DeepSeek硅基Translation site

In today’s rapidly developing AI technology landscape, SiliconFlow and Huawei Cloud have joined forces to launch DeepSeek R1&V3 inference services based on Ascend Cloud, providing developers with efficient and stable large model deployment solutions.

Website Introduction

SiliconFlow’s SiliconCloud platform focuses on providing standardized, high-performance generative AI infrastructure, supporting the most advanced large language models and multimodal model inference globally.

Key Features

• Launched DeepSeek R1&V3 model inference services based on Huawei Cloud’s Ascend Cloud.
• Self-developed inference acceleration engine enables model deployment performance comparable to global high-end GPUs.
• Provides stable, production-grade inference services to meet large-scale production environment needs.
• Zero deployment threshold; developers can directly call SiliconCloud API, focusing on application development.
• Offers highly competitive pricing strategies consistent with DeepSeek’s official promotional period prices.

Related Projects

SiliconFlow has also launched six accelerated versions of DeepSeek-R1 distilled models, including DeepSeek-R1-Distill-Llama-70B, DeepSeek-R1-Distill-Qwen-32B, etc., with some models available for free.

Advantages

The service achieves a single-card decode throughput exceeding 1920 Tokens/s under a single-user 20 TPS level, comparable to H100 deployment performance, with model accuracy consistent with DeepSeek’s official standards.

Pricing

Offers both free and paid versions; DeepSeek-V3‘s promotional period price is ￥1/M tokens (input) & ￥2/M tokens (output), and DeepSeek-R1’s price is ￥4/M tokens (input) & ￥16/M tokens (output).

Summary

Founded in August 2023 and based in China, SiliconFlow is dedicated to providing efficient and stable AI inference services. Through these innovative features, users can achieve a more efficient and convenient large model deployment experience.