inspect-ai 0.3.68__py3-none-any.whl → 0.3.69__py3-none-any.whl

inspect_ai/model/_providers/mistral.py

@@ -111,16 +111,12 @@ class MistralAPI(ModelAPI):
         if base_url:
             model_args["server_url"] = base_url
 
-        # create client
-        self.client = Mistral(
-            api_key=self.api_key,
-            timeout_ms=(config.timeout if config.timeout else DEFAULT_TIMEOUT) * 1000,
-            **model_args,
-        )
+        self.model_args = model_args
 
     @override
     async def close(self) -> None:
-        await self.client.sdk_configuration.async_client.aclose()
+        # client is created and destroyed in generate
+        pass
 
     async def generate(
         self,
@@ -149,7 +145,13 @@ class MistralAPI(ModelAPI):
 
         # send request
         try:
-            response = await self.client.chat.complete_async(**request)
+            with Mistral(
+                api_key=self.api_key,
+                timeout_ms=(config.timeout if config.timeout else DEFAULT_TIMEOUT)
+                * 1000,
+                **self.model_args,
+            ) as client:
+                response = await client.chat.complete_async(**request)
         except SDKError as ex:
             if ex.status_code == 400:
                 return self.handle_bad_request(ex), mistral_model_call(request, None)

inspect_ai/model/_providers/providers.py

@@ -93,8 +93,8 @@ def vertex() -> type[ModelAPI]:
 @modelapi(name="google")
 def google() -> type[ModelAPI]:
     FEATURE = "Google API"
-    PACKAGE = "google-generativeai"
-    MIN_VERSION = "0.8.4"
+    PACKAGE = "google-genai"
+    MIN_VERSION = "1.2.0"
 
     # workaround log spam
     # https://github.com/ray-project/ray/issues/24917
@@ -102,7 +102,7 @@ def google() -> type[ModelAPI]:
 
     # verify we have the package
     try:
-        import google.generativeai  # type: ignore  # noqa: F401
+        import google.genai  # type: ignore  # noqa: F401
     except ImportError:
         raise pip_dependency_error(FEATURE, [PACKAGE])
 
@@ -110,9 +110,9 @@ def google() -> type[ModelAPI]:
     verify_required_version(FEATURE, PACKAGE, MIN_VERSION)
 
     # in the clear
-    from .google import GoogleAPI
+    from .google import GoogleGenAIAPI
 
-    return GoogleAPI
+    return GoogleGenAIAPI
 
 
 @modelapi(name="hf")

inspect_ai/solver/_plan.py

@@ -2,6 +2,7 @@ import inspect
 from logging import getLogger
 from typing import Any, Awaitable, Callable, TypeVar, cast
 
+from inspect_ai._util.interrupt import check_sample_interrupt
 from inspect_ai._util.registry import (
     RegistryInfo,
     is_registry_object,
@@ -115,6 +116,7 @@ class Plan(Solver):
                 with solver_transcript(self.finish, state) as st:
                     state = await self.finish(state, generate)
                     st.complete(state)
+                check_sample_interrupt()
 
             # mark completed
             state.completed = True
@@ -124,6 +126,7 @@ class Plan(Solver):
             if self.cleanup:
                 try:
                     await self.cleanup(state)
+                    check_sample_interrupt()
                 except Exception as ex:
                     logger.warning(f"Exception occurred during plan cleanup: {ex}")
 

inspect_ai/solver/_solver.py

@@ -15,6 +15,7 @@ from typing import (
 from typing_extensions import Unpack
 
 from inspect_ai._util._async import is_callable_coroutine
+from inspect_ai._util.interrupt import check_sample_interrupt
 from inspect_ai._util.registry import (
     RegistryInfo,
     registry_add,
@@ -200,6 +201,7 @@ def solver(
                     state: TaskState, generate: Generate
                 ) -> TaskState:
                     state = await original_call(state, generate)
+                    check_sample_interrupt()
                     set_sample_state(state)
                     return state
 
@@ -215,6 +217,7 @@ def solver(
                     state: TaskState, generate: Generate
                 ) -> TaskState:
                     state = await solver(state, generate)
+                    check_sample_interrupt()
                     set_sample_state(state)
                     return state
 

inspect_ai/solver/_task_state.py

@@ -8,6 +8,7 @@ from typing import Any, Iterable, SupportsIndex, Type, Union, cast, overload
 
 from pydantic_core import to_jsonable_python
 
+from inspect_ai._util.interrupt import check_sample_interrupt
 from inspect_ai.dataset._dataset import MT, Sample, metadata_as
 from inspect_ai.model import (
     ChatMessage,
@@ -333,7 +334,7 @@ class TaskState:
     def completed(self) -> bool:
         """Is the task completed.
 
