language:
- en
tags:
- llama2
- llama-2
- llama
- llama2 architecture
- litellama
datasets:
- Redpajama
metrics:
- MMLU
license: mit
LiteLLama: Reduced-Scale, Experimental Versions of Llama
In this series of repos, we present an open-source reproduction of Meta AI's LLaMA and LLaMa 2 large language models. However, with significantly reduced model sizes, the experimental version of llama1_s has 1.8B parameters, and the experimental version of llama2_xs has 460M parameters. ('s' stands for small, while 'xs' denotes extra small).
Dataset and Tokenization
We train our models on part of RedPajama dataset. We use the GPT2Tokenizer to tokenize the text.
Training Details
The model was trained with ~1T tokens (0.98T). num of tokens = stepslengthbatch_size=4996791024192=98240888832≈0.98T.
The training curve is at https://wandb.ai/ahxt/llama2_xs_460M_training_loss/reports/reduced_train_loss-23-09-05-20-25-43---Vmlldzo1MzIwNDUx?accessToken=x2ch3n30jo77p1x8y7q9js4h4d8zpjtz1tzot4xxullyefixp4jwt7au2q37k2q6
Using with HuggingFace Transformers
The experimental checkpoints can be directly loaded by Transformers library. The following code snippet shows how to load the our experimental model and generate text with it.
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
model_path = 'ahxt/LiteLlama-460M-1T'
model = AutoModelForCausalLM.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)
model.eval()
prompt = 'Q: What is the largest bird?\nA:'
input_ids = tokenizer(prompt, return_tensors="pt").input_ids
tokens = model.generate(input_ids, max_length=20)
print( tokenizer.decode(tokens[0].tolist(), skip_special_tokens=True) )
# Q: What is the largest bird?\nA: The largest bird is the bald eagle.
Evaluation
We evaluate our models on the MMLU task.
| Models | #parameters | zero-shot | 5-shot |
|---|---|---|---|
| llama | 7B | 28.46 | 35.05 |
| openllama | 3B | 24.90 | 26.71 |
| TinyLlama-1.1B-step-50K-105b | 1.1B | 19.00 | 26.53 |
| LiteLlama-460M-1T | 0.46B | 21.13 | 26.39 |
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 26.65 |
| ARC (25-shot) | 24.91 |
| HellaSwag (10-shot) | 38.47 |
| MMLU (5-shot) | 26.17 |
| TruthfulQA (0-shot) | 41.59 |
| Winogrande (5-shot) | 49.88 |
| GSM8K (5-shot) | 0.0 |
| DROP (3-shot) | 5.51 |
Contact
This experimental version is developed by: Xiaotian Han from Texas A&M University. The model is released