Code to train and evaluate DSpark draft models for speculative decoding using the speculators library and vllm
-
Updated
Sep 17, 2026 - Jupyter Notebook
#
speculators
Here are 2 public repositories matching this topic...
Speculative-draft training pipeline (DFLASH/DSPARK/DFLASH2), orchestrating the vllm-project/speculators engine end to end, with preset configs for Qwen-family models
-
Updated
Sep 21, 2026 - Python
Add this topic to your repo
To associate your repository with the speculators topic, visit your repo's landing page and select "manage topics."