-        Additionally, checks message and token limits and raises if they are exceeded.
+        Additionally, checks message and token limits and raises if they are exceeded, and also checks for an operator interrupt of the sample.
         """
         from inspect_ai.log._samples import set_active_sample_total_messages
 
@@ -356,6 +357,7 @@ class TaskState:
                 "token", value=self.token_usage, limit=self.token_limit, state=self
             )
         else:
+            check_sample_interrupt()
             return self._completed
 
     @completed.setter

inspect_ai/util/_sandbox/docker/cleanup.py

@@ -56,17 +56,22 @@ async def project_cleanup_shutdown(cleanup: bool) -> None:
                     title_style="bold",
                     title_justify="left",
                 )
+                table.add_column("Sample ID")
+                table.add_column("Epoch")
                 table.add_column("Container(s)", no_wrap=True)
-                table.add_column("Cleanup")
                 for project in shutdown_projects:
                     containers = await compose_ps(project, all=True)
                     table.add_row(
+                        str(project.sample_id) if project.sample_id is not None else "",
+                        str(project.epoch if project.epoch is not None else ""),
                         "\n".join(container["Name"] for container in containers),
-                        f"[blue]inspect sandbox cleanup docker {project.name}[/blue]",
                     )
                 print(table)
                 print(
-                    "\nCleanup all environments with: [blue]inspect sandbox cleanup docker[/blue]\n"
+                    "\n"
+                    "Cleanup all containers  : [blue]inspect sandbox cleanup docker[/blue]\n"
+                    "Cleanup single container: [blue]inspect sandbox cleanup docker <container-id>[/blue]",
+                    "\n",
                 )
 
         # remove auto-compose files

inspect_ai/util/_sandbox/docker/compose.py

@@ -28,7 +28,7 @@ COMPOSE_WAIT = 120
 
 async def compose_up(
     project: ComposeProject, services: dict[str, ComposeService]
-) -> None:
+) -> ExecResult[str]:
     # compute the maximum amount of time we will
     up_command = ["up", "--detach", "--wait"]
 
@@ -49,7 +49,8 @@ async def compose_up(
     # passing the --wait flag (see https://github.com/docker/compose/issues/10596).
     # In practice, we will catch any errors when calling compose_check_running()
     # immediately after we call compose_up().
-    await compose_command(up_command, project=project, timeout=timeout)
+    result = await compose_command(up_command, project=project, timeout=timeout)
+    return result
 
 
 async def compose_down(project: ComposeProject, quiet: bool = True) -> None:
@@ -121,14 +122,9 @@ async def compose_check_running(
             unhealthy_services = services
             for successful_service in successful_services:
                 unhealthy_services.remove(successful_service["Service"])
-
-            msg = (
-                "One or more docker containers failed to start from "
-                f"{project.config}: {','.join(unhealthy_services)}"
-            )
-            raise RuntimeError(msg)
+            return []
     else:
-        raise RuntimeError("No services started")
+        return []
 
     return [service["Service"] for service in running_services]
 

inspect_ai/util/_sandbox/docker/docker.py

@@ -139,8 +139,15 @@ class DockerSandboxEnvironment(SandboxEnvironment):
                     env[key] = str(value)
 
         # create project
+        from inspect_ai.log._samples import sample_active
+
+        sample = sample_active()
         project = await ComposeProject.create(
-            name=task_project_name(task_name), config=config, env=env
+            name=task_project_name(task_name),
+            config=config,
+            sample_id=sample.id if sample is not None else None,
+            epoch=sample.epoch if sample is not None else None,
+            env=env,
         )
 
         try:
@@ -148,13 +155,18 @@ class DockerSandboxEnvironment(SandboxEnvironment):
             services = await compose_services(project)
 
             # start the services
-            await compose_up(project, services)
+            result = await compose_up(project, services)
 
             # check to ensure that the services are running
             running_services = await compose_check_running(
                 list(services.keys()), project=project
             )
 
+            if not running_services:
+                raise RuntimeError(
+                    f"No services started.\nCompose up stderr: {result.stderr}"
+                )
+
             # note that the project is running
             project_startup(project)
 

inspect_ai/util/_sandbox/docker/util.py

@@ -21,6 +21,8 @@ logger = getLogger(__name__)
 class ComposeProject:
     name: str
     config: str | None
+    sample_id: int | str | None
+    epoch: int | None
     env: dict[str, str] | None
 
     @classmethod
@@ -28,6 +30,9 @@ class ComposeProject:
         cls,
         name: str,
         config: SandboxEnvironmentConfigType | None,
+        *,
+        sample_id: int | str | None = None,
+        epoch: int | None = None,
         env: dict[str, str] = {},
     ) -> "ComposeProject":
         # resolve config to full path if we have one
@@ -58,16 +63,20 @@ class ComposeProject:
         ensure_auto_compose_file(config)
 
