Avatar for the BerriAI user
BerriAI
litellm
BlogDocsChangelog

Performance History

Latest Results

Update litellm/proxy/guardrails/guardrail_hooks/singulr/singulr.py Co-authored-by: veria-ai[bot] <224490171+veria-ai[bot]@users.noreply.github.com>
madan-singulr:m.singulr.25jun
13 minutes ago
fix(responses): build a Responses-API-shaped tool_choice for ResponsesAPIResponse ResponsesAPIResponse.tool_choice crashed with 13 Pydantic validation errors whenever a caller forced a named tool call (tool_choice={"type": "function", "function": {"name": "..."}}, the Chat Completion shape) through the litellm_completion Responses API bridge. That field only accepts strings or the flat Responses API shape ({"type": "function", "name": "..."}), never the nested Chat Completion dict. Reproduced against Bedrock/Anthropic models specifically, not because the bug is model-specific, but because OpenAI/Azure/xAI models get native Responses API handling that never touches this bridge, while Bedrock/ Anthropic models always fall back to it (no responses_api_provider_config exists for that provider). Any forced named tool call against a Claude model on Bedrock reproduces this identically, regardless of which Claude variant. Root cause: both the streaming _default_response_created_event_data and the non-streaming transform_chat_completion_response_to_responses_api_response reused the Chat-Completion-shaped value from _transform_tool_choice (which builds the outbound provider request) to also populate the inbound ResponsesAPIResponse.tool_choice field. That function's docstring says it "passes through as-is" for the OpenAI dict format - correct for the outbound request, wrong for the Responses API response shape. Adds _transform_tool_choice_for_responses_api_response, which rewrites the Chat Completion nested shape to the flat Responses API shape, and uses it at both call sites instead of reusing the Chat-Completion-shaped value.
TobiMayr:fix/responses-api-tool-choice-validation
16 minutes ago
fix(anthropic): drop non-1 temperature when downgrading adaptive thinking
axelray-dev:litellm_fix_anthropic_adaptive_temp_33203
30 minutes ago
improved
madan-singulr:m.singulr.25jun
38 minutes ago
fix(router): offload auto-router semantic route to thread and lock cold start
axelray-dev:litellm_fix_auto_router_async_to_thread_33204
43 minutes ago
fix: fetch MCP REST tools concurrently
caldito:litellm_concurrent_mcp_tool_listing
52 minutes ago
fix(mcp): hide public server configuration
caldito:litellm_hide_public_mcp_upstream_details
1 hour ago
feat(providers): add StepFun provider support
Ftgn-dpA:agent/add-stepfun-provider
3 hours ago

Latest Branches

CodSpeed Performance Gauge
0%
feat(guardrails): add Singulr guardrail integration for LiteLLM gateway#31302
34 minutes ago
806bc5b
madan-singulr:m.singulr.25jun
CodSpeed Performance Gauge
0%
fix(responses): build a Responses-API-shaped tool_choice for Response…#33208
1 hour ago
31b4a50
TobiMayr:fix/responses-api-tool-choice-validation
CodSpeed Performance Gauge
0%
fix(anthropic): drop non-1 temperature when downgrading adaptive thinking#33207
51 minutes ago
a74b353
axelray-dev:litellm_fix_anthropic_adaptive_temp_33203
© 2026 CodSpeed Technology
Home Terms Privacy Docs