DCore-EAGLE3-Qwen3-8B

Draft model for speculative decoding with Qwen/Qwen3-8B as the target model.

EAGLE-3 baseline, trained by us for comparison in the DCore paper.

This is a draft model: it is conditioned on hidden states of the target model and is not meant to be used as a standalone language model.

Paper: arXiv:2610.24993

Model Details

Item Value
Target model Qwen/Qwen3-8B (36 layers)
Draft architecture LlamaForCausalLMEagle3
Draft layers 1
Hidden size / heads / KV heads 4096 / 32 / 8
Draft vocabulary 32000
Vocabulary 151936 (Qwen3 tokenizer)
Precision bfloat16

Files

  • README.md
  • config.json
  • model.safetensors

Usage

This checkpoint uses the SpecForge / SGLang EAGLE-3 draft format and is meant to be loaded by SGLang as the draft model of Qwen/Qwen3-8B, e.g. --speculative-algorithm EAGLE3 --speculative-draft-model-path ernie-research/DCore-EAGLE3-Qwen3-8B.

Collection

Part of the DCore collection by ERNIE Research.

Downloads last month
178
Safetensors
Model size
0.4B params
Tensor type
I64
·
BF16
·
BOOL
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for ernie-research/DCore-EAGLE3-Qwen3-8B

Finetuned
Qwen/Qwen3-8B
Finetuned
(2,366)
this model

Collection including ernie-research/DCore-EAGLE3-Qwen3-8B

Paper for ernie-research/DCore-EAGLE3-Qwen3-8B