Llama-3.2-3B-Instruct-Eagle3

EAGLE3 Draft Head — Llama-3.2-3B-Instruct

A speculative decoding draft head for meta-llama/Llama-3.2-3B-Instruct, trained using the EAGLE3 method on Google Cloud TPU with the SpecJAX framework.

EAGLE3 draft heads accelerate autoregressive generation by proposing multiple tokens per step that a target model then verifies in parallel — typically achieving 2-3x throughput gains with no change in output quality.

Usage

SGLang (GPU)

Llama-3.2 EAGLE3 is natively supported in SGLang.

python -m sglang.launch_server \
    --model meta-llama/Llama-3.2-3B-Instruct \
    --speculative-algorithm EAGLE3 \
    --speculative-draft-model-path thoughtworks/Llama-3.2-3B-Instruct-Eagle3 \
    --speculative-num-steps 5 \
    --speculative-eagle-topk 4 \
    --dtype bfloat16

sglang-jax (TPU)

Llama-3.2 EAGLE3 is natively supported in sglang-jax. Note: sglang-jax's EAGLE3 pipeline is functional but not yet performance-optimized.

python -m sgl_jax.launch_server \
    --model-path meta-llama/Llama-3.2-3B-Instruct \
    --speculative-algorithm EAGLE3 \
    --speculative-draft-model-path thoughtworks/Llama-3.2-3B-Instruct-Eagle3 \
    --speculative-eagle-topk 1 \
    --speculative-num-steps 3 \
    --speculative-num-draft-tokens 4 \
    --tp-size 4 --dtype bfloat16

Python (SGLang client)

import sglang as sgl

llm = sgl.LLM(
    model="meta-llama/Llama-3.2-3B-Instruct",
    speculative_algorithm="EAGLE3",
    speculative_draft_model_path="thoughtworks/Llama-3.2-3B-Instruct-Eagle3",
    speculative_num_steps=5,
    speculative_eagle_topk=4,
    dtype="bfloat16",
)

Note: Access to meta-llama/Llama-3.2-3B-Instruct requires accepting Meta's Llama 3.2 Community License on the model page.

Training Details

Training Method

This model uses EAGLE3's Test-Time Training (TTT) objective with a rollout length of 7. At each training step, the draft head autoregressively proposes 7 tokens; the target model provides ground-truth hidden states and logits for all positions; a geometric loss (0.8^k weighting) trains the draft to match the target at each position.

Performance

Token acceptance rates on generic instruction-following data (ShareGPT-style prompts):

Measured on held-out evaluation data. Actual throughput gains depend on hardware, prompt distribution, and runtime version.

Model Architecture

The draft head is a single-layer transformer that operates on the target model's hidden states:

Limitations

• Trained on English-dominant instruction data; performance may degrade on non-English inputs or highly domain-specific content.
• Acceptance rates are measured on generic chat data and will vary by prompt distribution.
• This is a v1 checkpoint trained on generic data. A v2 with target-model-regenerated training data is planned.

License

This model is released under the Meta Llama 3.2 Community License. Usage is subject to Meta's acceptable use policy.

References

@article{li2025eagle3,
  title={EAGLE3: Scalable Speculative Decoding with Training-Free Multi-Draft Speculation},
  author={Li, Yuhui and Wei, Fangyun and Zhang, Chao and Zhang, Hongyang},
  journal={arXiv preprint arXiv:2503.01840},
  year={2025}
}