         # return project
-        return ComposeProject(name, config, env)
+        return ComposeProject(name, config, sample_id=sample_id, epoch=epoch, env=env)
 
     def __init__(
         self,
         name: str,
         config: str | None,
+        sample_id: int | str | None,
+        epoch: int | None,
         env: dict[str, str],
     ) -> None:
         self.name = name
         self.config = config
+        self.sample_id = sample_id
+        self.epoch = epoch
         self.env = env
 
     def __eq__(self, other: object) -> bool:

inspect_ai/util/_sandbox/self_check.py

@@ -445,7 +445,8 @@ async def test_exec_stdout_is_limited(sandbox_env: SandboxEnvironment) -> None:
     assert "limit of 10 MiB was exceeded" in str(e_info.value)
     truncated_output = e_info.value.truncated_output
     # `yes` outputs 'y\n' (ASCII) so the size equals the string length.
-    assert truncated_output and len(truncated_output) == 10 * 1024**2
+    # some shells additionally output 'canceled\n' so we add fudge factor for that
+    assert truncated_output and (len(truncated_output) - 10 * 1024**2) < 10
 
 
 async def test_exec_stderr_is_limited(sandbox_env: SandboxEnvironment) -> None:

inspect_ai/util/_subprocess.py

@@ -199,7 +199,10 @@ async def subprocess(
                 else:
                     result = await asyncio.wait_for(anext(rc), timeout=timeout)
                     return cast(Union[ExecResult[str], ExecResult[bytes]], result)
-            except asyncio.exceptions.TimeoutError:
+            # wait_for raises asyncio.TimeoutError under Python 3.10, but TimeoutError
+            # under Python > 3.11! asynio.timeout (introduced in Python 3.11) always
+            # raises the standard TimeoutError
+            except (TimeoutError, asyncio.exceptions.TimeoutError):
                 # terminate timed out process -- try for graceful termination
                 # then be more forceful if requied
                 try:

{inspect_ai-0.3.68.dist-info → inspect_ai-0.3.69.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: inspect_ai
-Version: 0.3.68
+Version: 0.3.69
 Summary: Framework for large language model evaluations
 Author: UK AI Security Institute
 License: MIT License
@@ -53,7 +53,7 @@ Requires-Dist: anthropic; extra == "dev"
 Requires-Dist: aioboto3; extra == "dev"
 Requires-Dist: azure-ai-inference; extra == "dev"
 Requires-Dist: google-cloud-aiplatform; extra == "dev"
-Requires-Dist: google-generativeai; extra == "dev"
+Requires-Dist: google-genai; extra == "dev"
 Requires-Dist: goodfire; extra == "dev"
 Requires-Dist: griffe; extra == "dev"
 Requires-Dist: groq; extra == "dev"
@@ -71,7 +71,7 @@ Requires-Dist: pytest-asyncio; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
 Requires-Dist: pytest-dotenv; extra == "dev"
 Requires-Dist: pytest-xdist; extra == "dev"
-Requires-Dist: ruff==0.9.5; extra == "dev"
+Requires-Dist: ruff==0.9.6; extra == "dev"
 Requires-Dist: textual-dev>=0.86.2; extra == "dev"
 Requires-Dist: types-Markdown; extra == "dev"
 Requires-Dist: types-PyYAML; extra == "dev"

