chadhurley25075-png / pd-bridge
Heterogeneous prefill/decode for DeepSeek-V4-Flash: CUDA prefill (DGX Spark, vLLM) -> Metal decode (Mac Studio, oMLX) over plain 10GbE
Why this job: topics: apple-silicon, llm-inference, vllm, mlx
What it does
Two production inference engines that share no cache format, no framework, no vendor and no quantization, serving one request together. DeepSeek-V4-Flash is 284B total / 13B active, 256 routed experts, MLA + sparse attention, 149 GB resident on the prefill side and 156 GB on the decode side:
Get started
No single install command. Read the steps in the README (opens in new tab)
- Maintenance
- Active
- Last commit 32 days ago
- Stars
- 151
- On GitHub
- Licence
- Apache-2.0
- Allows business use
- Downloads for
- None listed
- No release files
- Latest version
- None yet
- No releases
- Built with
- Python
- 84% of the code
Is it looked after?
- Contributors
- 1
- Last 90 days
- Issues
- 0% closed
- 1 still open
- Pull requests
- 2 merged
- 1 waiting
- Releases
- None
- Never published
- Contributing guide (opens in new tab)
- No security policy
- No code of conduct
Similar repos
| Repo | Stars | Last commit | Licence | Downloads for |
|---|---|---|---|---|
| vllm-project / vllm-metal | 1,974 | today | Apache-2.0 | Mac |
| raullenchai / twinspark | 176 | 2 months ago | MIT | No downloads |
| jundot / omlx | 23,475 | 4 days ago | Apache-2.0 | Mac |
More: languages, topics and project record
Languages
- Python 84%
- C 10%
- Shell 5%
- Makefile 1%
Topics
- apple-silicon
- deepseek
- dgx-spark
- heterogeneous-inference
- kv-cache
- llm-inference
- mlx
- omlx
- prefill-decode
- vllm
Project record
- JobLocal LLMs
- StatusListed
- Commits31
- Forks13
- Watching0
- Stars in 28 days0
- CreatedSep 6, 2026
- First seen by usNov 7, 2026
- CheckedNov 7, 2026