[https://nvbugs/5644187][fix] Llava-Next MMMU bugfix and Phi4 test bugfix (NVIDIA#9034)

Signed-off-by: yechank <[email protected]>

Author: yechank-nvidia

tests/integration/defs/accuracy/references/mmmu.yaml

@@ -4,7 +4,7 @@ Qwen/Qwen2.5-VL-7B-Instruct:
   - accuracy: 51.22
 nvidia/Nano-v2-VLM:
   - accuracy: 43.78
-llava-hf/llava-v1.6-mistral-7b:
+llava-hf/llava-v1.6-mistral-7b-hf:
   - accuracy: 35.33
 Efficient-Large-Model/NVILA-8B:
   - accuracy: 47.77

tests/integration/defs/accuracy/test_llm_api_pytorch_multimodal.py

@@ -80,8 +80,8 @@ def test_auto_dtype(self):
 
 
 class TestLlava_V1_6_Mistral_7B(LlmapiAccuracyTestHarness):
-    MODEL_NAME = "llava-hf/llava-v1.6-mistral-7b"
-    MODEL_PATH = f"{llm_models_root()}/llava-v1.6-mistral-7b"
+    MODEL_NAME = "llava-hf/llava-v1.6-mistral-7b-hf"
+    MODEL_PATH = f"{llm_models_root()}/llava-v1.6-mistral-7b-hf"
     MAX_NUM_TOKENS = 16384
 
     # NOTE: MMMU adds <|endoftext|> to the stop token.

tests/integration/defs/test_e2e.py

@@ -2658,7 +2658,7 @@ def test_ptp_quickstart_multimodal_kv_cache_reuse(llm_root, llm_venv,
                 ],
             ] * num_same_requests,
         },
-        "phi4-multimodal-instruct-fp8": {
+        "phi4-multimodal-instruct": {
             "image": [
                 [
                     "image", "depicts", "natural", "environment", "ocean",