Qwen3-8B-Thinking-speculator.eagle3

Qwen3-8B-Thinking-speculator.eagle3

Model Overview

• Verifier: Qwen3-8B
• Speculative Decoding Algorithm: EAGLE-3
• Model Architecture: Eagle3Speculator
• Release Date: 3/12/2026
• Version: 1.0
• Model Developers: RedHat

This is a speculator model designed for use with Qwen3-8B, based on the EAGLE-3 speculative decoding algorithm. It was trained using the speculators library on a combination of the Magpie-Align/Magpie-Pro-300K-Filtered and the HuggingFaceH4/ultrachat_200k datasets. The model was trained with thinking enabled. This model should be used with the 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-Thinking-speculator.eagle3",
    "num_speculative_tokens": 5,
    "method": "eagle3"
  }'

<h3>Acceptance lengths (draft length)</h3> <table> <thead> <tr> <th>Dataset</th> <th>k=1</th> <th>k=2</th> <th>k=3</th> <th>k=4</th> <th>k=5</th> </tr> </thead> <tbody> <tr> <td>HumanEval</td> <td>1.83</td> <td>2.43</td> <td>2.90</td> <td>3.23</td> <td>3.42</td> </tr> <tr> <td>math_reasoning</td> <td>1.85</td> <td>2.53</td> <td>3.04</td> <td>3.44</td> <td>3.74</td> </tr> <tr> <td>qa</td> <td>1.77</td> <td>2.30</td> <td>2.67</td> <td>2.90</td> <td>3.11</td> </tr> <tr> <td>question</td> <td>1.80</td> <td>2.37</td> <td>2.78</td> <td>3.12</td> <td>3.31</td> </tr> <tr> <td>rag</td> <td>1.77</td> <td>2.30</td> <td>2.69</td> <td>2.94</td> <td>3.11</td> </tr> <tr> <td>summarization</td> <td>1.70</td> <td>2.15</td> <td>2.42</td> <td>2.59</td> <td>2.69</td> </tr> <tr> <td>translation</td> <td>1.74</td> <td>2.25</td> <td>2.59</td> <td>2.81</td> <td>2.93</td> </tr> </tbody> </table> <details> <summary>Details</summary>

<strong>Configuration</strong>

• Model: Qwen3-8B-Thinking-speculator.eagle3
• temperature: 0.0
• backend: vLLM chat_completions
• rate-type: throughput
• max-seconds per run: 300
• hardware: 8× GPU (tensor parallel 8)
• Benchmark data: RedHatAI/speculator_benchmarks
• vLLM serve: --no-enable-prefix-caching, --max-num-seqs 64, --enforce-eager

<strong>Command</strong>

GUIDELLM__PREFERRED_ROUTE="chat_completions" \
GUIDELLM__MAX_CONCURRENCY=128 \
guidellm benchmark \
  --target "http://localhost:8000/v1" \
  --data "RedHatAI/speculator_benchmarks" \
  --data-args '{"data_files": "HumanEval.jsonl"}' \
  --rate-type throughput \
  --max-seconds 300 \
  --backend-args '{"extra_body": {"chat_completions": {"temperature": 0.0}}}'

</details>