RichardErkhov/neuralmagic_-_Llama-2-7b-pruned70-retrained-gguf

Quantization made by Richard Erkhov.

Github

Discord

Request more models

Llama-2-7b-pruned70-retrained - GGUF

• Model creator: https://huggingface.co/neuralmagic/
• Original model: https://huggingface.co/neuralmagic/Llama-2-7b-pruned70-retrained/

Name	Quant method	Size
Llama-2-7b-pruned70-retrained.Q2_K.gguf	Q2_K	2.36GB
Llama-2-7b-pruned70-retrained.Q3_K_S.gguf	Q3_K_S	2.75GB
Llama-2-7b-pruned70-retrained.Q3_K.gguf	Q3_K	3.07GB
Llama-2-7b-pruned70-retrained.Q3_K_M.gguf	Q3_K_M	3.07GB
Llama-2-7b-pruned70-retrained.Q3_K_L.gguf	Q3_K_L	3.35GB
Llama-2-7b-pruned70-retrained.IQ4_XS.gguf	IQ4_XS	3.4GB
Llama-2-7b-pruned70-retrained.Q4_0.gguf	Q4_0	3.56GB
Llama-2-7b-pruned70-retrained.IQ4_NL.gguf	IQ4_NL	3.58GB
Llama-2-7b-pruned70-retrained.Q4_K_S.gguf	Q4_K_S	3.59GB
Llama-2-7b-pruned70-retrained.Q4_K.gguf	Q4_K	3.8GB
Llama-2-7b-pruned70-retrained.Q4_K_M.gguf	Q4_K_M	3.8GB
Llama-2-7b-pruned70-retrained.Q4_1.gguf	Q4_1	3.95GB
Llama-2-7b-pruned70-retrained.Q5_0.gguf	Q5_0	4.33GB
Llama-2-7b-pruned70-retrained.Q5_K_S.gguf	Q5_K_S	4.33GB
Llama-2-7b-pruned70-retrained.Q5_K.gguf	Q5_K	4.45GB
Llama-2-7b-pruned70-retrained.Q5_K_M.gguf	Q5_K_M	4.45GB
Llama-2-7b-pruned70-retrained.Q5_1.gguf	Q5_1	4.72GB
Llama-2-7b-pruned70-retrained.Q6_K.gguf	Q6_K	5.15GB
Llama-2-7b-pruned70-retrained.Q8_0.gguf	Q8_0	6.67GB

Original model description:

base_model: neuralmagic/Llama-2-7b-pruned50-retrained inference: true model_type: llama pipeline_tag: text-generation datasets: - cerebras/SlimPajama-627B tags: - sparse

Llama-2-7b-pruned70-retrained

This repo contains model files for a Llama 2 7B model that has had 50% of the parameters pruned in one-shot with SparseGPT, then retrained by Cerebras with 50B tokens from SlimPajama while maintaining sparsity. It was then one-shot pruned to 70% sparsity and trained for another 100B tokens.

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 has not been fine-tuned for instruction-following but 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")
model = AutoModelForCausalLM.from_pretrained("neuralmagic/Llama-2-7b-pruned70-retrained", device_map="auto")

input_text = "Write me a poem about Machine Learning."
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. [UPDATE]

Benchmark	Metric	Llama-2-7b	Llama-2-7b-pruned70-retrained
MMLU	5-shot	46.9%	36.5%
HellaSwag	0-shot	78.6%	74.1%
WinoGrande	5-shot	74.0%	69.5%
ARC-c	25-shot	53.1%	45.4%
TruthfulQA	5-shot	38.8%	36.7%
GSM8K	5-shot	14.5%	8.0%
HumanEval	pass@1	13.4%	14.4%

Model Training Details

[UPDATE]

Help

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