inspect-ai 0.3.91__py3-none-any.whl → 0.3.92__py3-none-any.whl

inspect_ai/_cli/eval.py

@@ -1164,6 +1164,10 @@ def eval_retry_command(
     elif fail_on_error == 0.0:
         fail_on_error = True
 
+    # resolve retry on error
+    if retry_on_error == 0:
+        retry_on_error = None
+
     # resolve log file
     retry_log_files = [
         log_file_info(filesystem(log_file).info(log_file)) for log_file in log_files

inspect_ai/model/_openai_responses.py

@@ -1,3 +1,4 @@
+import json
 from itertools import chain
 from typing import TypedDict, cast
 
@@ -306,6 +307,14 @@ def _openai_input_items_from_chat_message_assistant(
     """
     (output_message_id, tool_message_ids) = _ids_from_assistant_internal(message)
 
+    # we want to prevent yielding output messages in the case where we have an
+    # 'internal' field (so the message came from the model API as opposed to
+    # being user synthesized) AND there is no output_message_id (indicating that
+    # when reading the message from the server we didn't find output). this could
+    # happen e.g. when a react() agent sets the output.completion in response
+    # to a submit() tool call
+    suppress_output_message = message.internal is not None and output_message_id is None
+
     # if we are not storing messages on the server then blank these out
     if not store:
         output_message_id = None
@@ -341,6 +350,9 @@ def _openai_input_items_from_chat_message_assistant(
                         )
                     )
             case ContentText(text=text, refusal=refusal):
+                if suppress_output_message:
+                    continue
+
                 new_content = (
                     ResponseOutputRefusalParam(type="refusal", refusal=text)
                     if refusal
@@ -415,7 +427,7 @@ def _tool_call_items_from_assistant_message(
                 type="function_call",
                 call_id=call.id,
                 name=_responses_tool_alias(call.function),
-                arguments=call.function,
+                arguments=json.dumps(call.arguments),
             )
 
             # add id if available

{inspect_ai-0.3.91.dist-info → inspect_ai-0.3.92.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inspect_ai
-Version: 0.3.91
+Version: 0.3.92
 Summary: Framework for large language model evaluations
 Author: UK AI Security Institute
 License: MIT License

{inspect_ai-0.3.91.dist-info → inspect_ai-0.3.92.dist-info}/RECORD

@@ -3,7 +3,7 @@ inspect_ai/__main__.py,sha256=oWX4YwDZDg3GS3-IG0yPGoSEOfSzWihELg7QmrUlxjM,67
 inspect_ai/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_ai/_cli/cache.py,sha256=RVGuBYwwk3c45JfyfcSFJ419etSsv3-Z7AyfQE-Hul0,3912
 inspect_ai/_cli/common.py,sha256=WbAgGbLcxABcWoEWiGCWSbkkxYr2jmL9i0cyseWHmRA,4165
-inspect_ai/_cli/eval.py,sha256=A__oE6kYtcZC3aevIDkEoAoaqK0OFccR_Gl8NVMrVrw,38442
+inspect_ai/_cli/eval.py,sha256=g1hNhMG79mtGYq1YAC-FKgdSwpTQK6jOj7yYu2IkOlY,38530
 inspect_ai/_cli/info.py,sha256=QMxaTG9TmzW95EiLrOgkzubvavoR-VHxo3eV7ppmrzI,1789
 inspect_ai/_cli/list.py,sha256=M8mazI8Zuq8Hp99YWKnxQd9UWx1Qi87zfXRzZYAAakk,2459
 inspect_ai/_cli/log.py,sha256=O-w7GqsE7tLojPnEtpWXPSh0Vu2Hbrbxjneyc1BFfpk,5817
@@ -559,7 +559,7 @@ inspect_ai/model/_model_call.py,sha256=VJ8wnl9Y81JaiClBYM8eyt1jVb3n-yc6Dd88ofRiJ
 inspect_ai/model/_model_output.py,sha256=UNodmzkeFE1JXfVtnu0i8q6rwgTj2BClJqcftLSGXdA,7829
 inspect_ai/model/_openai.py,sha256=WlsWWMzP55_pxP_6yp_5eCKo4Ie418OgcgEDKmqjq34,24364
 inspect_ai/model/_openai_computer_use.py,sha256=vbKkYLhqNuX16zuWfg5MaGp9H8URrPcLhKQ1pDsZtPo,5943
-inspect_ai/model/_openai_responses.py,sha256=-mCLQmcR7JVZbTzaHdMBBsGHOyYGAwKQqUbEQ_jnLmI,21912
+inspect_ai/model/_openai_responses.py,sha256=PW17R217NBy0exq4LLfM-eI0GTcG4gm6cxjVG9Uy3d8,22528
 inspect_ai/model/_reasoning.py,sha256=qmR8WT6t_cb7NIsJOQHPyFZh2eLV0HmYxKo2vtvteQ4,929
 inspect_ai/model/_registry.py,sha256=Cr2y32EqLnOqLbSWoXHVK4ivTTzCUhJuACxoTyPt8kY,2032
 inspect_ai/model/_render.py,sha256=rWypNUjgrH4NGp0r-ESAze9gZz7lYNjheEP438vRYZE,922
@@ -724,9 +724,9 @@ inspect_ai/util/_sandbox/docker/internal.py,sha256=c8X8TLrBPOvsfnq5TkMlb_bzTALyc
 inspect_ai/util/_sandbox/docker/prereqs.py,sha256=0j6_OauBBnVlpBleADcZavIAAQZy4WewVjbRn9c0stg,3355
 inspect_ai/util/_sandbox/docker/service.py,sha256=hhHIWH1VDFLwehdGd19aUBD_VKfDO3GCPxpw1HSwVQk,2437
 inspect_ai/util/_sandbox/docker/util.py,sha256=EeInihCNXgUWxaqZ4dNOJd719kXL2_jr63QCoXn68vA,3154
-inspect_ai-0.3.91.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
-inspect_ai-0.3.91.dist-info/METADATA,sha256=h1x-Ak_MDOKNR8pUHwQ7l_3CnHwo6wVceyiKUrFpIbI,5143
-inspect_ai-0.3.91.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
-inspect_ai-0.3.91.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
-inspect_ai-0.3.91.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
-inspect_ai-0.3.91.dist-info/RECORD,,
+inspect_ai-0.3.92.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
+inspect_ai-0.3.92.dist-info/METADATA,sha256=BtQ8ClHkDw2ChlQl_rV7AS4wtisP49cxajyuKIhCmkI,5143
+inspect_ai-0.3.92.dist-info/WHEEL,sha256=SmOxYU7pzNKBqASvQJ7DjX3XGUF92lrGhMb3R6_iiqI,91
+inspect_ai-0.3.92.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
+inspect_ai-0.3.92.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
+inspect_ai-0.3.92.dist-info/RECORD,,