{inspect_ai-0.3.68.dist-info → inspect_ai-0.3.69.dist-info}/RECORD

@@ -25,10 +25,10 @@ inspect_ai/_display/core/results.py,sha256=aFLmG1Ij0fxYk2848QgQlesfMeRdHVEg_W9es
 inspect_ai/_display/core/rich.py,sha256=GPzc-0PWZVOPWxnjfQmNSK66uZXc3x8joz4ethgv_4M,2729
 inspect_ai/_display/core/textual.py,sha256=kzMTt8ijrodwhDB5V50pP2IBhnUCusVbP86TytU_rA8,870
 inspect_ai/_display/plain/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-inspect_ai/_display/plain/display.py,sha256=fGITQj6w4-2A4zNMKFW32wdyVB5JpK-LsXMbFckzXTA,6743
+inspect_ai/_display/plain/display.py,sha256=WF16dmlKNQojdSWlu1RY6wfw2D0o5iA0tTaZ8vYKT8c,6700
 inspect_ai/_display/rich/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 inspect_ai/_display/rich/display.py,sha256=aJiuaEL8aEdOvdkKWobhlDuMr2X2j1M_SiAezPd3k-A,10920
-inspect_ai/_display/textual/app.py,sha256=PyDkwzyEHLjObvwB3PtPlP0CSU9L7frYb1eoQ5MhnR0,15991
+inspect_ai/_display/textual/app.py,sha256=wZ5nV_d1YA6pMVKobhhm4FLvDoVrfmgnyI9wDAg7Tpk,16030
 inspect_ai/_display/textual/app.tcss,sha256=ymx47Y2fAbce4vPOrpeij1p9qwSipT7Dd6fSxCTknT8,249
 inspect_ai/_display/textual/display.py,sha256=zTG8k67iT-xq1H8CVwtqZdvuo_O2hsgh4q4y8EKRkf8,2027
 inspect_ai/_display/textual/theme.py,sha256=SVf2UhQVuEOzzNx1Pl2bqLrofw13ykQKAVnevUpJBYs,588
@@ -36,7 +36,7 @@ inspect_ai/_display/textual/widgets/clock.py,sha256=pxXQOtadf6qdNMLQh_D3bx3SIPoB
 inspect_ai/_display/textual/widgets/console.py,sha256=lp5lbT9erPjxE1NWzvuJ5Bj8mN2ZZSBTgKQWHinMKgA,1590
 inspect_ai/_display/textual/widgets/footer.py,sha256=_l9BgJWXX_vXO2qrSIkeUPZzniYHa0qXaiSiy6479Tg,1106
 inspect_ai/_display/textual/widgets/port_mappings.py,sha256=mxQJGYeZh1aXNmW5z-Ukss7Zwul5qfH5CqtvTLl0BAU,2974
-inspect_ai/_display/textual/widgets/samples.py,sha256=rBaBP__MbKEX18B00nQCiLrWO9MmtSn0arh-zL90c8I,18206
+inspect_ai/_display/textual/widgets/samples.py,sha256=CfNV3YHPMSeidClWw8HCX0rnePYOypSyzzJ5b4oxzck,19687
 inspect_ai/_display/textual/widgets/sandbox.py,sha256=_Ivpba2oIxs_Lw2Pic7BRoHeVu9ibaGPFhiyfzbcc9E,1146
 inspect_ai/_display/textual/widgets/task_detail.py,sha256=aaI626vbvieGbZtOKWmnaNWLivFMjnlDt4oqeTmKEZg,8389
 inspect_ai/_display/textual/widgets/tasks.py,sha256=2cqYJ3PGn4F-riSdB3Fp1H5YCWQ-PVh7L8OrLCPSiso,11283
@@ -89,10 +89,11 @@ inspect_ai/_util/hooks.py,sha256=8QnHCQQY_2XMYPkiPvixUgFY0E_niZvQhQDMI-eCdhM,435
 inspect_ai/_util/html.py,sha256=X62FY8gpEJ2ZQoDu2y8aQAbiBUIHKsd7DA9rWCIleo8,168
 inspect_ai/_util/http.py,sha256=c4yvH48ZkETZ7sNDuNzBR0NUS4r-6WzCaolW9my13ns,3628
 inspect_ai/_util/images.py,sha256=W7QJHyzuXhfy3VsLhKTzddSo1g69O9RxnTyhat48Wyo,1312
+inspect_ai/_util/interrupt.py,sha256=atMrIdpQsk94HlAM5xnTwPgSGIo1CPvjl9z1hamgKFA,221
 inspect_ai/_util/json.py,sha256=qxr8KPQ3NLPDbSSuDQjRQqZcwCOOWpPpuYc36fXsj-w,3786
 inspect_ai/_util/kvstore.py,sha256=z2IXLWP4QqqGqsq5_MbYjBQPcEJqfWK4IyZXgV-kppA,2398
 inspect_ai/_util/list.py,sha256=6_5r5jI5RKK34kCmIqqVQ5hYG-G8v0F5H7L-DmQQ2E4,279
