Model Card for Llama-2-7b-alpaca-cleaned
This model checkpoint is the Llama-2-7b fine-tuned on alpaca-cleaned dataset with the original Alpaca fine-tuning hyper-parameters.
Model Details
Model Description
This model checkpoint is the Llama-2-7b fine-tuned on alpaca-cleaned dataset with the original Alpaca fine-tuning hyper-parameters.
The original Alpaca model is fine-tuned on Llama with the alpaca dataset by researchers from Stanford University
- Developed by: NEU Human-centered AI Lab
- Shared by [optional]: NEU Human-centered AI Lab
- Model type: Text-generation
- Language(s) (NLP): English
- License: cc-by-nc-4.0 (comply with the alpaca-cleaned dataset)
- Finetuned from model [optional]: Llama-2-7b
Model Sources
Uses
Direct Use
The model is intended to be used for research purposes only in English, complying with stanford_alpaca project.
The model has been fine-tuned on the alpaca-cleaned dataset for assistant-like chat and general natural language generation tasks.
The use of this model should also comply with the restrictions from Llama-2-7b.
Out-of-Scope Use
The out-of-Scope use of this model should also comply with stanford_alpaca project and Llama-2-7b.
Bias, Risks, and Limitations
{{ bias_risks_limitations | default("[More Information Needed]", true)}}
How to Get Started with the Model
Use the code below to get started with the model.
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("NEU-HAI/Llama-2-7b-alpaca-cleaned")
model = AutoModelForCausalLM.from_pretrained("NEU-HAI/Llama-2-7b-alpaca-cleaned")
Training Details
Training Data
We use the alpaca-cleaned dataset, which is the cleaned version of the original alpaca dataset created by researchers from Stanford University.
Training Procedure
We follow the same training procedure and mostly same hyper-parameters to fine-tune the original Alpaca model on Llama. The procedure can be found in stanford_alpaca project.
Training Hyperparameters
--bf16 True \
--num_train_epochs 3 \
--per_device_train_batch_size 4 \
--per_device_eval_batch_size 4 \
--gradient_accumulation_steps 8 \
--evaluation_strategy "no" \
--save_strategy "steps" \
--save_steps 2000 \
--save_total_limit 1 \
--learning_rate 2e-5 \
--weight_decay 0. \
--warmup_ratio 0.03 \
--lr_scheduler_type "cosine" \
--logging_steps 1 \
--fsdp "full_shard auto_wrap" \
--fsdp_transformer_layer_cls_to_wrap 'LlamaDecoderLayer' \
--tf32 True
Evaluation
Testing Data, Factors & Metrics
Testing Data
N/A
Factors
N/A
Metrics
N/A
Results
N/A
Summary
N/A
Citation
Please cite the stanford_alpaca project
@misc{alpaca,
author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
title = {Stanford Alpaca: An Instruction-following LLaMA model},
year = {2023},
publisher = {GitHub},
journal = {GitHub repository},
howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
}
Model Card Authors
Northeastern Human-centered AI Lab
Model Card Contact
- Downloads last month
- 909