---
title: "ThinkingCap×Gemma-4：少thinking、分不太掉的开源向工作 - Angryecho"
description: "BottleCap AI公开了对gemma-4-26B-A4B-it的ThinkingCap结果：thinking token平均约少39%，极限约69%，准确率变化约0.15pp。属于围绕开源模型的后训练/压缩实践分享，对要自托管推理的人有参考价值。[1] 落地前核对他们的评测表和发布物是否开放权重。"
canonical_url: "https://www.angryecho.com/c/Python/p/P2TBBiRrDoNU9bKhciYPp"
meta:
  "og:description": "BottleCap AI公开了对gemma-4-26B-A4B-it的ThinkingCap结果：thinking token平均约少39%，极限约69%，准确率变化约0.15pp。属于围绕开源模型的后训练/压缩实践分享，对要自托管推理的人有参考价值。[1] 落地前核对他们的评测表和发布物是否开放权重。"
  "og:image": "https://www.angryecho.com/_og/d/c_Post%2Ctitle_~5ZCQ5qe9572R%2Cdescription_~5ZCQ5qe95LiA5LiL77yM5bCx5LiK5ZCQ5qe9572R44CC%2CpostId_P2TBBiRrDoNU9bKhciYPp%2Crevision_~MjAyNi0xMC0yN1QwMjo1Nzo0OS4yMjMzNTBa%2Cp_Ii9jL1B5dGhvbi9wL1AyVEJCaVJyRG9OVTliS2hjaVlQcCI%2Cs_2vJmI3o6CMTbLJPH.png"
  "og:image:height": "630"
  "og:image:type": image/png
  "og:image:width": "1200"
  "og:title": ThinkingCap×Gemma-4：少thinking、分不太掉的开源向工作
---

> Index of every page on this site: [llms.txt](https://www.angryecho.com/llms.txt)

# **ThinkingCap×Gemma-4：少thinking、分不太掉的开源向工作**

BottleCap AI公开了对`gemma-4-26B-A4B-it`的ThinkingCap结果：thinking token平均约少39%，极限约69%，准确率变化约0.15pp。属于围绕开源模型的后训练/压缩实践分享，对要自托管推理的人有参考价值。\[1]

落地前核对他们的评测表和发布物是否开放权重。文章： https://bottlecapai.com/post/thinkingcap-gemma-4-26b/

> \[1] [BottleCap AI](https://bottlecapai.com/post/thinkingcap-gemma-4-26b/)，2026年10月23日。

#开源模型 #Gemma #推理优化 #Python #推荐

排序方式 :

### **成为第一条评论！**

Nobody's responded to this post yet. Add your thoughts and get the conversation going.

```json
{"@context":"https://schema.org","@graph":[{"@id":"https://www.angryecho.com/c/Python/p/P2TBBiRrDoNU9bKhciYPp#post","@type":"DiscussionForumPosting","commentCount":0,"dateModified":"2026-10-27T02:57:49.223Z","datePublished":"2026-10-23T04:00:00.000Z","description":"BottleCap AI公开了对gemma-4-26B-A4B-it的ThinkingCap结果：thinking token平均约少39%，极限约69%，准确率变化约0.15pp。属于围绕开源模型的后训练/压缩实践分享，对要自托管推理的人有参考价值。[1] 落地前核对他们的评测表和发布物是否开放权重。","headline":"ThinkingCap×Gemma-4：少thinking、分不太掉的开源向工作","isPartOf":"https://www.angryecho.com/c/Python","mainEntityOfPage":"https://www.angryecho.com/c/Python/p/P2TBBiRrDoNU9bKhciYPp","text":"BottleCap AI公开了对gemma-4-26B-A4B-it的ThinkingCap结果：thinking token平均约少39%，极限约69%，准确率变化约0.15pp。属于围绕开源模型的后训练/压缩实践分享，对要自托管推理的人有参考价值。[1]\n\n落地前核对他们的评测表和发布物是否开放权重。文章：https://bottlecapai.com/post/thinkingcap-gemma-4-26b/\n\n\n\n[1] BottleCap AI，2026年10月23日。\n\n#开源模型 #Gemma #推理优化 #Python #推荐","url":"https://www.angryecho.com/c/Python/p/P2TBBiRrDoNU9bKhciYPp","author":{"@type":"Person","name":"Shivam","url":"https://www.angryecho.com/u/P7H10043798"},"interactionStatistic":[{"@type":"InteractionCounter","interactionType":"https://schema.org/LikeAction","userInteractionCount":0},{"@type":"InteractionCounter","interactionType":"https://schema.org/DislikeAction","userInteractionCount":0},{"@type":"InteractionCounter","interactionType":"https://schema.org/ViewAction","userInteractionCount":0},{"@type":"InteractionCounter","interactionType":"https://schema.org/ShareAction","userInteractionCount":0}]}]}
```

## Sitemap

See the full [sitemap](https://www.angryecho.com/sitemap.md) for all pages.