-inspect_ai/_util/logger.py,sha256=wLxB8fsY0_7dkpRYk7tkpK--vNGNd03dc6o2iPJkPyc,7170
+inspect_ai/_util/logger.py,sha256=Tz1QkqCFKcfhsXaAFNS7znziTOq0BPDP4NxuzYcDaEw,7304
 inspect_ai/_util/notebook.py,sha256=Mgz3J4uBh-MqVBRmpiJqDHRpn2hd7HIOBeJBwLG-bbk,2998
 inspect_ai/_util/notgiven.py,sha256=zkn6AYflKLf8YlnwTAMxPLQ-4LyIVmKpGcNcXf-Ssng,457
 inspect_ai/_util/package.py,sha256=2ntItRYaYBaVWI5eDaB4FdpI1IUBiBWNRxq7FChvk1I,2729
@@ -105,7 +106,7 @@ inspect_ai/_util/retry.py,sha256=fAupOVgGJ0ImnmtXmCHBKRa3AQC7jDA-Zf_zilSCRl0,189
 inspect_ai/_util/rich.py,sha256=sNWEsGlGmkkZZLo4AcEv-_yJI1bI0HcpZVt7wNJFsXg,692
 inspect_ai/_util/samples.py,sha256=uobAN2i1U-6YBxCBvaW6z1-xFufQIuFXHnnnK-oTDKc,507
 inspect_ai/_util/terminal.py,sha256=I4NDy7Ln5YSCzxbd0O9OPslEHQMBVKZfqJl3TOCegTg,4166
-inspect_ai/_util/text.py,sha256=4A4-ehftR7QXsOiN4VCBuy8IkgUIkoTIKfA_HZ_e9ac,3909
+inspect_ai/_util/text.py,sha256=cB67HEVxt7bP1Jo4YiEY4z_D-OodTnnI4fidvKZMujE,9673
 inspect_ai/_util/thread.py,sha256=yao8tzMkX4860ulFWTb9x4BT7aHghibxRmfIxWBnZIA,114
 inspect_ai/_util/throttle.py,sha256=JczSG_y0v60m4gQCt28uw_WPjJTbHuq8gWcxY3-vFsc,855
 inspect_ai/_util/timeouts.py,sha256=-iC1LjpNqB6Hx-i36MfSrLy819RVhKNo4KXjZDuskZQ,5193
@@ -139,7 +140,7 @@ inspect_ai/_view/www/.vscode/settings.json,sha256=g5hrVnMaYxM06JpiJD2EuE2xjcbF6x
 inspect_ai/_view/www/dist/index.html,sha256=gpdu6SR-SOH9EWx15cCWHzujMZujnZR5tRlEfROJg2A,997
 inspect_ai/_view/www/dist/assets/favicon.svg,sha256=b9AHYZaO2zBzeKH6G4PwXZMGGW_UxY0omKHam-c9MAs,1508
 inspect_ai/_view/www/dist/assets/index.css,sha256=Igio8rHcaIaA6wnwH3bQhr10yXHD_K9idKsALnb-b7A,893895
-inspect_ai/_view/www/dist/assets/index.js,sha256=74oJ-IiBQ1HnwzGQIlUweY18hYZBbNRhGVJsuyb3_Rc,2631819
+inspect_ai/_view/www/dist/assets/index.js,sha256=asoYOf8KgDCYvayM3hkqMbFVXwEt41RvqbbMPK_maog,2631827
 inspect_ai/_view/www/src/App.tsx,sha256=EvpnQfpWugx4PJivUN8GSUo8RXI4sHAFiIfacVXZ1Z0,28854
 inspect_ai/_view/www/src/AppErrorBoundary.tsx,sha256=RyhZWbIMZj1QeUOUUXh9hUFvq6LoDEoHuTY0giswmL0,1169
 inspect_ai/_view/www/src/constants.ts,sha256=UIxGbDscs61CcOQLQiW6MsZAU1uupSYNVLGxx2pp14A,1169
@@ -259,7 +260,7 @@ inspect_ai/_view/www/src/samples/descriptor/score/BooleanScoreDescriptor.tsx,sha
 inspect_ai/_view/www/src/samples/descriptor/score/CategoricalScoreDescriptor.tsx,sha256=17we32TGRoqayxI4yw99MqlaJfLcJgtZXCtoHBznGOg,460
 inspect_ai/_view/www/src/samples/descriptor/score/NumericScoreDescriptor.tsx,sha256=SiiqqaodZPT1aMW1xBgFGSPQ6Z7OCbtx2n-9sewW7cM,836
 inspect_ai/_view/www/src/samples/descriptor/score/ObjectScoreDescriptor.module.css,sha256=lWNDBHma-aaeeKRRig5GUIYag95UISwcN1QuF46-zR8,202
-inspect_ai/_view/www/src/samples/descriptor/score/ObjectScoreDescriptor.tsx,sha256=EDcEUMYLNgL6Jev2w0CvCpfXeVODEY2ntjhmY_A6uVI,2016
+inspect_ai/_view/www/src/samples/descriptor/score/ObjectScoreDescriptor.tsx,sha256=mq-MFIu4AwEe7ZoQZuOCUUfaguAsgHrsVFG-qtFmfos,2024
