make 3.18b use SanctumAI weights rather than attempt local irpa

nod-ai · Jan 30, 2025 · 6de4446 · 6de4446
1 parent 99b2c5a
commit 6de4446
Showing 1 changed file with 3 additions and 5 deletions.
diff --git a/app_tests/integration_tests/llm/shortfin/conftest.py b/app_tests/integration_tests/llm/shortfin/conftest.py
@@ -26,11 +26,9 @@
         device_settings=device_settings.CPU,
     ),
     "llama3.1_8b": ModelConfig(
-        source=ModelSource.LOCAL,
-        local_path=Path(
-            "/data/llama3.1/weights/8b/fp16/llama3.1_8b_fp16_instruct.irpa"
-        ),
-        model_file="llama3.1_8b_fp16_instruct.irpa",
+        source=ModelSource.HUGGINGFACE,
+        repo_id="SanctumAI/Meta-Llama-3.1-8B-Instruct-GGUF",
+        model_file="meta-llama-3.1-8b-instruct.f16.gguf",
         tokenizer_id="NousResearch/Meta-Llama-3.1-8B",
         batch_sizes=(1, 4),
         device_settings=device_settings.CPU,