Offline inference framework for running DeepSeek models at large scale: optimized kernels, MoE/Attention parallelism, deterministic sampling and structured outputs. Aimed at teams deploying DeepSeek-V4.1-Flash on NVIDIA SM100 or Huawei Ascend 950.
Useful for
- Run offline inference of DeepSeek-V4.1-Flash across a multi-GPU cluster
- Constrain generation to a JSON schema or grammar via structured outputs
- Get reproducible results across batch sizes with batch-invariant mode
Before using
- License: Apache-2.0
- Installation and requirements have not been verified; review the README before running
Why now
- Created on GitHub
- 2026-10-16 UTC
- First seen by GitNova
- 2026-10-20 UTC
- Magnitude
- 9.3 / 10 · growing interest
Only 3 complete days of history; a reliable usual pace is not established. On 2026-10-19: +3 stars (UTC)
The current gain already exceeds the maximum of the previous 3 complete days: +3 on 2026-10-19 (UTC)
Measured: October 20, 2026, 12:38 UTC
Measurement details
- Created on GitHub: 2026-10-16; first seen by GitNova: 2026-10-20 (UTC)
- +4,651 stars measured 2026-10-20, 09:29–12:24 UTC
- Only 3 complete days of history; a reliable usual pace is not established. On 2026-10-19: +3 stars (UTC)
- The current gain already exceeds the maximum of the previous 3 complete days: +3 on 2026-10-19 (UTC)
- Magnitude 9.3/10