Llama-3.1-8B-Instruct-speculator.eagle3

Model Overview

• Verifier: meta-llama/Llama-3.1-8B-Instruct
• 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 meta-llama/Llama-3.1-8B-Instruct, 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. This model should be used with the meta-llama/Llama-3.1-8B-Instruct chat template, specifically through the /chat/completions endpoint.

Use with vLLM

vllm serve meta-llama/Llama-3.1-8B-Instruct \
  -tp 1 \
  --speculative-config '{
    "model": "RedHatAI/Llama-3.1-8B-Instruct-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	1.85	2.55	3.07	3.43	3.69	3.85	3.96/td>
Math Reasoning	1.83	2.47	2.92	3.23	3.43	3.58	3.65
Text Summarization	1.74	2.27	2.61	2.86	3.00	3.10	3.17

Performance benchmarking (1xA100)

Details

Configuration

• temperature: 0
• repetitions: 5
• time per experiment: 3min
• 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/SpeculativeDecoding" \
  --rate-type sweep \
  --max-seconds 180 \
  --output-path "Llama-3.1-8B-Instruct-HumanEval.json" \
  --backend-args '{"extra_body": {"chat_completions": {"temperature":0.0}}}'
</details>