 inspect_ai/_view/www/src/samples/descriptor/score/OtherScoreDescriptor.tsx,sha256=argRDR7qkZ--Sb6UnuMc1mYBbEY5CXVX0q1vfiTVYA8,511
 inspect_ai/_view/www/src/samples/descriptor/score/PassFailScoreDescriptor.module.css,sha256=8TVK-2jTTzFAMHxLxDUc3FKm8F1Wkq689L_bzGKCVKI,530
 inspect_ai/_view/www/src/samples/descriptor/score/PassFailScoreDescriptor.tsx,sha256=oTVPJGvuyuYVeyCXUWxV2rtp5Q2sapfRSGzazZDPcog,2058
@@ -433,7 +434,7 @@ inspect_ai/log/_file.py,sha256=o5DNFq0tRV8l7INzIEG76kP9AQM8Ta5nCUgroaTTI6c,17996
 inspect_ai/log/_log.py,sha256=Hr5XlxH7AGXFTj9q-5SE5gLE_WpElXy-fFQq_FV_8QU,21670
 inspect_ai/log/_message.py,sha256=QofM_JZF_x3k_5ta1uQzoN_VnMoUhXFnqWurIn9FXOY,1999
 inspect_ai/log/_retry.py,sha256=e7a2hjl3Ncl8b8sU7CsDpvK8DV0b1uSRLeokRX1mt34,2109
-inspect_ai/log/_samples.py,sha256=uuc9-XJ1_Ly3E1YiBaRn0-pqQ4mCN4fIsVX8f6yvg38,4244
+inspect_ai/log/_samples.py,sha256=0mnXsXGKqfUQgg7y-UgBSujpxYFzbrT-F6S_su8kgEs,4110
 inspect_ai/log/_transcript.py,sha256=Pql8u17vVy0PIZPICeAVjb2WvrNpzWnC-6Y3xyg25pE,11267
 inspect_ai/log/_recorders/__init__.py,sha256=-ECELTfjeWwDuEIBSXO2oxEtJ6Dn0ZQYUxrEy0klN34,350
 inspect_ai/log/_recorders/create.py,sha256=WB-fms0dBDHlTtTa_a_r0fFc6UPRvQZKZT7d_Inp-EU,1103
@@ -448,7 +449,7 @@ inspect_ai/model/_chat_message.py,sha256=Y0X1b1uaZhw0brSGfXVU7lADrXh2nR6iSWgBR_m
 inspect_ai/model/_conversation.py,sha256=cd5ru6lD2xsfkdB9lfWYPbuvzdqjv9geOVFl2HXSad8,2163
 inspect_ai/model/_generate_config.py,sha256=jKsb5qpiV4o2Wdkaez6EzSHhnUc3_j4dTZIOzV9P6Sc,8976
 inspect_ai/model/_image.py,sha256=kpO2Bn_-c-dK80HuPOPH1eSNmcoc39kofwf4yTTiTFE,477
-inspect_ai/model/_model.py,sha256=7ByKcGdmrmpI3BZlNXXLVN-xjyGbaKObHDRS7ZJMBAw,42388
+inspect_ai/model/_model.py,sha256=EgEKg13UhNAN4u72ZNl16q1yIWwq7zhgSHH-w38bti0,42488
 inspect_ai/model/_model_call.py,sha256=r6ObHZwm7jy1qX2qxvAP6iPV3BhdGThL-VH-QEawQhA,2017
 inspect_ai/model/_model_output.py,sha256=QM2K4zVqUumfhAFzDkvxb60qT4DbCvPjqva0WfJ4gWU,7321
 inspect_ai/model/_openai.py,sha256=Z17GAEU-Zraqy-wRcY3KGC34gN1YvRoy0hdYZpt-BjI,13948
@@ -460,18 +461,18 @@ inspect_ai/model/_providers/azureai.py,sha256=L0aZrx-mcRLZWXmNQG2gEmGYVQ7QtLF-o7
 inspect_ai/model/_providers/bedrock.py,sha256=bNQq5Z9w9fLq3SxOKFDuopuLWi0QTDcSE9ppzX51eX0,23482
 inspect_ai/model/_providers/cloudflare.py,sha256=OOpU4EE97s7M6u_qk7uuCILSLU84DoCVrUNsa-qPbmA,4228
 inspect_ai/model/_providers/goodfire.py,sha256=L5RHmpXcp7Se30hJUjK0jES8ckGhxOKDM5xYfOjBcIU,8811
-inspect_ai/model/_providers/google.py,sha256=CVOWm5Q3n8AAIhCz-5sKJ1o-b1v3Vor5UvC8oRYGiU8,23750
+inspect_ai/model/_providers/google.py,sha256=V8N0MbJrbOx6Boz1Vq5h_6TLmRdMjJVQoSS_4V0PiBY,26640
 inspect_ai/model/_providers/grok.py,sha256=dS88ueXiD-kHAFr0jCoTpTGLGa2VsUlB_TFP8L_2lBM,995
 inspect_ai/model/_providers/groq.py,sha256=aVaWrGn6oqWopNuh8WfvbGux3P0tcRQg_uq8DdTJ2Ng,10042
 inspect_ai/model/_providers/hf.py,sha256=KIHcfiRQ8XQEQBi7UzCIulZh1Ku5sCjWaa1EznGPHzU,17605
 inspect_ai/model/_providers/llama_cpp_python.py,sha256=i2I56Damgb8VDhMuPxPca24fVhlajDHzxCTYFnT41uI,702
