Qwen3.5-9B Eagle3 Draft Model

Eagle3 speculative decoding draft model for Qwen/Qwen3.5-9B.

Trained on ShareGPT data using mini-specforge.

Usage with SGLang

Requires Eagle3 support for Qwen3.5 (pending).

python -m sglang.launch_server \
  --model Qwen/Qwen3.5-9B \
  --speculative-algorithm EAGLE3 \
  --speculative-draft-model-path BLR2/Qwen3.5-9B-Eagle3-ShareGPT \
  --trust-remote-code \
  --mamba-scheduler-strategy extra_buffer

Training Details

  • Architecture: LlamaForCausalLMEagle3 (single decoder layer)
  • Target model: Qwen/Qwen3.5-9B
  • Training data: ShareGPT conversations
  • Training steps: 71,000
  • Draft vocabulary: 32,000 most frequent tokens
  • Precision: BF16
Downloads last month
101
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for BLR2/Qwen3.5-9B-Eagle3-ShareGPT

Finetuned
Qwen/Qwen3.5-9B
Finetuned
(204)
this model