[PR #891] docs: update native Transformers support for DeepSeek-R1 #801

New issue

Open

opened 2026-10-26 21:26:53 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 21:26:53 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepSeek-R1/pull/891
Author: @KennyMcSimpson
Created: 10/24/2026
Status: 🔄 Open

Base: main ← Head: docs/native-transformers-support

📝 Commits (1)

8136a52 docs: update native Transformers support for DeepSeek-R1

📊 Changes

1 file changed (+3 additions, -1 deletions)

View changed files

📝 README.md (+3 -1)

📄 Description

The local-running section still says Hugging Face Transformers has no direct support. The published R1 config declares model_type: deepseek_v3, and Transformers provides a native DeepSeek-V3 implementation.

Replace the outdated note with a link to the official Transformers documentation. This describes architecture support without adding unverified full-checkpoint deployment instructions.

Validation: with Transformers 4.51.3 and PyTorch 2.4.1+cpu, the published R1 configuration resolves to the native DeepseekV3Config with trust_remote_code=False. A reduced, unquantized configuration resolves to native DeepseekV3ForCausalLM, produces finite CPU logits, and generates two tokens. No full model weights or FP8 loading were tested. git diff --check passes.

Related to #207.

_{🔄 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/deepseek-ai/DeepSeek-R1/pull/891 **Author:** [@KennyMcSimpson](https://github.com/KennyMcSimpson) **Created:** 10/24/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `docs/native-transformers-support` --- ### 📝 Commits (1) - [`8136a52`](https://github.com/deepseek-ai/DeepSeek-R1/commit/8136a5286bf104886a0b7a3cf024f0e240b8b074) docs: update native Transformers support for DeepSeek-R1 ### 📊 Changes **1 file changed** (+3 additions, -1 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+3 -1) </details> ### 📄 Description The local-running section still says Hugging Face Transformers has no direct support. The [published R1 config](https://huggingface.co/deepseek-ai/DeepSeek-R1/blob/56d4cbbb4d29f4355bab4b9a39ccb717a14ad5ad/config.json) declares `model_type: deepseek_v3`, and Transformers provides a native DeepSeek-V3 implementation. Replace the outdated note with a link to the official Transformers documentation. This describes architecture support without adding unverified full-checkpoint deployment instructions. Validation: with Transformers 4.51.3 and PyTorch 2.4.1+cpu, the published R1 configuration resolves to the native `DeepseekV3Config` with `trust_remote_code=False`. A reduced, unquantized configuration resolves to native `DeepseekV3ForCausalLM`, produces finite CPU logits, and generates two tokens. No full model weights or FP8 loading were tested. `git diff --check` passes. Related to #207. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 21:26:53 +00:00