---
license: mit
task_categories:
- text-generation
language:
- en
tags:
- llama-2
- benchmark
- evaluation
- mt-bench
- alpaca-eval
---

# Llama 2 13B Model Outputs

This repository contains all concatenated text outputs from Llama 2 13B models (base and chat) for MT-Bench and AlpacaEval benchmarks.

## Quick Download

Download the output files directly:

```bash
# Base model outputs (885 completions: 80 MT-Bench + 805 AlpacaEval)
wget https://huggingface.co/datasets/jeqcho/llama-2-13b-outputs/resolve/main/llama-2-13b-base.txt

# Chat model outputs (885 completions: 80 MT-Bench + 805 AlpacaEval)
wget https://huggingface.co/datasets/jeqcho/llama-2-13b-outputs/resolve/main/llama-2-13b-chat.txt
```

Or with curl:

```bash
curl -L -o llama-2-13b-base.txt https://huggingface.co/datasets/jeqcho/llama-2-13b-outputs/resolve/main/llama-2-13b-base.txt
curl -L -o llama-2-13b-chat.txt https://huggingface.co/datasets/jeqcho/llama-2-13b-outputs/resolve/main/llama-2-13b-chat.txt
```

## Files

- **`llama-2-13b-base.txt`**: All outputs from Llama 2 13B base model
  - 80 MT-Bench first-turn completions
  - 805 AlpacaEval completions
  - Total: 885 outputs concatenated

- **`llama-2-13b-chat.txt`**: All outputs from Llama 2 13B chat model
  - 80 MT-Bench first-turn completions
  - 805 AlpacaEval completions
  - Total: 885 outputs concatenated

## Structured Data

For structured JSONL/JSON outputs with metadata, see the separate repositories:

- [jeqcho/llama-2-13b-hf-mt-bench](https://huggingface.co/datasets/jeqcho/llama-2-13b-hf-mt-bench) - Base MT-Bench
- [jeqcho/llama-2-13b-hf-alpaca-eval](https://huggingface.co/datasets/jeqcho/llama-2-13b-hf-alpaca-eval) - Base AlpacaEval
- [jeqcho/llama-2-13b-chat-hf-mt-bench](https://huggingface.co/datasets/jeqcho/llama-2-13b-chat-hf-mt-bench) - Chat MT-Bench
- [jeqcho/llama-2-13b-chat-hf-alpaca-eval](https://huggingface.co/datasets/jeqcho/llama-2-13b-chat-hf-alpaca-eval) - Chat AlpacaEval

## Models Used

- **Base**: [meta-llama/Llama-2-13b-hf](https://huggingface.co/meta-llama/Llama-2-13b-hf)
- **Chat**: [meta-llama/Llama-2-13b-chat-hf](https://huggingface.co/meta-llama/Llama-2-13b-chat-hf)

## Generation Details

- **Hardware**: NVIDIA L40S (46GB VRAM)
- **Batch Size**: 8
- **Temperature**: 0.7
- **Top-p**: 0.9
- **Max New Tokens**: 512

## Citation

```bibtex
@misc{touvron2023llama,
    title={Llama 2: Open Foundation and Fine-Tuned Chat Models},
    author={Hugo Touvron et al.},
    year={2023},
    eprint={2307.09288},
    archivePrefix={arXiv}
}
```