +7,8 @@
 export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/w4a16_channel_quant_moe.yaml"
 pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-int4.log
 
-export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/fp4_nvfp4_moe.yaml"
-pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-fp4.log
+# export TEST_DATA_FILE="${REPOS}/llm-compressor/tests/e2e/vLLM/configs/fp4_nvfp4_moe.yaml"
+# pytest tests/e2e/vLLM/test_vllm.py -vs 2>&1 | tee log-fp4.log
 
 
 

log-int4.log

@@ -77,7 +77,7 @@ def data_collator(batch):
                 }
 
             oneshot_kwargs["data_collator"] = data_collator
-
+    print("CDH", recipe, scheme)
     oneshot_kwargs["model"] = loaded_model
     if recipe:
         oneshot_kwargs["recipe"] = recipe

run.sh

@@ -2,9 +2,9 @@ cadence: "nightly"
 test_type: "regression"
 model: Qwen/Qwen3-VL-30B-A3B-Instruct
 model_class: "Qwen3VLMoeForConditionalGeneration"
-scheme: W4A16
+# scheme: W4A16
 num_calibration_samples: 20
 dataset_id: neuralmagic/calibration
 dataset_split: train
 dataset_config: LLM
-quant_type: "GPTQ"
+recipe: "tests/e2e/vLLM/recipes/WNA16/recipe_w4a16_gptq.yaml"

tests/e2e/e2e_utils.py

@@ -18,7 +18,7 @@ def parse_args():
     except json.JSONDecodeError as e:
         raise ValueError(f"Invalid JSON input: {e}")
 
-    if "W4A16_2of4" in scheme:
+    if scheme is not None and "W4A16_2of4" in scheme:
         # required by the kernel
         llm_kwargs["dtype"] = torch.float16