---
base_model: neuralmagic/Llama-2-7b-pruned70-retrained
inference: true
model_type: llama
pipeline_tag: text-generation
datasets:
  - garage-bAInd/Open-Platypus
  - Open-Orca/OpenOrca
  - cognitivecomputations/dolphin
tags:
- sparse
- instruct
---

# Llama-2-7b-pruned70-retrained-instruct

This repo contains a [70% sparse Llama 2 7B](https://huggingface.co/neuralmagic/Llama-2-7b-pruned70-retrained) finetuned for instruction-following tasks using a blend of the Platypus + Open Orca + Dolphin datasets.

**Authors**: Neural Magic, Cerebras

## Usage

Below we share some code snippets on how to get quickly started with running the model.

### Sparse Transfer

By leveraging a pre-sparsified model's structure, you can efficiently fine-tune on new data, leading to reduced hyperparameter tuning, training times, and computational costs. Learn about this process [here](https://neuralmagic.github.io/docs-v2/get-started/transfer).

### Running the model

This model may be run with the transformers library. For accelerated inference with sparsity, deploy with [nm-vllm](https://github.com/neuralmagic/nm-vllm) or [deepsparse](https://github.com/neuralmagic/deepsparse).

```python
# pip install transformers accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Llama-2-7b-pruned70-retrained-instruct")
model = AutoModelForCausalLM.from_pretrained("Llama-2-7b-pruned70-retrained-instruct", device_map="auto")

input_text = "Write a recipe for banana bread:\n"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))
```

## Evaluation Benchmark Results

Model evaluation metrics and results.

| Benchmark                                      | Metric        | Llama-2-7b-instruct  | Llama-2-7b-pruned70-retrained-instruct |
|------------------------------------------------|---------------|-------------|-------------------------------|
| [MMLU](https://arxiv.org/abs/2009.03300)       | 5-shot        | 48.60%        | 42.33%                        |
| [HellaSwag](https://arxiv.org/abs/1905.07830)  | 10-shot       | 79.45%        | 77.21%                        |
| [WinoGrande](https://arxiv.org/abs/1907.10641) | 5-shot        | 75.69%        | 71.90%                        |
| [ARC-c](https://arxiv.org/abs/1911.01547)      | 25-shot       | 53.92%        | 47.35%                        |
| [TruthfulQA](https://arxiv.org/abs/2109.07958) | 0-shot        | 43.63%        | 42.25%                        |
| [GSM8K](https://arxiv.org/abs/2110.14168)      | 5-shot        | 15.92%        | 14.25%                        |

## Model Training Details

This model was obtained by sparse-tranfer of the sparse foundational model [Llama-2-7b-pruned50-retrained](https://huggingface.co/neuralmagic/Llama-2-7b-pruned70-retrained) on a blend of [Open Platypus](https://huggingface.co/datasets/garage-bAInd/Open-Platypus), 10% [Open Orca](https://huggingface.co/datasets/Open-Orca/OpenOrca) and 10% [Dolphin](https://huggingface.co/datasets/cognitivecomputations/dolphin) datasets. Training was perfomerd for 6 epochs.

## Help

For further support, and discussions on these models and AI in general, join [Neural Magic's Slack Community](https://join.slack.com/t/discuss-neuralmagic/shared_invite/zt-q1a1cnvo-YBoICSIw3L1dmQpjBeDurQ)