inspect-ai 0.3.97__py3-none-any.whl → 0.3.98__py3-none-any.whl

inspect_ai/_eval/run.py

@@ -475,7 +475,12 @@ async def startup_sandbox_environments(
     sandboxenvs: Set[TaskSandboxEnvironment] = set()
     for task in tasks:
         # resolve each sample and add to sandboxenvs
-        dataset = slice_dataset(task.task.dataset, config.limit, config.sample_id)
+        resolved_task_sample_ids = resolve_task_sample_ids(
+            task.task.name, config.sample_id
+        )
+        dataset = slice_dataset(
+            task.task.dataset, config.limit, resolved_task_sample_ids
+        )
         for sample in dataset:
             sandbox = await resolve_sandbox_for_task_and_sample(
                 eval_sandbox, task.task, sample

inspect_ai/dataset/_dataset.py

@@ -51,7 +51,6 @@ class Sample(BaseModel):
                 or narrative text to be used by a model grader.
             id: Optional. Unique identifier for sample.
             metadata: Optional. Arbitrary metadata associated with the sample.
-                sandbox (SandboxEnvironmentType | None): Sandbox environment type (or optionally a str or tuple with a shorthand spec)
             sandbox: Optional. Sandbox specification for this sample.
             files: Optional. Files that go along with the sample (copied to
                 SandboxEnvironment). Files can be paths, inline text, or inline binary (base64 encoded data URL).

inspect_ai/model/_providers/google.py

@@ -350,6 +350,12 @@ class GoogleGenAIAPI(ModelAPI):
             self.is_gemini() and not self.is_gemini_1_5() and not self.is_gemini_2_0()
         )
         if has_thinking_config:
+            if config.reasoning_tokens == 0:
+                # When reasoning_tokens is set to zero, we disable reasoning and return None.
+                # We cannot return a ThinkingConfig with reasoning_tokens set to 0,
+                # as this will cause the Gemini API to return a 400 INVALID_ARGUMENT error.
+                return None
+
             return ThinkingConfig(
                 include_thoughts=True, thinking_budget=config.reasoning_tokens
             )

{inspect_ai-0.3.97.dist-info → inspect_ai-0.3.98.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: inspect_ai
-Version: 0.3.97
+Version: 0.3.98
 Summary: Framework for large language model evaluations
 Author: UK AI Security Institute
 License: MIT License
@@ -49,7 +49,7 @@ Requires-Dist: semver>=3.0.0
 Requires-Dist: shortuuid
 Requires-Dist: sniffio
 Requires-Dist: tenacity
-Requires-Dist: textual>=0.86.2
+Requires-Dist: textual<v3.0.0,>=0.86.2
 Requires-Dist: typing_extensions>=4.9.0
 Requires-Dist: zipp>=3.19.1
 Provides-Extra: dev