-inspect_ai/model/_providers/mistral.py,sha256=afhEAfsR4QeBigrINU3Dgf5EZp5zNvkk4Ub-Z1SigcU,15329
+inspect_ai/model/_providers/mistral.py,sha256=XDXSB9JMff2u3wuJ_T0htM3wvJ2MyVJl88Px2bGCvaw,15385
 inspect_ai/model/_providers/mockllm.py,sha256=gL9f-f5TOdE4a0GVENr3cOIIp2kv8zVXWPZ608rouGk,2440
 inspect_ai/model/_providers/ollama.py,sha256=mBPSxaEkiH_RnlHKqOyFBlXObQhc2dfjL-rCKrea5u8,675
 inspect_ai/model/_providers/openai.py,sha256=kD3ypqlPnqA1YdF3zQWzpHOHkF_8OTGMmS86ZqarCjI,11731
 inspect_ai/model/_providers/openai_o1.py,sha256=kiCzFYPBzUR-4U_AcQCoFpYRg-RUiWL6m2_XhK0JhW0,12262
 inspect_ai/model/_providers/openrouter.py,sha256=5G8qS8xA7Gy4IGodEJd04xwjsN-O_as4oeU8DTsKB5s,2932
-inspect_ai/model/_providers/providers.py,sha256=PlzYzv-muCSkmb8p04R0cNHS6DyUWPtB4c0hxW6sWUs,6407
+inspect_ai/model/_providers/providers.py,sha256=ALa8vNhjaVF-B8SpOB4PwQ_Yk4Gqlk6PKCGTRCmMHn0,6403
 inspect_ai/model/_providers/together.py,sha256=0KpFLKbnP_a_AH7VN18eLtlm7kGvJkBIura9drdc7BU,9483
 inspect_ai/model/_providers/vertex.py,sha256=-4ArhaW_H_KGj8P0HdmDtyKRwXdR3-UuZurI7lzO0tI,16531
 inspect_ai/model/_providers/vllm.py,sha256=RGKMkY9nlScoVx0RTiUCl4usDBSughVsGJjPNQDPDaQ,14230
@@ -508,10 +509,10 @@ inspect_ai/solver/_critique.py,sha256=8GhB9OEsDCOqS6_cHAaTWNNeAimjV3JM_xXr8k1n3X
 inspect_ai/solver/_fork.py,sha256=XKBDwZXRiWkxK_uFtQs-RDJb9gbL6JvfOr_5LtAgS-4,2844
 inspect_ai/solver/_limit.py,sha256=zaZseJgjbJaBnGdXQHQ5MpU4tzgUyD8FzLvJMGDk3jA,1122
 inspect_ai/solver/_multiple_choice.py,sha256=uYr3E6LA6Gv5ktqdmQJX6rYxbc6maTy3ZjyZXabdBug,11298
-inspect_ai/solver/_plan.py,sha256=Dp1DDTtGe2iTo8CYWKqCOdfBFfTK_0wi2JzIr6qrikI,7042
+inspect_ai/solver/_plan.py,sha256=lKeFBIVw_VCLT_V06z-AUS_ABGu3yw-5F1lyC1_0nkw,7190
 inspect_ai/solver/_prompt.py,sha256=jqJ15ysz5_Wn5kadthsZDJWzSoFA7q3IMqSvYW475e8,4901
-inspect_ai/solver/_solver.py,sha256=S3C82Ttan7hGz2yl1OAwjH_GxhwhJVXPnAkHV3OrEEI,9081
-inspect_ai/solver/_task_state.py,sha256=qxcAUL4MUGUuu-SqF8fv2YbUVmZ9VzJTIfeiYl71iXA,16323
+inspect_ai/solver/_solver.py,sha256=41XTYG8a7NyPQKD3ep3Qy2O5Iqtyr3hdxzvof7xc32A,9233
+inspect_ai/solver/_task_state.py,sha256=1EsKFBW5dsIR-VQmobrUxdaR27STwHg9B9USNi-PCI0,16479
 inspect_ai/solver/_transcript.py,sha256=gkH9CC5gYbz7ZzrFD0TkjtKYjWxQP5EthJOkq8NXDOc,1049
 inspect_ai/solver/_use_tools.py,sha256=3Oprsk5FlG5paHLqAurNgj23-endW3_t3cxe28wkMt8,1942
 inspect_ai/solver/_util.py,sha256=pthrf-CzC6FnQYSUFLXTYM4wFEJptZrh5POTmV-Jtow,446
@@ -598,7 +599,7 @@ inspect_ai/util/_panel.py,sha256=MdZxOt0F01ddn_NsRfwn0es6UjQasK1_EKIQ6jtQyG8,312
 inspect_ai/util/_resource.py,sha256=X280aW_7VCkVTGk812tuU5qnZlGM5Qt1-kANr-DaGOs,3389
 inspect_ai/util/_store.py,sha256=QemJe2M-RK6zSFNcd07_92XFjvNtWKgHzBr5eT3KF1I,3786
 inspect_ai/util/_store_model.py,sha256=Z2dvxfh8Re-85pqart4hWlWauerPrK7nKxJoC-0zMhA,4079
-inspect_ai/util/_subprocess.py,sha256=bn1HSHSOH4PX0FZhCvqcJFKKoaEJMgZteEYh_kmZ6fs,8289
