Skip to content

Pull requests: UKGovernmentBEIS/inspect_ai

New pull request New
58 Open 3,228 Closed
58 Open 3,228 Closed
Author
Filter by author
Loading
Label
Filter by label
Loading
Use alt + click/return to exclude labels
or ⇧ + click/return for logical OR
Projects
Filter by project
Loading
Milestones
Filter by milestone
Loading
Reviews
Filter by reviews
No reviews Review required Approved review Changes requested
Assignee
Filter by who’s assigned
Assigned to nobody Loading
Sort
Sort by
Newest Oldest Most commented Least commented Recently updated Least recently updated Best match
Most reactions
👍 👎 😄 🎉 😕 ❤️ 🚀 👀

Pull requests list

Parallel tools: Strengthened system-prompt guidance encouraging react() and deepagent() to issue independent tool calls in parallel
#4085 opened May 29, 2026 by jjallaire Collaborator Loading…
DRAFT: control channel
#4082 opened May 29, 2026 by ransomr Collaborator • Draft
5 tasks
fix(openrouter): replay reasoning_content for deepseek v4
#4080 opened May 28, 2026 by g6000 Loading…
1
scoring-phase resume via solver_done marker
#4078 opened May 28, 2026 by epatey Collaborator • Draft
3 of 5 tasks
docs(scorer): clarify chat_history comment for include_history=True
#4073 opened May 28, 2026 by RecreationalMath Contributor Loading…
1 of 5 tasks
1
docs: clarify include_history grading behavior
#4072 opened May 28, 2026 by he-yufeng Contributor Loading…
1
1
Extract SampleRunner from task_run_sample (PR 1 of EvalSession RFC)
#4070 opened May 28, 2026 by sinamoeini • Draft
5 tasks
Bound transcript memory for long-running samples qualified
#4062 opened May 27, 2026 by rasmusfaber Contributor Loading…
3 of 5 tasks
8
Categorical score support: frequency() metric, StrEnum values, value_schema round-trip qualified
#4058 opened May 27, 2026 by kaifronsdal Collaborator Loading…
6 tasks done
Add OrcaRouter model provider
#4057 opened May 27, 2026 by xilema2 Loading…
2 of 5 tasks
vllm-completions: accept pre-tokenized prompts
#4055 opened May 26, 2026 by Butanium Contributor Loading…
1
test: cover eval_set bundling after retries
#4052 opened May 26, 2026 by deepujain Contributor • Draft
1 of 6 tasks
1
fix: recreate sandbox context during scoring
#4051 opened May 26, 2026 by deepujain Contributor • Draft
1 of 5 tasks
1
fix: warn on unbounded message conversations
#4049 opened May 26, 2026 by deepujain Contributor • Draft
2 of 5 tasks
1
fix(scorer): return NOANSWER (not INCORRECT) when model_graded grade parser fails
#4048 opened May 26, 2026 by vladmesh Loading…
1 of 5 tasks
1
fix: align vLLM logging with Inspect log level
#4047 opened May 26, 2026 by deepujain Contributor • Draft
2 of 5 tasks
1
feat: add OpenRouter app attribution headers
#4046 opened May 26, 2026 by deepujain Contributor • Draft
1
fix: resolve registry task sandbox paths from module dir
#4045 opened May 26, 2026 by deepujain Contributor • Draft
1 of 5 tasks
1
fix: support explicit Azure OpenAI AD token auth
#4044 opened May 26, 2026 by deepujain Contributor • Draft
2 of 5 tasks
1
fix: map metrics to solver score names
#4043 opened May 26, 2026 by deepujain Contributor • Draft
2 of 5 tasks
1
docs: clarify ReAct loop depth controls
#4042 opened May 26, 2026 by deepujain Contributor • Draft
1 of 5 tasks
1
docs: add LLM-to-LLM conversation eval example
#4041 opened May 26, 2026 by deepujain Contributor • Draft
1 of 5 tasks
1
Add Krippendorff's α metric for multi-judge agreement
#4035 opened May 25, 2026 by joesposito8 Contributor Loading…
2 of 5 tasks
Arena: Add pairwise comparison with win rate and Elo metrics
#4034 opened May 25, 2026 by showpiecep • Draft
Fix Bedrock provider to support adaptive thinking and output_config (closes #3765)
#4020 opened May 22, 2026 by ernestprovo23 Contributor Loading…
2 tasks done
1
1
ProTip! Adding no:label will show everything without a label.