Deploy DeepSeek-V3 and R1 Models with One-Click on Alibaba Cloud PAI Model Gallery

DeepSeek-V3 and R1 Series Overview

DeepSeek‑V3 is a Mixture‑of‑Experts (MoE) large language model released by DeepSeek, with a total of 671 billion parameters and 37 billion active parameters per token. It employs Multi‑head Latent Attention (MLA) and the DeepSeekMoE architecture, introduces a load‑balancing strategy without auxiliary loss, and adopts multi‑token prediction as a training objective. The model was pretrained on 14.8 trillion high‑quality tokens and subsequently refined through supervised fine‑tuning (SFT) and reinforcement learning.

DeepSeek‑R1 is a high‑performance AI inference model with 660 billion parameters. After extensive reinforcement‑learning fine‑tuning, its inference capabilities on mathematics, code, and natural‑language reasoning tasks match those of OpenAI’s o1 model.

Supported Models and Distillation Variants

DeepSeek‑R1 is available in several distilled versions that pair with open‑source base models:

DeepSeek‑R1‑Distill‑Qwen‑1.5B → Qwen2.5‑Math‑1.5B

DeepSeek‑R1‑Distill‑Qwen‑7B → Qwen2.5‑Math‑7B

DeepSeek‑R1‑Distill‑Llama‑8B → Llama‑3.1‑8B

DeepSeek‑R1‑Distill‑Qwen‑14B → Qwen2.5‑14B

DeepSeek‑R1‑Distill‑Qwen‑32B → Qwen2.5‑32B

DeepSeek‑R1‑Distill‑Llama‑70B → Llama‑3.3‑70B‑Instruct

The PAI Model Gallery now supports one‑click deployment of DeepSeek‑V3, DeepSeek‑R1, and all distilled small‑parameter models.

PAI Model Gallery Overview

PAI Model Gallery is a component of Alibaba Cloud’s AI platform that aggregates high‑quality pretrained models from global open‑source communities, covering LLM, AIGC, computer vision, and NLP domains such as Qwen and DeepSeek series. By adapting these models on PAI, users can develop, train, deploy, and infer without writing code, streamlining the AI development workflow.

Access the gallery at https://pai.console.aliyun.com/#/quick-start/models .

One‑Click Deployment Guide

Log in to the PAI console.

Select the appropriate region in the top‑left corner.

Choose a workspace from the left navigation and enter it.

Navigate to Quick Start > Model Gallery .

In the model list, click the desired model card (e.g., “DeepSeek‑R1‑Distill‑Qwen‑7B”) to open its detail page.

Click Deploy in the upper‑right corner. DeepSeek‑R1 can be accelerated with vLLM; DeepSeek‑V3 supports vLLM and Web deployment; distilled models support BladeLLM or vLLM.

After deployment, retrieve the endpoint and token from the service page and follow the provided usage instructions.

Developers are encouraged to follow the gallery for future SOTA model releases and can join the user community via DingTalk (group 79680024618) for support.