[ci] Change trtllm correctness testing to llama3-8b

deepjavalibrary · Jul 31, 2024 · bbbbe49 · bbbbe49
1 parent 5154298
commit bbbbe49
Show file tree

Hide file tree

Showing 3 changed files with 7 additions and 7 deletions.
diff --git a/tests/integration/llm/client.py b/tests/integration/llm/client.py
@@ -739,7 +739,7 @@ def get_model_name():
             "return_full_text": True
         }
     },
-    "trtllm-llama3-1-8b": {
+    "trtllm-llama3-8b": {
         "batch_size": [213],
         "seq_length": [1],
         "num_run": 66,

diff --git a/tests/integration/llm/prepare.py b/tests/integration/llm/prepare.py
@@ -939,10 +939,10 @@
         "option.max_rolling_batch_size": 41,
         "option.model_loading_timeout": 1800
     },
-    "trtllm-llama3-1-8b": {
+    "trtllm-llama3-8b": {
         "engine": "Python",
         "option.task": "text-generation",
-        "option.model_id": "s3://djl-llm/llama-3.1-8b-hf/",
+        "option.model_id": "s3://djl-llm/llama-3-8b-hf/",
         "option.rolling_batch": "trtllm",
         "option.tensor_parallel_degree": 4,
         "option.max_rolling_batch_size": 213

diff --git a/tests/integration/tests.py b/tests/integration/tests.py
@@ -849,11 +849,11 @@ def test_codestral_22b(self):
             r.launch("CUDA_VISIBLE_DEVICES=0,1,2,3")
             client.run("correctness trtllm-codestral-22b".split())
 
-    def test_llama3_1_8b(self):
-        with Runner('tensorrt-llm', 'llama3-1-8b') as r:
-            prepare.build_correctness_model("trtllm-llama3-1-8b")
+    def test_llama3_8b(self):
+        with Runner('tensorrt-llm', 'llama3-8b') as r:
+            prepare.build_correctness_model("trtllm-llama3-8b")
             r.launch("CUDA_VISIBLE_DEVICES=0,1,2,3")
-            client.run("correctness trtllm-llama3-1-8b".split())
+            client.run("correctness trtllm-llama3-8b".split())
 
 
 @pytest.mark.correctness