vLLM Supports DFlash Inference Acceleration Checkpoints
vllm_project · x · 2026-07-15
Red Hat AI has released DFlash speculative checkpoints for two of NVIDIA's open models:
- Nemotron Ultra 550B
- Nemotron Super 120B
Reported performance metrics include:
- Average acceptance of about 5/7 draft tokens on math/reasoning tasks
- Average acceptance of about 3.4/7 draft tokens on coding tasks (HumanEval)
These checkpoints were trained using the open-source Speculators library, are licensed under Apache 2.0, and have been validated on the NVIDIA B200. Users can enable them simply by adding the --speculative-config parameter in vLLM.
Related event: Red Hat AI Releases DFlash Inference Checkpoints for NVIDIA Models(2 posts)→
More from Infra
- 12 KV Cache Reduction Techniques Every AI Engineer Should Understand, Explained — blaizedsouza · 2026-09-11
- The shadow GPU capacity market is formalizing, with Meta selling excess compute to outside buyers — DavidLinthicum · 2026-09-11
- Engram's random reads don't suit SSDs; CPU-memory over NVLink could serve all 72 GPUs — bookwormengr · 2026-09-11
- 80% of the DIY LLM inference hype posters have already quit — it's brutally hard systems work — abhijithneil · 2026-09-11
- Hugging Face's Ultra Scale Playbook: a free book on training LLMs on GPU clusters — mdancho84 · 2026-09-11
- Is inference latency becoming the biggest bottleneck for production AI agents? — Euphoric_Sea632 · 2026-09-11