From e9e9ed09a531562d6ccbe56d114162128136aa22 Mon Sep 17 00:00:00 2001 From: nesquena-hermes Date: Wed, 22 Jul 2026 02:51:37 -0700 Subject: [PATCH] Release: #6148 alias-prefix model routing fix (exp-v0.52.141) (#6418) * fix: resolve provider alias before matching _PROVIDER_MODELS (#6148) * fix(backend): resolve model prefix through _PROVIDER_ALIASES before matching - Fix IndentationError: dedent _canon_prefix assignment to 8 spaces - Rebased onto upstream/master - Add regression test: alias prefix (z-ai) resolves to canonical key (zai) and routes cross-provider through OpenRouter (#6131) * fix(#6131): canonicalize both sides via _canonicalise_provider_id (idempotent) + regression tests Codex found _resolve_provider_alias breaks Qwen (qwen->alibaba, not in _PROVIDER_MODELS) and cross-routes alias-of-active-provider (raw vs canon asymmetry). Use the idempotent _canonicalise_provider_id for both membership and equality. Adds 2 regression tests. * docs(changelog): stamp #6148 alias-prefix model routing fix --------- Co-authored-by: webtecnica <75556242+webtecnica@users.noreply.github.com> Co-authored-by: webtecnica Co-authored-by: nesquena-hermes Co-authored-by: nesquena-hermes --- CHANGELOG.md | 2 ++ api/config.py | 8 +++++- tests/test_model_resolver.py | 54 ++++++++++++++++++++++++++++++++++++ 3 files changed, 63 insertions(+), 1 deletion(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index 16e49929..0638ecb4 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -19,6 +19,8 @@ ### Fixed +- **Alias-prefixed models now route to the right provider.** When you pick a namespaced model whose prefix is a provider *alias* (e.g. `z-ai/glm-5.2`, where `z-ai` is an alias of the canonical `zai`) under a different active provider, the WebUI resolver now canonicalizes the prefix before deciding whether to cross-route through OpenRouter — so it reaches OpenRouter instead of falling back to the stale configured provider (which produced auth/404 failures). The comparison canonicalizes both sides, so a model selected under an alias of its *own* provider is correctly treated as same-provider (not cross-routed), and OpenRouter selections like Qwen/xAI keep routing to OpenRouter. Thanks @webtecnica. (#6148, #6131) + - **Per-model reasoning-effort metadata is now honored by the WebUI reasoning picker.** When resolving which reasoning modes a model exposes, a `models..reasoning_efforts` list (as Hermes Agent stores under `custom_providers[].models`) now takes precedence over the provider-wide `reasoning_efforts` list, falling back to the provider list and then heuristics. This lets a model with its own declared modes (e.g. an unrecognized custom model) expose them without a provider-wide override. Configs that only set a provider-level list are completely unaffected — output is byte-identical to before. Thanks @CharlesMcquade. (#6107) - **Stream teardown no longer leaks per-stream runtime state on long-running servers.** When a stream ended, its `STREAM_SESSION_OWNERS` owner mapping and (for gateway turns) the `AGENT_INSTANCES` entry were retained, so a high-volume, long-lived server slowly accumulated dead per-stream bookkeeping. Teardown now unregisters the stream owner and drops the gateway agent instance. The offline-frame buffer is deliberately **not** cleared on teardown so a late reconnect can still replay buffered `token`/`done`/`stream_end` frames; buffers remain bounded (8,192 frames) and channels are reclaimed once outstanding reconnect references exit. Thanks @webtecnica. (#6389, #6351) diff --git a/api/config.py b/api/config.py index 2002e923..0647b945 100644 --- a/api/config.py +++ b/api/config.py @@ -2976,7 +2976,13 @@ def resolve_model_provider(model_id: str, *, explicitly_picked: bool = False) -> # branch (#3872). _cp_lower_cross = (config_provider or "").strip().lower() _is_custom_cross = _cp_lower_cross == "custom" or _cp_lower_cross.startswith("custom:") - if prefix in _PROVIDER_MODELS and prefix != config_provider and not _is_custom_cross: + _canon_prefix = _canonicalise_provider_id(prefix) + _canon_config_provider = _canonicalise_provider_id(config_provider) + if ( + _canon_prefix in _PROVIDER_MODELS + and _canon_prefix != _canon_config_provider + and not _is_custom_cross + ): return model_id, "openrouter", None return model_id, config_provider, config_base_url diff --git a/tests/test_model_resolver.py b/tests/test_model_resolver.py index 4c0a7c42..0dc559f4 100644 --- a/tests/test_model_resolver.py +++ b/tests/test_model_resolver.py @@ -121,6 +121,60 @@ def test_cross_provider_routes_through_openrouter(): assert base_url is None # openrouter uses its own endpoint +def test_cross_provider_routes_through_openrouter_with_alias_prefix(): + """A model whose prefix is a _PROVIDER_ALIASES entry (not itself a + _PROVIDER_MODELS key) but resolves to a canonical provider key still + routes through the cross-provider OpenRouter branch. + + Regression test for #6131 — before the fix, ``z-ai/glm-5.2`` with + config provider ``anthropic`` would fall through and return the stale + config provider because ``z-ai`` was checked directly against + ``_PROVIDER_MODELS`` instead of being resolved through + ``_PROVIDER_ALIASES`` first. + """ + model, provider, base_url = _resolve_with_config( + 'z-ai/glm-5.2', provider='anthropic', + ) + assert model == 'z-ai/glm-5.2' + assert provider == 'openrouter' + assert base_url is None # openrouter uses its own endpoint + + +def test_cross_provider_openrouter_qwen_selection_not_stale_routed(): + """An OpenRouter Qwen selection under a different config provider must + still route through openrouter — NOT inherit the stale config provider. + + Regression guard for #6131's fix: ``qwen`` canonicalises to ``qwen`` + (a _PROVIDER_MODELS key), but its hermes_cli *alias* target is + ``alibaba`` (NOT a _PROVIDER_MODELS key). An earlier fix that resolved + through _PROVIDER_ALIASES instead of the idempotent + _canonicalise_provider_id would have mapped ``qwen`` → ``alibaba``, + missed the membership check, and wrongly returned the stale provider. + """ + model, provider, base_url = _resolve_with_config( + 'qwen/qwen3-max', provider='anthropic', + ) + assert model == 'qwen/qwen3-max' + assert provider == 'openrouter' + assert base_url is None + + +def test_alias_prefix_matching_active_provider_not_cross_routed(): + """A namespaced model whose prefix is an *alias of its own active + provider* must NOT be cross-routed to openrouter. + + ``z-ai/glm-5.2`` under provider ``zai`` — the prefix ``z-ai`` and the + config provider ``zai`` canonicalise to the SAME id (``zai``), so this + is a same-provider selection, not a cross-provider one. The fix compares + canonical prefix against canonical config provider (not the raw prefix) + so the inequality guard correctly suppresses the openrouter cross-route. + """ + model, provider, base_url = _resolve_with_config( + 'z-ai/glm-5.2', provider='zai', + ) + assert provider != 'openrouter' + + # ── Bare model names ───────────────────────────────────────────────────── def test_bare_model_uses_config_provider():