Skip to content
Other

Mistral-7B-Instruct-v0.3 EAGLE3 draft

a draft checkpoint for speculative decoding

2 min readupdated

This is a specialized EAGLE-3 draft model built to speed up text generation for mistralai/Mistral-7B-Instruct-v0.3. It uses speculative decoding—where a lightweight auxiliary model quickly drafts several tokens ahead for the main model to verify in a single pass—boosting throughput without sacrificing quality or altering the output distribution. Trained on AWS Trainium2 across the UltraChat, Magicoder, and GSM8K datasets, the checkpoint has 293.6M trainable parameters and is optimized for the speculators format. It is a solid drop-in option for developers and engineers running Mistral-7B services who want lower inference latency.

Key features

  • EAGLE-3 architecture: speculative decoding that verifies every drafted token against the target model while preserving its exact output distribution.
  • Faster inference: cuts per-token latency by 1.5–2.1x under low-to-moderate server load.
  • Curated training data: trained on 61.2M tokens generated directly by Mistral-7B-Instruct-v0.3 using AWS Trainium2.
  • vLLM compatibility: works out of the box with standard speculative decoding libraries.

What you can do

  • Speed up local or cloud deployments: plug the checkpoint into a vLLM pipeline alongside mistralai/Mistral-7B-Instruct-v0.3 to cut response latency.
  • Benchmark on coding and math: maximize token acceptance lengths on programming tasks (Magicoder) and multi-step reasoning (GSM8K).
  • Deploy on AWS Trainium2: run the model using the vllm-neuron framework with tensor parallelism.

Example

Running speculative decoding with vLLM:

vllm serve mistralai/Mistral-7B-Instruct-v0.3 \
  --speculative-config '{"method": "eagle3", "model": "", "num_speculative_tokens": 3}'
Read moreDetails
Benchmarks: mean acceptance length reaches 2.70 on UltraChat, 3.25 on Magicoder, and 3.16 on GSM8K (at k=3).
Hardware: trained on a single trn2.3xlarge node (AWS Trainium2).
Limitations: reported speedups apply to greedy decoding; efficiency drops when sampling with temperature > 0. Optimized primarily for English, code, and math.
Discuss Telegram 0 7

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.0 GB · ready files1.7 GB
  • FP8 / INT8recommendedweights 0.5 GB1.1 GB
  • 4-bitweights 0.3 GB0.8 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # jburtoft/Mistral-7B-Instruct-v0.3-EAGLE3
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download jburtoft/Mistral-7B-Instruct-v0.3-EAGLE3 --local-dir ~/models/Mistral-7B-Instruct-v0.3-EAGLE3
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

  1. Mistral-7B-v0.3family root
  2. Mistral-7B-Instruct-v0.3
  3. Mistral-7B-Instruct-v0.3-EAGLE3

New posts about versions of Mistral-7B-v0.3 (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 12All on Hugging Face
Merges 3All on Hugging Face

About the model

jburtoft/Mistral-7B-Instruct-v0.3-EAGLE3
Author
jburtoft
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

70

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments