wip

llama_stack/apis/evals/client.py

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

llama_stack/providers/impls/meta_reference/evals/evals.py

@@ -14,10 +14,6 @@ from llama_stack.providers.impls.meta_reference.evals.datas.dataset_registry imp
     get_dataset,
 )
 
-# from llama_stack.providers.impls.meta_reference.evals.tasks.task_registry import (
-#     get_task,
-# )
-
 from .config import MetaReferenceEvalsImplConfig
 
 
@@ -45,7 +41,9 @@ class MetaReferenceEvalsImpl(Evals):
 
         # TODO: replace w/ batch inference & async return eval job
         generation_outputs = []
+        print("generation start")
         for msg in x1[:5]:
+            print("generation for msg: ", msg)
             response = self.inference_api.chat_completion(
                 model=model,
                 messages=[msg],

tests/examples/local-run.yaml

@@ -39,18 +39,18 @@ api_providers:
     config: {}
 routing_table:
   inference:
-  # - provider_type: meta-reference
-  #   config:
-  #     model: Llama3.2-1B-Instruct
-  #     quantization: null
-  #     torch_seed: null
-  #     max_seq_len: 4096
-  #     max_batch_size: 1
-  #   routing_key: Llama3.2-1B-Instruct
-  - provider_type: remote::tgi
+  - provider_type: meta-reference
     config:
-      url: http://127.0.0.1:5009
-    routing_key: Llama3.1-8B-Instruct
+      model: Llama3.2-1B-Instruct
+      quantization: null
+      torch_seed: null
+      max_seq_len: 4096
+      max_batch_size: 1
+    routing_key: Llama3.2-1B-Instruct
+  # - provider_type: remote::tgi
+  #   config:
+  #     url: http://127.0.0.1:5009
+  #   routing_key: Llama3.1-8B-Instruct
   safety:
   - provider_type: meta-reference
     config: