[https://nvbugs/6663062][fix] remove Mistral moe tests

tensorrt_llm/_torch/model_config.py

@@ -641,7 +641,9 @@ def load_hf_quant_config(hf_quant_config, moe_backend, checkpoint_dir=None):
                 or hf_quant_config.get("activation_scheme") == "static"):
             quant_config.quant_algo = QuantAlgo.FP8_BLOCK_SCALES
 
-            block_size = hf_quant_config.get("weight_block_size", [128, 128])
+            block_size = hf_quant_config.get("weight_block_size")
+            if block_size is None:
+                block_size = [128, 128]
             assert tuple(block_size) == (
                 128, 128), "FP8_BLOCK_SCALES only supports block_size=(128,128)"
             quant_config.group_size = block_size[0]

tests/integration/defs/llmapi/test_llm_api_pytorch_moe_lora.py

@@ -267,7 +267,7 @@ def test_mixtral_moe_routed_expert_fp8_multi_lora_varying_ranks(moe_lora_mode: s
     """
     cuda_graph_config = CudaGraphConfig(max_batch_size=10) if moe_lora_mode == "cudagraph" else None
 
-    model_dir = f"{llm_models_root()}/Mistral-Small-4-119B-2603"
+    model_dir = f"{llm_models_root()}/Mistral-Small-4-119B-2603/Mistral-Small-4-119B-2603"
 
     ranks = _RANKS
     max_rank = max(ranks)

tests/unittest/_torch/test_hf_quant_config.py

@@ -62,3 +62,20 @@ def test_load_hf_quant_config_parses_nvfp4_with_kv_cache_scheme():
     assert quant_config.group_size == 16
     assert quant_config.kv_cache_quant_algo == QuantAlgo.FP8
     assert set(quant_config.exclude_modules) == {gate_exclude, "lm_head"}
+
+
+def test_load_hf_quant_config_defaults_null_fp8_weight_block_size():
+    quant_config, layer_quant_config = ModelConfig.load_hf_quant_config(
+        {
+            "activation_scheme": "static",
+            "modules_to_not_convert": ["model.vision_tower", "lm_head"],
+            "quant_method": "fp8",
+            "weight_block_size": None,
+        },
+        moe_backend="CUTLASS",
+    )
+
+    assert layer_quant_config is None
+    assert quant_config.quant_algo == QuantAlgo.FP8_BLOCK_SCALES
+    assert quant_config.group_size == 128
+    assert {"model.vision_tower", "lm_head"} <= set(quant_config.exclude_modules)