Text Generation Model

Llama 3 70B Instruct

Best-in-class 70B model with enhanced reasoning, code generation, and instruction following.

Start Building with Llama 3 70B Instruct View All Models

Publisher

70B instruction-tuned model from Meta

Llama 3 70B Instruct is a 70-billion parameter large language model developed by Meta and released in April 2024. It is the instruction-tuned variant of the Llama 3 70B base model, fine-tuned to follow user instructions, engage in dialogue, and complete tasks described in natural language. The model was trained with a context window of 8,000 tokens and has a training data cutoff of December 2023.

Compared to its predecessor Llama 2, Llama 3 70B Instruct incorporates improvements in both pretraining and post-training procedures, resulting in reduced false refusal rates and more diverse responses. The model demonstrates notable capabilities in reasoning, code generation, and instruction following. It is well-suited for tasks such as summarization, question answering, coding assistance, and general-purpose text generation in developer and enterprise applications.

Capabilities

What Llama 3 70B Instruct supports

Instruction Following

Follows natural language instructions to complete tasks, with post-training improvements that reduced false refusals and improved response diversity.

Code Generation

Generates and explains code across common programming languages, a capability explicitly improved in Llama 3's training process.

Reasoning

Handles multi-step reasoning tasks including logical inference and problem decomposition, with reasoning cited as a key improvement over Llama 2.

Text Summarization

Condenses long-form text into concise summaries within the model's 8,000-token context window.

Dialogue & Chat

Supports multi-turn conversational interactions, tuned for alignment and steerability through instruction fine-tuning.

Multilingual Text

Processes and generates text in multiple languages, reflecting broad multilingual coverage in Llama 3's pretraining data.

Ready to build with Llama 3 70B Instruct?

Get Started Free

FAQ

Common questions about Llama 3 70B Instruct

What is the context window for Llama 3 70B Instruct?

The model supports a context window of 8,000 tokens, meaning the combined length of your input and output cannot exceed that limit in a single request.

What is the knowledge cutoff date for this model?

The training data for Llama 3 70B Instruct has a cutoff of December 2023, so the model does not have knowledge of events that occurred after that date.

Who developed Llama 3 70B Instruct?

Llama 3 70B Instruct was developed and released by Meta. It is the instruction-tuned version of the Llama 3 70B base model.

What is the difference between the base model and the instruct variant?

The base model (Meta-Llama-3-70B) is a pretrained model suited for text completion tasks, while the instruct variant has been fine-tuned to follow natural language instructions and is better suited for dialogue and task-oriented use cases.

Is Llama 3 70B Instruct open source?

Meta released Llama 3 under a custom community license. The model weights are publicly available on Hugging Face, but usage is subject to Meta's Llama 3 license terms, which include restrictions for certain commercial use cases above a defined user threshold.

Resources