Qwen3-8B-speculator.eagle3

Model Overview

Verifier: Qwen/Qwen3-8B
Speculative Decoding Algorithm: EAGLE-3
Model Architecture: Eagle3Speculator
Release Date: 07/27/2025
Version: 1.0
Model Developers: RedHat

This is a speculator model designed for use with Qwen/Qwen3-8B, based on the EAGLE-3 speculative decoding algorithm. It was trained using the speculators library on a combination of the Aeala/ShareGPT_Vicuna_unfiltered and the HuggingFaceH4/ultrachat_200k datasets. The model was trained with thinking turned disabled. This model should be used with the Qwen/Qwen3-8B chat template, specifically through the /chat/completions endpoint.

Use with vLLM

vllm serve Qwen/Qwen3-8B \
  -tp 1 \
  --speculative-config '{
    "model": "RedHatAI/Qwen3-8B-speculator.eagle3",
    "num_speculative_tokens": 3,
    "method": "eagle3"
  }'

Evaluations

Use cases

Use Case	Dataset	Number of Samples
Coding	HumanEval	168
Math Reasoning	gsm8k	80
Text Summarization	CNN/Daily Mail	80

Acceptance lengths

Use Case	k=1	k=2	k=3	k=4	k=5	k=6	k=7
Coding	1.72	2.17	2.39	2.59	2.60	2.59	2.69
Math Reasoning	1.73	2.20	2.48	2.63	2.72	2.79	2.81
Text Summarization	1.62	1.96	2.13	2.24	2.25	2.29	2.30

Performance benchmarking (1xA100)

Details Configuration

temperature: 0.6
top_p: 0.95
top_k: 20
repetitions: 3
time per experiment: 10min
hardware: 1xA100
vLLM version: 0.11.0
GuideLLM version: 0.3.0

Command

GUIDELLM__PREFERRED_ROUTE="chat_completions" \
guidellm benchmark \
  --target "http://localhost:8000/v1" \
  --data "RedHatAI/speculator_benchmarks" \
  --data-args '{"data_files": "HumanEval.jsonl"}' \
  --rate-type sweep \
  --max-seconds 600 \
  --output-path "Qwen3-8B-HumanEval.json" \
  --backend-args '{"extra_body": {"chat_completions": {"temperature":0.6, "top_p":0.95, "top_k":20}}}'

GuideLLM interface changed, so for compatibility with the latest version (v0.6.0), please use the following command:

GUIDELLM__PREFERRED_ROUTE="chat_completions" \
guidellm benchmark \
  --target "http://localhost:8000/v1" \
  --data "RedHatAI/speculator_benchmarks" \
  --data-args '{"data_files": "HumanEval.jsonl"}' \
  --profile sweep \
  --max-seconds 1800 \
  --output-path "my_output.json" \
  --backend-args '{"extras": {"body": {"temperature":0.6, "top_p":0.95, "top_k":20}}}'