Llama 3 Euryale 70B v2.1

Llama 3 Euryale 70B v2.1 (L3-70B-Euryale-v2.1) is a community fine-tune published by the author Sao10K on Hugging Face at Sao10K/L3-70B-Euryale-v2.1. The author calls it "Stheno's Sister Model, designed to impress". It is a roleplay and creative-writing model of about 71B parameters in BF16 Safetensors format, using the Llama 3 Instruct prompt format. It is not a Meta release, and the repository is still online as of October 29, 2026.

Key Features

• LoRA fine-tune: the card says it is a LoRA rather than a full fine-tune because a full fine-tune was too costly. Training used 8x H100 SXM GPUs, followed by extra training, on the same dataset as Stheno v3.2.
• Author claims: the author reports better prompt adherence, better anatomy and spatial awareness, and stronger handling of custom formats, and describes it as very creative and unrestrictive in roleplay. These are the author's own statements, not independent measurements.
• Recommended sampling: temperature 1.17, min_p 0.075, repetition penalty 1.10, with the Llama-3-Instruct-Names context template in SillyTavern.
• License: CC-BY-NC-4.0, which does not allow commercial use.
• Availability: usable with Transformers, vLLM, SGLang and Docker Model Runner, with quantized versions for llama.cpp, Ollama and LM Studio listed on the card.

Use Cases

• Interactive fiction and character roleplay in tools such as SillyTavern.
• Creative-writing experiments with a large open model.
• Research on community fine-tunes of Llama 3.

Limitation: the card states no context length and no benchmark scores, and the non-commercial license rules out most business uses.

Pricing and availability

The weights are free to download. The card lists Featherless AI as an inference provider; this entry quotes no hosted price because none was confirmed on a first-party page.

Getting Started

1. Download the model from Hugging Face, or a quantized build for your runtime.
2. Use the Llama-3-Instruct prompt format and the sampling settings above as a starting point.
3. Check the CC-BY-NC-4.0 terms before using outputs or the model commercially.

FAQ

Can I use Euryale 70B v2.1 commercially?

The card lists CC-BY-NC-4.0, a non-commercial license. Also review the terms of the Llama 3 base model.

What is Euryale designed for?

Roleplay and creative writing, according to the author.

Is there a newer Euryale?

The catalog also lists Llama 3.1 Euryale 70B v2.2, another Sao10K release.

What is its context length?

The model card does not state one.

Alternatives

• Llama 3.1 Euryale 70B v2.2: the later Euryale release in this catalog.
• Llama 3.1 70B Instruct: Meta's own 70B instruct model.
• Hermes 3 405B Instruct: Nous Research's larger Llama 3.1 fine-tune.

Conclusion

Euryale 70B v2.1 is a niche, creativity-focused fine-tune, free to run but restricted to non-commercial use. For general or commercial work, start from Meta's official models. See more in the models category.