iomgaa

iomgaa closed issue iomgaa/PolyGateway#5

2026-08-06 12:27:32 +08:00

minimax 的 enable_thinking 是空声明：设 false 完全不生效，实际开关是 reasoning_effort

iomgaa commented on issue iomgaa/PolyGateway#6

2026-08-06 12:26:22 +08:00

采集 completion_tokens_details.reasoning_tokens：推理开销目前无法与生成开销分开归因

已在 1.0.6 落地，关闭。

89ff916 按建议照搬 _coerce_cached_tokens 的形态，新增 transports/openai_compat.py 的 _coerce_reasoning_tokens，从 usage.completion_tokens_details 取值，非负整数…

iomgaa commented on issue iomgaa/PolyGateway#5

2026-08-06 12:25:47 +08:00

minimax 的 enable_thinking 是空声明：设 false 完全不生效，实际开关是 reasoning_effort

已在 1.0.6 落地，关闭。

82f4ec4 把推理开关建模为「形态(provider 级) + 能力(model 级)」两层：ProviderProfile 的 thinking 两档类型放宽为 `Mapping

iomgaa opened issue iomgaa/PolyGateway#7

2026-08-05 00:21:27 +08:00

错误分类：GovernanceBackendError 不在 GatewayUnavailableError 之下，且公开错误面没有文档

iomgaa pushed to main at iomgaa/PolyGateway

2026-08-02 20:22:57 +08:00

a1a9212ba1 docs: state the real downstream impact of the M2.x refusal

iomgaa pushed tag v1.0.6 to iomgaa/PolyGateway

2026-08-02 20:17:32 +08:00

iomgaa pushed to main at iomgaa/PolyGateway

2026-08-02 20:17:32 +08:00

7adcfff0fa feat: make the thinking switch real and collect reasoning tokens (issue #5, #6)

5eb01a0096 chore: release 1.0.6 and keep the live matrix out of the CI gate

48805cb9fb fix: address the independent verification findings (issue #5, #6)

4c135075b3 test: verify the thinking switch against the live API (issue #5, #6)

82f4ec4910 feat: model the thinking switch as shape plus capability (issue #5)

Compare 8 commits »

iomgaa pushed to main at iomgaa/PolyGateway

2026-08-02 14:18:48 +08:00

acc419a29b chore: add a mechanical wiki-vs-source alignment checker

de7273598e docs: correct the scope normalization comment on Redis key impact

Compare 2 commits »

iomgaa commented on issue iomgaa/PolyGateway#6

2026-08-02 11:13:23 +08:00

采集 completion_tokens_details.reasoning_tokens：推理开销目前无法与生成开销分开归因

更正上文的一处论证错误：我把 K3 说成了"反思模型"，实际它在下游项目里是被试底座之一，不是反思模型。这条 issue 的诉求不变，但理由应当改成下面两条，它们比原来的更贴切：

**一，底座开着推理时，推…

iomgaa opened issue iomgaa/PolyGateway#6

2026-08-02 11:03:15 +08:00

采集 completion_tokens_details.reasoning_tokens：推理开销目前无法与生成开销分开归因

iomgaa opened issue iomgaa/PolyGateway#5

2026-08-02 11:02:49 +08:00

minimax 的 enable_thinking 是空声明：设 false 完全不生效，实际开关是 reasoning_effort

iomgaa closed issue iomgaa/PolyGateway#4

2026-08-02 10:18:20 +08:00

chat() 无法传递采样参数：temperature / seed / max_tokens 都够不着，下游无法固定解码

iomgaa commented on issue iomgaa/PolyGateway#4

2026-08-02 10:18:02 +08:00

chat() 无法传递采样参数：temperature / seed / max_tokens 都够不着，下游无法固定解码

已在 v1.0.5 实现，两个层次都支持（ce630a3 合入 main，包已发到 Gitea PyPI）。

实现

iomgaa pushed tag v1.0.5 to iomgaa/PolyGateway

2026-08-01 12:03:42 +08:00

iomgaa pushed to main at iomgaa/PolyGateway

2026-08-01 12:03:25 +08:00

ce630a37ef feat: pass sampling parameters through to the provider (issue #4)

dfda59fec2 chore: release 1.0.5 with sampling parameter passthrough

15b9b02e96 fix: make the sampling invariant test actually enforce the constraint

cce7562d07 test: verify sampling parameters through the full governance stack

2958dc8231 docs: document sampling passthrough and the empty thinking profiles

Compare 17 commits »

iomgaa closed issue iomgaa/PolyGateway#3

2026-07-31 23:16:22 +08:00

LLMResponse 缺两个可观测字段：供应商 prompt cache 命中 token 数、API 实际返回的模型版本串

iomgaa commented on issue iomgaa/PolyGateway#3

2026-07-31 23:16:18 +08:00

LLMResponse 缺两个可观测字段：供应商 prompt cache 命中 token 数、API 实际返回的模型版本串

已随 v1.0.4 发布并合并到 main(486809b,tag v1.0.4)。

版本号取 patch 而非 minor:虽然加了字段、扩了遥测端口、动了表结构,但对下游是零改动——新字段带默认值,旧价格表与旧缓存条目照常工作,遥…

iomgaa pushed tag v1.0.4 to iomgaa/PolyGateway

2026-07-31 23:16:04 +08:00

iomgaa pushed to main at iomgaa/PolyGateway

2026-07-31 23:16:04 +08:00

486809b08b feat: expose provider cache tokens and reported model (issue #3)

58cb55b869 chore: release 1.0.4 instead of a minor bump

86fb4d5536 fix: keep the postgres backfill from disabling telemetry or locking the table

32d7869043 fix: harden the observability fields against the verifier findings

966d548245 chore: release 1.1.0 with the response observability fields

Compare 18 commits »

iomgaa commented on issue iomgaa/PolyGateway#3

2026-07-31 22:44:32 +08:00

LLMResponse 缺两个可观测字段：供应商 prompt cache 命中 token 数、API 实际返回的模型版本串

已在 v1.1.0 实现,两个字段的最终形态如下。

新增字段

cached_prompt_tokens: int