Upstar existe aussi en français.

Voir cette page en français
DA

deepseek-ai/DeepInfer

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

#1 in Python NewPythonApache-2.0 license Last commit 22 hours ago

Updated 8 min ago

Open on GitHub(opens in new tab)

Maintain this repo? Add this badge to your README.

Growth

Last 24 hours
+2.3k
stars
Last 7 days
+17.8k
stars
Last 30 days
+17.8k
stars
Total stars
17,849
created 5 days ago
Forks
1,625

Top 3 in Python this week

  1. 1 deepseek-ai/DeepInfer +17.8k stars this week
  2. 2 google-research/harness2 +17.2k stars this week
  3. 3 Leonxlnx/earshot +2k stars this week

This repo: #1 of 1606 this week, #9 of 422 today, #1 of 2788 this month.

See all 1606 Python repos

Star history

05k10k15k20kAug 6Aug 20Sep 3Sep 17Oct 1Oct 15

Total stars at the end of each week, from GitHub’s daily star counts.

Health signals

Too new to assessChecked Oct 20, 2026

Too new to assess: the repository is less than 60 days old or has fewer than 10 commits. The signals below are shown for information only.

Commit activity
Commits this week (created this week) 3 commits on the default branch in 90 days
Maintainers
2 commit authors in 90 days Top author: 67% of the last 3 commits
Issue responsiveness
Too few issues to tell 0 issues opened by others in 90 days
Pull requests
0 merged, 6 closed, 23 open Of 29 pull requests opened in 90 days
Releases
No GitHub releases Not counted: many projects ship without them
Project hygiene
License; no CI or security policy
OpenSSF Scorecard
Not scanned Unknown, not counted against the project

Signals are computed automatically from public GitHub data. They are not an endorsement or a security audit. How health is computed

README

README.md

DeepInfer

DeepInfer is the high-performance inference framework that powers DeepSeek's large-scale production services. This open-source release provides offline inference with the core features needed for serving at scale.

Supported models:

  • DeepSeek-V4.1-Flash

Supported features:

  • Optimized kernels, model and sampler runtime
  • Parallelism: Attention DP / balanced CP, MoE EP + EPLB
  • DSpark and confidence-scheduled verification
  • SWA bounded replay
  • Asynchronous scheduling (zero-overhead batch scheduler)
  • Batch-invariant inference
  • Deterministic sampling
  • Structured outputs

Supported hardware:

Read the full README on GitHub(opens in new tab)