{inspect_ai-0.3.97.dist-info → inspect_ai-0.3.98.dist-info}/RECORD

@@ -50,7 +50,7 @@ inspect_ai/_eval/evalset.py,sha256=Gvj22yy5WI_SOJYfVdtC-qmMjzdBOFTSq_RZznIQnZo,2
 inspect_ai/_eval/list.py,sha256=VbZ-2EI6MqrXvCN7VTz21TQSoU5K5_Q0hqhxmj5A_m0,3744
 inspect_ai/_eval/loader.py,sha256=dafv4TlQDqdvzPyrQrBsNiCzhvqjwmcVQzweX-AL1os,24805
 inspect_ai/_eval/registry.py,sha256=8Cm-qyDB6Fthea8DUe-QES9plly_Pf2MUuCgeNQ3fOY,5303
-inspect_ai/_eval/run.py,sha256=cv54aTFOi8nMWSG0Y40Z9o07V3nA-0_9uufgQDFP89k,20586
+inspect_ai/_eval/run.py,sha256=QOu4OMaQey--c3OPRxEtg-01jUfKiMFUDcpweMW8rR0,20731
 inspect_ai/_eval/score.py,sha256=ns5X3NJBfQ9bcue54wrLIdc2674ofcua1ZtX3ZOwkOI,10477
 inspect_ai/_eval/task/__init__.py,sha256=6FvojMW3yo36L7xDacppCHDxt6A8_tzj_ftg5bQ6eNk,199
 inspect_ai/_eval/task/constants.py,sha256=quAKMw-4-3xKd1T_KwXCZvHYoKRXt1ZGuaHbBcWJwnA,72
@@ -549,7 +549,7 @@ inspect_ai/approval/_human/manager.py,sha256=Igae35VS99TejSWUShNwFuVnmhwByK30H84
 inspect_ai/approval/_human/panel.py,sha256=UaO309bn7zMDV1I5GJcrZThhU8c8D4gOW7bwPRPPdkc,7672
 inspect_ai/approval/_human/util.py,sha256=DPxpA9_pzeoQyHpGtRE3hx8sGV7MyrlElCBvsh1FNgA,1996
 inspect_ai/dataset/__init__.py,sha256=4uTSHpN_ccdtbZulUMDetSSP-dXRkFGYsa2FzA5mLEw,534
-inspect_ai/dataset/_dataset.py,sha256=V2tS5vuUgsplUD_CUYG5vXsi5kqnvVp4XT7pfw9zj4Q,12071
+inspect_ai/dataset/_dataset.py,sha256=CrDsUaoyJg5ijApVPxQgCiZRbkO-QA6xgv39q3LzQRc,11936
 inspect_ai/dataset/_util.py,sha256=u2IDIJdsa5lOsC67SNWZo0mRSey5DJdxS4DCCxSx7kE,7857
 inspect_ai/dataset/_examples/bias_detection.jsonl,sha256=ufXUZMjsJhY2_lJ9j_iJ6w0fiuyw7TQmJeMFoOqKYzM,20756
 inspect_ai/dataset/_examples/biology_qa.jsonl,sha256=Hfkr8XRT2x6Cgjd-zKbWGfFD-ZEfi566FqLX4w9USKs,2009
@@ -609,7 +609,7 @@ inspect_ai/model/_providers/anthropic.py,sha256=8R7lX1lKdVmTRksTpQ7RIFU_gj6rB5v0
 inspect_ai/model/_providers/azureai.py,sha256=EUtgc-qyIqzV0Oy3C0ztKgYU4ID4z5xwuOxtoRn8WoY,14607
 inspect_ai/model/_providers/bedrock.py,sha256=G252v6gUXtT56M4JaLLY7tEw2AJVQFucjeFgv0okhgo,23999
 inspect_ai/model/_providers/cloudflare.py,sha256=9yHfA5qbKWjzOfOzCJ_u8CZsH_U7AolAWLxvLBXKrhM,2375
-inspect_ai/model/_providers/google.py,sha256=qL3cK1HbEnFrWAXR9SR7DpF5LfXjEK95TEfX5ga4OYM,32259
+inspect_ai/model/_providers/google.py,sha256=yp_UWN2JHbeqic9MV8nDMtkxyb0Z4pwHSwJ9ygTRXy8,32603
 inspect_ai/model/_providers/grok.py,sha256=iAPXmZMR7VWPq6EIwRsoUJr_TR6b5kTt-Fkba1pogGQ,1267
 inspect_ai/model/_providers/groq.py,sha256=q9o4sy0uUyLQbSThB-MMTmSc5AtKmb6GJfgHBpf5amM,12262
 inspect_ai/model/_providers/hf.py,sha256=jyXi4qyq2hdsp1waB2ON5m8f9mpE2h1GFD7Tu_phCEo,19115
@@ -744,9 +744,9 @@ 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-0.3.97.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
-inspect_ai-0.3.97.dist-info/METADATA,sha256=1jgLRbZ0oobGShNJf8VmtU4FAJv9D-OSjbclYCMs9i8,5438
-inspect_ai-0.3.97.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
-inspect_ai-0.3.97.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
-inspect_ai-0.3.97.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
-inspect_ai-0.3.97.dist-info/RECORD,,
+inspect_ai-0.3.98.dist-info/licenses/LICENSE,sha256=xZPCr8gTiFIerrA_DRpLAbw-UUftnLFsHxKeW-NTtq8,1081
+inspect_ai-0.3.98.dist-info/METADATA,sha256=AotZ1r5SK0mv2OC_fb3_1VKl9IowgTwrcTobo9f-z3s,5446
+inspect_ai-0.3.98.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+inspect_ai-0.3.98.dist-info/entry_points.txt,sha256=WGGLmzTzDWLzYfiyovSY6oEKuf-gqzSDNOb5V-hk3fM,54
+inspect_ai-0.3.98.dist-info/top_level.txt,sha256=Tp3za30CHXJEKLk8xLe9qGsW4pBzJpEIOMHOHNCXiVo,11
+inspect_ai-0.3.98.dist-info/RECORD,,