Commit Graph
6 Commits
Author SHA1 Message Date
Wesley Liddick 0ef2228ce9 Merge pull request #4906 from alfadb/feature/kimi-k3-support
feat: 支持 Kimi K3 与 Kimi Code 模型 ID
2026-07-28 11:23:53 +08:00
alfadb 7e65eafbe4 feat: add Kimi K3 support 2026-07-28 10:08:39 +08:00
HuntercodeT 83b368553d fix(openai): honor passthrough over model_mapping in IsModelSupported (#4936)
An OpenAI account with auto-passthrough enabled (extra.openai_passthrough=true,
"replace auth only, allow all models") was still filtered out during account
selection when it had a non-empty credentials.model_mapping that did not list the
requested model. isOpenAICompatibleAccountEligibleForRequest calls
Account.IsModelSupported directly, and IsModelSupported only bypassed the mapping
whitelist for passthrough accounts when the mapping was empty. A leftover mapping
(common after switching an account from whitelist mode to passthrough) therefore
excluded the account -> zero candidates -> ErrNoAvailableAccounts, and the client
saw 404 "Model ... is not supported by any configured account in this group" even
though a direct account test with the same model succeeded (that path already
honored passthrough via isModelSupportedByAccount).

Fix: short-circuit passthrough at the top of Account.IsModelSupported, before the
model_mapping check, so it is consistent with isModelSupportedByAccount. Add a
regression test covering passthrough + non-empty leftover mapping.
2026-07-27 14:20:12 +08:00
li fa01aec80d fix: 前缀全部加连字符,与 thinking_protocol.go 风格对齐
deepseek → deepseek-、moonshot → moonshot-、qwen → qwen-/qwen2-/qwen3-/qwen4-、
minimax → minimax-、llama → llama-/llama2-/llama3-、mistral → mistral-、
mixtral → mixtral-。避免无连字符前缀误伤假想的连写模型名。
2026-07-07 15:00:00 +08:00
li 9b75c7b76c fix(scheduler): 改用异族厂商前缀黑名单,兼容渠道级模型映射
自审发现白名单守卫有误伤面:渠道级模型映射在账号选定之后才改写
请求体模型名,调度过滤看到的是改写前的原始别名。若按 Codex 白名单
判定,「渠道映射自定义别名 → gpt-5.x + 空映射 OAuth 账号」的既有
部署会被误判为 model_not_found。

改为保守 fail-open 黑名单:空映射默认仍允许,仅排除 deepseek/glm/
kimi/qwen/gemini/grok 等明确异族厂商前缀(Codex 上游绝无可能服务,
透传必然换来不可重试 400)。前缀分类先例见 ResolveThinkingProtocol。
2026-07-07 14:19:58 +08:00
li 2a3dcb499f fix(scheduler): 空 model_mapping 的 OpenAI OAuth 账号不再吸收全部模型
空映射此前一律视为「允许所有」,导致混合部署下请求非 OpenAI 模型
别名(deepseek-v4 等)时,OAuth 账号可能被调度选中;转发阶段对未知
模型原样透传,Codex 上游返回不可重试的 400,请求无法 failover 到
真正支持该模型的 API Key 账号。

IsModelSupported 对空映射的 OpenAI OAuth(未开透传)账号改为与转发
归一化对齐的判定 isOpenAIOAuthServableModel:可归一到已知 Codex 模型
集合(含 gpt-image-*、推理后缀、别名拼写)或 claude-* dispatch 系列
才视为可服务。API Key/显式映射/透传/其他平台语义不变。

Fixes #3662
2026-07-07 14:06:21 +08:00