Hacker GitHub

← All repos

deepseek-ai/DeepInfer

View on GitHub ↗ ★ 25,730 Python

DeepInfer: High-performance, large-scale inference framework for DeepSeek models

6 mentions on Hacker News by 6 people: 1 story, 5 comments · first 2026-10-20 · latest 2026-11-03

Mentions per year 6 mentions · busiest year 2026 (6)
20082026
  1. Deep seek have released their inference engine. https://github.com/deepseek-ai/DeepInfer
  2. …inference engine, revealing that they're still running on only H100s and Huawei Ascend chips ( https://github.com/deepseek-ai/DeepInfer/blob/main/README.md ). Anthropic will be running their 2027 Opus and Fable models on Vera Rubin, a chip…
  3. …155 trillion tokens per day, with overall revenue of US$38 per GPU-hour (at valley pricing)." https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer... So this implies they actually hit about double that during peak pricing.
  4. …just disabuse us of that notion? > …with overall revenue of US$38 per GPU-hour (at valley pricing) https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer...
  5. They use 3FS [0] also open sourced [1] [0] https://github.com/deepseek-ai/DeepInfer/blob/main/doc/infer... [1] https://github.com/deepseek-ai/3fs
  6. ▲ 2,355
    DeepInfer: High-performance inference framework for DeepSeek models
    story by ggcr · 2026-10-20 · 892 comments