+inspect_ai/util/_subprocess.py,sha256=zq6EoiYegmPmGekvnEPKbbLCpGBP2vdCdVbRF4MWxeQ,8524
 inspect_ai/util/_subtask.py,sha256=CFLxdLyj5T1nXmaPKhdYsZARJD8ukoeYF8Wj5m345LM,4729
 inspect_ai/util/_throttle.py,sha256=JczSG_y0v60m4gQCt28uw_WPjJTbHuq8gWcxY3-vFsc,855
 inspect_ai/util/_sandbox/__init__.py,sha256=08yrm-GWL-o7K-NltssDc9zATv2DY7fqAW8p7v6C3Cs,874
@@ -607,19 +608,19 @@ inspect_ai/util/_sandbox/environment.py,sha256=zUVCSEZlIJ16MHT3zMarOxI38FghdE9x0
 inspect_ai/util/_sandbox/limits.py,sha256=K-GjKfSugOq8KP0wW_oF6qFrXsOnMV0C88QUWkjPJ9o,2164
 inspect_ai/util/_sandbox/local.py,sha256=NkHnR_e7s7RFsBdwfaSR7Yzp6lSUc7Em0Pc9_CFuN4c,3534
 inspect_ai/util/_sandbox/registry.py,sha256=mQwWwqzaCXF1FZ2fcVujpp3WMA35GWnh1w43SoIJAVM,2145
-inspect_ai/util/_sandbox/self_check.py,sha256=YaonS0VQyutDpOlHchGhrIUm13juUzEkrkwqLosjgyw,19246
+inspect_ai/util/_sandbox/self_check.py,sha256=ebCSCgJZZkRaKQvrysdzTpsbJ5vXxlOD3swxbb-bOio,19335
 inspect_ai/util/_sandbox/service.py,sha256=2os7W8NYBDcaBoaHVfZ1YrI9hvldksmiwqkUYrCRCPo,11258
-inspect_ai/util/_sandbox/docker/cleanup.py,sha256=MK6UlADcWtTDotppeVJga2ibf9Ud-e4V-5ReoNbmhqg,4793
-inspect_ai/util/_sandbox/docker/compose.py,sha256=g98fwb33EsQQDHKEUYic8tPmi51IA7v3aWm3YoEBEgA,11904
+inspect_ai/util/_sandbox/docker/cleanup.py,sha256=rtpI1NvZX75eDrhpnCO5U7AMCAFB_fEhw73DusPm-Xc,5072
+inspect_ai/util/_sandbox/docker/compose.py,sha256=x5sr-feqGjP96KvsYlfhYBF6eOnv6otIu7zxa2oZv3I,11723
 inspect_ai/util/_sandbox/docker/config.py,sha256=I2sxkN2mTS3kXoAGtJ2w7yuhMoacECgkdxiXftlAvKA,2918
-inspect_ai/util/_sandbox/docker/docker.py,sha256=Dhp2-H-CFPvHD0BlDNTZZtnXj0Jh9_IZjLl7qoyUaD4,17195
+inspect_ai/util/_sandbox/docker/docker.py,sha256=1tJNdCOrgiJ_2N0feCKFvVnjzvM50Pizm5mtMbL_xn4,17622
 inspect_ai/util/_sandbox/docker/internal.py,sha256=fATyk2pdtjSl-D0VPT4dmkXV-gOc5HrPH0EQDW4IAJY,1446
 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=CuYrt9iyLjPSVDEs_oGTas8wAwMwQc_45dZa2g3E4cY,2847
-inspect_ai-0.3.68.dist-info/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
-inspect_ai-0.3.68.dist-info/METADATA,sha256=HQ8Pd2TbpMviMgRkDDYOxIvF6KEJz1omDv_hP7g6E3o,4785
-inspect_ai-0.3.68.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-inspect_ai-0.3.68.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
-inspect_ai-0.3.68.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
-inspect_ai-0.3.68.dist-info/RECORD,,
+inspect_ai/util/_sandbox/docker/util.py,sha256=EeInihCNXgUWxaqZ4dNOJd719kXL2_jr63QCoXn68vA,3154
+inspect_ai-0.3.69.dist-info/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
+inspect_ai-0.3.69.dist-info/METADATA,sha256=Ebnaoeng30uKne0Ltimk6N3jqiLi_11AHzWd9H_Zx2k,4778
+inspect_ai-0.3.69.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+inspect_ai-0.3.69.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
+inspect_ai-0.3.69.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
+inspect_ai-0.3.69.dist-info/RECORD,,