Llama 2 70B
Meta
Meta's largest open-weights Llama 2 model with 70B parameters. Released July 2023, trained on 2 trillion tokens with 40% more data than Llama 1. Features 4K context length (expanded from 2K in Llama 1) and Grouped-Query Attention for improved inference scalability. Supports both research and commercial use with permissive license. Pretrained on data through September 2022 with tuning data through July 2023. Global batch size of 4M tokens during training. Fine-tuned chat variant optimized for dialogue using supervised fine-tuning and RLHF. Widely adopted as foundation for custom models and applications. Superseded by Llama 3.1 70B (July 2024).
Strengths
- Largest Llama 2 variant with strongest capabilities in the family
- Permissive license enables both research and commercial use
- Trained on 2T tokens with 40% more data than Llama 1
- Grouped-Query Attention improves inference efficiency
- Widely adopted as foundation for fine-tuning and custom models
- Strong ecosystem support and tooling
- Chat variant optimized via SFT and RLHF for dialogue
Caveats
- 4K context window extremely small vs modern models (Llama 3.1: 128K)
- Superseded by Llama 3.1 70B with major improvements (July 2024)
- September 2022 pretraining cutoff very outdated
- 70B size requires significant GPU resources (140GB+ VRAM)
- No tool use, vision, or multimodal capabilities
- Consider Llama 3.1 70B or Llama 3.3 70B for new projects
- Vendor
- Meta
- Context Window
- 4,096 tokens
- Max Output
- 4,096 tokens
- License
- Llama 2 Community License
- Release Date
- 2023-07-18
- Modalities
- text
Capabilities
Pricing
Open weights - free to download. Requires significant GPU resources. API providers typically $0.70-$0.90 per 1M input tokens