rename tasks

llama_stack/apis/evals/client.py

@@ -44,7 +44,7 @@ async def run_main(host: str, port: int):
 
     # CustomDataset
     response = await client.run_evals(
-        "Llama3.2-1B-Instruct",
+        "Llama3.1-8B-Instruct",
         "mmlu-simple-eval-en",
         "mmlu",
     )

llama_stack/providers/impls/third_party/evals/eleuther/eleuther.py

@@ -90,7 +90,6 @@ class EleutherEvalsWrapper(LM):
     def generate_until(self, requests, disable_tqdm: bool = False) -> List[str]:
         res = []
         for req in requests:
-            print("generation for msg: ", req.args[0])
             response = self.inference_api.chat_completion(
                 model=self.model,
                 messages=[
@@ -144,7 +143,7 @@ class EleutherEvalsAdapter(Evals):
         output = evaluate(
             eluther_wrapper,
             task_dict,
-            limit=1,
+            limit=10,
         )
 
         formatted_output = lm_eval.utils.make_table(output)

tests/examples/local-run.yaml

@@ -21,7 +21,7 @@ providers:
   - provider_id: meta-reference
     provider_type: meta-reference
     config:
-      model: Llama3.2-1B-Instruct
+      model: Llama3.1-8B-Instruct
       quantization: null
       torch_seed: null
       max_seq_len: 4096
@@ -54,8 +54,8 @@ providers:
     provider_type: meta-reference
     config: {}
 models:
-- identifier: Llama3.2-1B-Instruct
-  llama_model: Llama3.2-1B-Instruct
+- identifier: Llama3.1-8B-Instruct
+  llama_model: Llama3.1-8B-Instruct
   provider_id: meta-reference
 shields:
 - identifier: llama_guard