metadata

datasets:
  - openai/gsm8k
language:
  - en
metrics:
  - accuracy
base_model: meta-llama/Llama-2-7b-hf
inference: true
model_type: llama
pipeline_tag: text-generation

Llama-2-7b-gsm8k

This repo contains a dense Llama 2 7B finetuned for arithmetic reasoning task using the GSM8k 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.

Running the model

# pip install transformers accelerate
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("neuralmagic/Llama-2-7b-gsm8k")
model = AutoModelForCausalLM.from_pretrained("neuralmagic/Llama-2-7b-gsm8k", device_map="auto")

input_text = "Natalia sold clips to 48 of her friends in April, and then she sold half as many clips in May. How many clips did Natalia sell altogether in April and May?"
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-gsm8k
GSM8K	0-shot	35.5%

Model Training Details

This model was obtained by fine-tuning the dense Llama 2 7B on the GSM8k dataset. Fine-tuning was performed for 2 epochs with batch-size of 32, with linearly decaying learning-rate from initial value of 3e-5 and warm-up phase of 20 steps.

Help

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