Repository navigation
fix(providers): wire MiMo thinking control on gateway providers (#3845) - #3851
Merged
Merged
Conversation
…S#3845) The xiaomi_mimo ProviderSpec carries thinking_style="thinking_type", but gateway providers (OpenRouter etc.) route MiMo under their own spec which has no thinking_style. As a result, `reasoning_effort="none"` was silently ignored: `{"thinking": {"type": "disabled"}}` was never injected and responses still contained reasoning_content. Mirror the Kimi pattern that already handles the same problem: add an explicit _MIMO_THINKING_MODELS allowlist (mimo-v2.5-pro, mimo-v2.5, mimo-v2-pro, mimo-v2-omni — per Xiaomi docs), an _is_mimo_thinking_model helper that strips publisher prefixes ("xiaomi/mimo-v2.5-pro" matches), and a sibling branch in _build_kwargs that injects the thinking payload by model name. mimo-v2-flash is intentionally excluded — it has no thinking mode. Also include MiMo in the explicit_thinking predicate so the reasoning_content backfill (HKUDS#3554, HKUDS#3584) covers the gateway path consistently with the direct path. Tests cover the gateway disable/enable signals, bare-slug fallback, flash exclusion, and a non-MiMo sanity check.
|
I don't know why... It seems the logic to disable thinking is in effect, and I don't have time to delve deeper into the nanobot code right now, but I have a vague guess—is there a similar situation requiring fallingangel@fallingangel-vps~nanob.txt The vim command between the two runs was switching the provider from openrouter to xiaomiMimo — everything else stayed the same. |
Merged
4 of 5 tasks
Contributor
Author
|
@ClearPlume Thanks, you are right: extra_body.thinking is MiMo's upstream shape and OpenRouter strips it and uses its own reasoning field, so the disable signal never reached the model. Follow-up in #3867, it also injects extra_body.reasoning = {effort: "none"} (by OR documentation) when routing Kimi/MiMo through OR. Both fields go out so direct paths unchanged, OR now honors the setting. Would appreciate a verification against your setup once it's merged. |
This was referenced May 17, 2026
Re-bin
pushed a commit
that referenced
this pull request
May 21, 2026
Follow-up to #3851: that PR added `extra_body.thinking={type: disabled}` for MiMo via OpenRouter, but OR doesn't forward provider-specific thinking shapes to upstream — it strips unknown extra_body fields and uses its own unified `reasoning` parameter. So MiMo via OR kept thinking despite the injection (reproduced by @ClearPlume on #3851 with identical kwargs but provider switched from openrouter → xiaomi_mimo). For known thinking-capable models (Kimi, MiMo) routed via the openrouter spec, also inject `extra_body.reasoning = {effort: <effort>}` in OR's documented enum ("none"|"minimal"|"low"|"medium"|"high"|"xhigh"). OR translates this to the upstream model's native shape. Existing tests updated to expect both fields on the OR path. The direct xiaomi_mimo and moonshot paths are unchanged (the new branch is gated on spec.name == "openrouter"). Flash and non-MiMo models on OR continue to receive no injection.
Erlichooo
pushed a commit
to Erlichooo/nanobot
that referenced
this pull request
May 23, 2026
Follow-up to HKUDS#3851: that PR added `extra_body.thinking={type: disabled}` for MiMo via OpenRouter, but OR doesn't forward provider-specific thinking shapes to upstream — it strips unknown extra_body fields and uses its own unified `reasoning` parameter. So MiMo via OR kept thinking despite the injection (reproduced by @ClearPlume on HKUDS#3851 with identical kwargs but provider switched from openrouter → xiaomi_mimo). For known thinking-capable models (Kimi, MiMo) routed via the openrouter spec, also inject `extra_body.reasoning = {effort: <effort>}` in OR's documented enum ("none"|"minimal"|"low"|"medium"|"high"|"xhigh"). OR translates this to the upstream model's native shape. Existing tests updated to expect both fields on the OR path. The direct xiaomi_mimo and moonshot paths are unchanged (the new branch is gated on spec.name == "openrouter"). Flash and non-MiMo models on OR continue to receive no injection.
StefanoParadisi
pushed a commit
to StefanoHernandez/nanobot
that referenced
this pull request
Oct 16, 2026
Follow-up to HKUDS#3851: that PR added `extra_body.thinking={type: disabled}` for MiMo via OpenRouter, but OR doesn't forward provider-specific thinking shapes to upstream — it strips unknown extra_body fields and uses its own unified `reasoning` parameter. So MiMo via OR kept thinking despite the injection (reproduced by @ClearPlume on HKUDS#3851 with identical kwargs but provider switched from openrouter → xiaomi_mimo). For known thinking-capable models (Kimi, MiMo) routed via the openrouter spec, also inject `extra_body.reasoning = {effort: <effort>}` in OR's documented enum ("none"|"minimal"|"low"|"medium"|"high"|"xhigh"). OR translates this to the upstream model's native shape. Existing tests updated to expect both fields on the OR path. The direct xiaomi_mimo and moonshot paths are unchanged (the new branch is gated on spec.name == "openrouter"). Flash and non-MiMo models on OR continue to receive no injection.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
{"thinking": {"type": "disabled"}}) is driven byProviderSpec.thinking_style, which only the dedicatedxiaomi_mimospec carries. When users route MiMo through a gateway (OpenRouter, etc.), the active spec has nothinking_style, soreasoning_effort="none"is silently dropped and the response still containsreasoning_content._MIMO_THINKING_MODELSallowlist +_is_mimo_thinking_model()helper (strips publisher prefix, soxiaomi/mimo-v2.5-promatches) + sibling injection branch in_build_kwargs.tests/providers/test_xiaomi_mimo_thinking.py(per Xiaomi docs):mimo-v2.5-pro,mimo-v2.5,mimo-v2-pro,mimo-v2-omni.mimo-v2-flashis intentionally excluded — it has no thinking mode.explicit_thinkingpredicate so thereasoning_content=""backfill (DeepSeek-V4 reasoning_content error still reproducible on v0.1.5.post3 with WebUI + exec tool on Windows #3554, [Fixed] DeepSeek API 'reasoning_content' error - Root cause identified and Patch provided #3584) covers the gateway path consistently with the direct path.Why an explicit allowlist (not a substring rule)
It matches Kimi's approach, keeps
mimo-v2-flashout without special-casing, and the canonical model list was already enumerated in the existing test docstring — no new policy decision needed.Test plan
pytest tests/providers/test_xiaomi_mimo_thinking.py -v— 11 passed (6 existing + 5 new)pytest tests/providers/ -q— 408 passed, no regressions in adjacent paths (Kimi, DeepSeek, DashScope)ruff check nanobot/providers/openai_compat_provider.py tests/providers/test_xiaomi_mimo_thinking.py— cleanmimo-v2-flashexclusion, non-MiMo non-Kimi sanityNotes for reviewers
xiaomi_mimorequests of allowlisted models (spec-driven + model-name). They write the same payload, so the dict update is an idempotent no-op — no behavior change on the direct path._is_mimo_thinking_model(currently it injects formimo-v2-flashtoo) is left as a follow-up if desired.