Smaug multi-turn recipe
According to the model card, the model was built with a Smaug recipe for improving real-world multi-turn conversation performance.
Open Source Model Profile · abacusai
Smaug-Llama-3-70B-Instruct
Smaug-Llama-3-70B-Instruct is a 70.55B-parameter Llama text-generation fine-tune from abacusai. According to the model card, it improves multi-turn conversation on Meta-Llama-3-70B-Instruct.
- Publisher
- abacusai
- Task
- text-generation
- Model type
- llama
- License
- llama3
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
Smaug-Llama-3-70B-Instruct is published by abacusai as a Transformers text-generation fine-tune. The captured configuration identifies LlamaForCausalLM with model type llama and about 70.55B Safetensors parameters. According to the model card, it applies a Smaug multi-turn conversation recipe to meta-llama/Meta-Llama-3-70B-Instruct.
Recorded capabilities
Llama 3 70B Instruct lineage
According to the model card, it is fine-tuned from meta-llama/Meta-Llama-3-70B-Instruct and developed by Abacus.AI.
70B-class Llama build
Captured configuration identifies LlamaForCausalLM with 70553706496 Safetensors parameters, or about 70.55B.
Documented Transformers use
The model card documents Transformers usage with the tokenizer chat template and a sampling configuration of temperature 0.6 and top-p 0.9.
Use cases in the source record
- Multi-turn conversational text generation using the documented Llama 3 Instruct prompt format.
- Transformers-based chat experiments using the model card's tokenizer chat-template and sampling example.
Limitations and unknowns
- Comparative claims about MT-Bench and Arena-Hard are publisher statements; no Ethen-measured evaluation results were extracted.
- No context-window value was extracted from this record.
- No training dataset, steps, hyperparameters, or hardware details were extracted.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: abacusai/Smaug-Llama-3-70B-Instruct
Captured: Unknown. Processed: 2026-09-07T19:34:39.058106+00:00.
Smaug-Llama-3-70B-Instruct Built with Meta Llama 3 This model was built using a new Smaug recipe for improving performance on real world multi-turn conversations applied to meta-llama/Meta-Llama-3-70B-Instruct . The model outperforms Llama-3-70B-Instruct substantially, and is on par with GPT-4-Turbo, on MT-Bench (see below). EDIT: Smaug-Llama-3-70B-Instruct is the top open source model on Arena-Hard currently! It is also nearly on par with Claude Opus - see below. We are conducting additional benchmark evaluations and will add those when available. Model Description Developed by: Abacus.AI License: https://llama.meta.com/llama3/lice…
F001F002F003F004F005F006F007F009F010F011F012F013F014F015F016F017F018F019