[PR #384] [MERGED] Add notes to DeepSeek-R1 tutorial documentation #1413

New issue

Closed

opened 2026-10-29 14:52:13 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 14:52:13 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/ktransformers/pull/384
Author: @ubergarm
Created: 2/16/2025
Status: ✅ Merged
Merged: 2/19/2025
Merged by: @Atream

Base: main ← Head: add-tutorial-notes

📝 Commits (1)

bb39eeb Add notes to DeepSeek-R1 tutorial documentation

📊 Changes

2 files changed (+6 additions, -0 deletions)

View changed files

📝 doc/en/DeepseekR1_V3_tutorial.md (+1 -0)
📝 doc/zh/DeepseekR1_V3_tutorial_zh.md (+5 -0)

📄 Description

在急于使用ktransformers结合unsloth最新动态量化技术（例如UD-Q2_K_XL）的过程中，发现存在一些需要澄清的混淆点：

Web界面与API接口目前尚未启用。当前版本会产生逻辑混乱的幻觉输出
仅本地对话功能可正常运作
使用本地对话时必须采用单行提示，任何包含换行符'\n'的粘贴内容都会触发多重生成，导致输出结果紊乱

以上认知是笔者在撰写相关实践指南后总结得出的。后续将在指南中补充说明以预防混淆，同时也为开发团队争取调试时间，避免因用户问题消耗开发精力。

There seems to be some confusion in the rush to use ktransformers
with the new dynamic quantization from unsloth such as
UD-Q2_K_XL

The Web and API interface do not work yet. They will hallucinate.
Only the local chat is working.
You must use a single line prompt with local chat as any copy paste '\n' will trigger multiple generations leading to confusing output.

I learned this after writing a guide on how to do this.

I will update my guide with some notes as well to help prevent confusion and give the devs time to make progress instead of dealing with people.

Thank you for your time.

Examples: https://github.com/kvcache-ai/ktransformers/issues/311#issuecomment-2661559090

#311
#364
#373
#349
#343
#326

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/ktransformers/pull/384 **Author:** [@ubergarm](https://github.com/ubergarm) **Created:** 2/16/2025 **Status:** ✅ Merged **Merged:** 2/19/2025 **Merged by:** [@Atream](https://github.com/Atream) **Base:** `main` ← **Head:** `add-tutorial-notes` --- ### 📝 Commits (1) - [`bb39eeb`](https://github.com/kvcache-ai/ktransformers/commit/bb39eeb005dbe0668d20bb4b02ccd38e8f3bb32c) Add notes to DeepSeek-R1 tutorial documentation ### 📊 Changes **2 files changed** (+6 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `doc/en/DeepseekR1_V3_tutorial.md` (+1 -0) 📝 `doc/zh/DeepseekR1_V3_tutorial_zh.md` (+5 -0) </details> ### 📄 Description 在急于使用ktransformers结合unsloth最新动态量化技术（例如[UD-Q2_K_XL](https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/DeepSeek-R1-UD-Q2_K_XL)）的过程中，发现存在一些需要澄清的混淆点： 1. **Web界面与API接口**目前**尚未启用**。当前版本会产生逻辑混乱的幻觉输出 2. **仅本地对话功能**可正常运作 3. 使用本地对话时**必须采用单行提示**，任何包含换行符'\n'的粘贴内容都会触发多重生成，导致输出结果紊乱以上认知是笔者在撰写[相关实践指南](https://github.com/ubergarm/r1-ktransformers-guide)后总结得出的。后续将在指南中补充说明以预防混淆，同时也为开发团队争取调试时间，避免因用户问题消耗开发精力。 --- There seems to be some confusion in the rush to use ktransformers with the new dynamic quantization from unsloth such as [UD-Q2_K_XL](https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/DeepSeek-R1-UD-Q2_K_XL) 1. The Web and API interface do *not* work yet. They will hallucinate. 2. Only the local chat is working. 3. You must use a single line prompt with local chat as any copy paste '\n' will trigger multiple generations leading to confusing output. I learned this after writing [a guide on how to do this.](https://github.com/ubergarm/r1-ktransformers-guide) I will update my guide with some notes as well to help prevent confusion and give the devs time to make progress instead of dealing with people. Thank you for your time. --- Examples: https://github.com/kvcache-ai/ktransformers/issues/311#issuecomment-2661559090 --- #311 #364 #373 #349 #343 #326 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 14:52:13 +00:00