alexmarques's picture
Update README.md
b7105bf verified
metadata
base_model: neuralmagic/Llama-2-7b-pruned70-retrained
inference: true
model_type: llama
pipeline_tag: text-generation
datasets:
  - cerebras/SlimPajama-627B
  - HuggingFaceH4/ultrachat_200k
tags:
  - sparse
  - chat

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

This repo contains a 70% sparse Llama 2 7B finetuned for chat tasks using the UltraChat 200k dataset.

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

This model may be run with the transformers library. For accelerated inference with sparsity, deploy with nm-vllm or deepsparse.

# pip install transformers accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM

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

input_text = "Write me a poem about Machine Learning."
input_ids = tokenizer.apply_chat_template(input_text, add_generation_prompt=True, 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-ultrachat Llama-2-7b-pruned70-retrained-ultrachat
MMLU 5-shot 46.1% 32.5%
HellaSwag 0-shot 75.9% 68.9%
WinoGrande 5-shot 72.6% 65.1%
ARC-c 25-shot 52.8% 45.3%
TruthfulQA 5-shot 44.8% 39.6%
GSM8K 5-shot 12.4% 4.8%
AlpacaEval (Llama-2-70b-chat-hf evaluator) Win rate 57.6% 57.4%
AlpacaEval (GPT-4 Turbo evaluator) Win rate 60.6% 54.0%

Model Training Details

This model was obtained by sparse-tranfer of the sparse foundational model Llama-2-7b-pruned70-retrained on the ultrachat_200k dataset. Training was performed for 2 epochs and used the SquareHead knowledge distillation with Llama-2-7b-ultrachat as teacher.

Help

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