Set csgmv as default lora backend. (#11488)

test/srt/lora/test_lora.py

@@ -81,13 +81,12 @@ class TestLoRA(CustomTestCase):
         for model_case in model_cases:
             for torch_dtype in TORCH_DTYPES:
                 max_new_tokens = 32
-                backend = "triton"
                 base_path = model_case.base
                 lora_adapter_paths = [a.name for a in model_case.adaptors]
                 assert len(lora_adapter_paths) >= 2
 
                 print(
-                    f"\n========== Testing multiple batches on base '{base_path}' with backend={backend}, dtype={torch_dtype} ---"
+                    f"\n========== Testing multiple batches on base '{base_path}', dtype={torch_dtype} ---"
                 )
 
                 # Initialize runners
@@ -97,7 +96,6 @@ class TestLoRA(CustomTestCase):
                     model_type="generation",
                     lora_paths=[lora_adapter_paths[0], lora_adapter_paths[1]],
                     max_loras_per_batch=len(lora_adapter_paths) + 1,
-                    lora_backend=backend,
                     sleep_on_idle=True,  # Eliminate non-determinism by forcing all requests to be processed in one batch.
                     attention_backend="torch_native",
                 )
@@ -142,7 +140,7 @@ class TestLoRA(CustomTestCase):
                             if rouge_score < rouge_tol:
                                 raise AssertionError(
                                     f"ROUGE-L score {rouge_score} below tolerance {rouge_tol} "
-                                    f"for base '{base_path}', adaptor '{lora_paths}', backend '{backend}', prompt: '{prompts}...'"
+                                    f"for base '{base_path}', adaptor '{lora_paths}', prompt: '{prompts}...'"
                                 )
 
                         print(f"--- Batch {i} Comparison Passed --- ")