neuralmagic/Llama-2-7b-ultrachat200k-pruned_70-quantized-deepsparse

Llama-2-7b-pruned70-retrained-ultrachat-quant-ds

This repo contains a 70% sparse Llama 2 7B finetuned for chat tasks using the UltraChat 200k dataset. It was then quantized to 8-bit weights + activations and exported to deploy with DeepSparse, a CPU inference runtime for sparse models.

Official model weights from Enabling High-Sparsity Foundational Llama Models with Efficient Pretraining and Deployment.

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.

Running the model

For accelerated inference with sparsity on CPUs, deploy with deepsparse.

# pip install deepsparse[llm]
from deepsparse import TextGeneration

model = TextGeneration(model_path="hf:neuralmagic/Llama-2-7b-pruned70-retrained-ultrachat-quant-ds")

input_text = "Write me a poem about Machine Learning."
outputs = model(input_text, max_new_tokens=100)
print(outputs.generations[0].text)

Evaluation Benchmark Results

Model evaluation metrics and results.

Benchmark	Metric	Llama-2-7b-ultrachat	Llama-2-7b-pruned70-retrained-ultrachat-quant-ds
AlpacaEval (Llama-2-70b-chat-hf evaluator)	Win rate	57.6%	57.1%

Help

For further support, and discussions on these models and AI in general, join Neural Magic's Slack Community