fixes

Author: benjibc

examples/tinker_math_rl/debug_dataset.py

@@ -27,7 +27,7 @@ def get_gsm8k_input_rows(limit: int = 10) -> List[EvaluationRow]:
 
 
 @evaluation_test(
-    input_rows=[gsm8k_input_rows],
+    input_rows=gsm8k_input_rows,
     completion_params=[
         {
             "max_tokens": 512,
@@ -48,7 +48,9 @@ def test_gsm8k_tinker(row: EvaluationRow) -> EvaluationRow:
     else:
         model_response = assistant_msgs[-1].content
         # The content might be a list of content parts, handle that
-        if not isinstance(model_response, str):
+        if model_response is None:
+            model_response = ""
+        elif not isinstance(model_response, str):
             # Simple join for now if it's a list
             model_response = "".join([p.text for p in model_response if hasattr(p, "text")])
 

examples/tinker_math_rl/test_gsm8k_eval.py

@@ -1,11 +1,13 @@
 import asyncio
+import copy
 import logging
 import os
 import sys
 from functools import partial
 from typing import Literal, Any, Optional
 
 import chz
+import datetime
 
 # Add tinker-cookbook to path if not installed
 # Assuming the directory structure:
@@ -153,7 +155,7 @@ async def cli_main(cli_config: CLIConfig):
     # Need to wrap in a factory as expected by Config.evaluator_builders
     def create_eval_protocol_evaluator():
         return EvalProtocolEvaluator(
-            rows=eval_rows,
+            rows=copy.deepcopy(eval_rows),
             eval_func=test_gsm8k_tinker,
             rollout_processor_cls=TinkerRolloutProcessor,
             model_name=cli_config.model_name,
@@ -203,7 +205,5 @@ def create_eval_protocol_evaluator():
 
 
 if __name__ == "__main__":
-    from datetime import datetime
-
     cli_config = chz.entrypoint(CLIConfig)
     asyncio.run(cli_main(cli_config))