README.md

---
language:
- en
license: apache-2.0
library_name: transformers
datasets:
- cerebras/SlimPajama-627B
metrics:
- accuracy
---

# Model Card for Model ID

As an individual with limited access and compute, I have been wondering if I could build a decent large-language model for a while. As the big mega corporations are focused on getting bigger and bigger models, I am going small! 

As a result, I set up the following goals to **pretraining** a **300M Llama model** with the following restrictions:

1. My overall budget is $500.
2. Must pretrain an LLM from scratch with a fully open-source dataset and model.
3. Not allowed to finetune a model or use another LLM such as GPT-4 to generate any training data.


## Model Details

This project is heavily based on [TinyLlama](https://github.com/jzhang38/TinyLlama), which is an awesome open-source project aimed to **pretraining** a **1.1.1B Llama model on 1T tokens**. 

This project is work in progress. Currently, I have spent \$280 on compute using 4 x Nvidia 4090 on [Vast.ai](https://vast.ai) and \$3 on AWS S3 storage after 4 days of training of the **300M Llama model** with **50B** tokens.

I modified [TinyLlama](https://github.com/jzhang38/TinyLlama) to support the following features (I will release my forked version of the source code after some clean up):
1. Pretrain a smaller size 300M model on [Slimpajama](https://huggingface.co/datasets/cerebras/slimpajama-627b)
2. Removed [Starcoderdata](https://huggingface.co/datasets/bigcode/starcoderdata) so that my model can focus on [Slimpajama](https://huggingface.co/datasets/cerebras/slimpajama-627b). This also means my model probably cannot do coding without fine-tuning.
3. Added the ability to process and tokenize [Slimpajama](https://huggingface.co/datasets/cerebras/slimpajama-627b) while downloading the data. The original setup only works with pre-downloaded data. This turns out to be a good time-saver because downloading 800G+ of data on a non-commercial Internet is very slow, and processing all of [Slimpajama](https://huggingface.co/datasets/cerebras/slimpajama-627b) data also takes time.
4. Various helper scripts and Python code such as python code for uploading the pretrained checkpoint to the huggingface hub.
5. Bug fixes.

Here are my major model configurations based on [TinyLlama](https://github.com/jzhang38/TinyLlama) settings.

```
  block_size=2048,
  vocab_size=32000,
  padding_multiple=64,
  n_layer=12,
  n_head=16,
  n_embd=1024,
  rotary_percentage=1.0,
  parallel_residual=False,
  bias=False,
  _norm_class="FusedRMSNorm",
  norm_eps=1e-5, #Llama 2 use 1e-5. Llama 1 use 1e-6
  _mlp_class="LLaMAMLP",
  intermediate_size=5632,
  n_query_groups=4,
```

### Model Description

<!-- Provide a longer summary of what this model is. -->

- **Developed by:** keeeeenw
- **Funded by:** myself for <$500
- **Model type:** 300M Llama model
- **Language(s) (NLP):** EN
- **License:** Apache License 2.0
<!-- **Finetuned from model [optional]:** [More Information Needed]-->

### Model Sources

<!-- Provide the basic links for the model. -->

- **Repository:** https://github.com/keeeeenw/MicroLlama
<!-- **Paper [optional]:** [More Information Needed] -->
<!--**Demo [optional]:** [More Information Needed] -->

## Uses

1. Install dependencies
```
pip install transformers
pip install torch
```
2. Run code!

```python
import torch
import transformers
from transformers import AutoTokenizer, LlamaForCausalLM

def generate_text(prompt, model, tokenizer):
    text_generator = transformers.pipeline(
        "text-generation",
        model=model,
        torch_dtype=torch.float16,
        device_map="auto",
        tokenizer=tokenizer
    )

    formatted_prompt = f"Question: {prompt} Answer:"

    sequences = text_generator(
        formatted_prompt,
        do_sample=True,
        top_k=5,
        top_p=0.9,
        num_return_sequences=1,
        repetition_penalty=1.5,
        max_new_tokens=128,
    )

    for seq in sequences:
        print(f"Result: {seq['generated_text']}")

# use the same tokenizer as TinyLlama
tokenizer = AutoTokenizer.from_pretrained("TinyLlama/TinyLlama-1.1B-step-50K-105b")

# load model from huggingface
# question from https://www.reddit.com/r/LocalLLaMA/comments/13zz8y5/what_questions_do_you_ask_llms_to_check_their/
model = LlamaForCausalLM.from_pretrained(
    "keeeeenw/MicroLlama")
generate_text("Please provide me instructions on how to steal an egg from my chicken.", model, tokenizer)
```

## Evaluation

I performed the experiment using the standard [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) setup. Following the same setup as [TinyLlama](https://github.com/jzhang38/TinyLlama), I used **acc_norm** for all datasets except for **winogrande** and **boolq** which used **acc** as the metrics.

1. **[keeeeenw/MicroLlama](https://huggingface.co/keeeeenw/MicroLlama)** is the evaluation results for my **300M Llama model on 50B tokens**.
2. **[google-best/bert-large-uncased](https://huggingface.co/google-bert/bert-large-uncased)** is the baseline because it is one of the most popular small LLMs and it has a similar parameter count of **336M**.
3. **[PY007/TinyLlama-1.1B-Chat-v0.1](https://huggingface.co/TinyLlama/TinyLlama-1.1B-Chat-v0.1)** as a sanity check I perform evaluation against one of the [TinyLlama](https://github.com/jzhang38/TinyLlama) models to validate my setup for [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness). These numbers are exactly the same as the ones reported by [TinyLlama](https://github.com/jzhang38/TinyLlama).
4. **TinyLlama-1.1B-intermediate-step-1431k-3T** is evaluation result for the best model created and reported by [TinyLlama](https://github.com/jzhang38/TinyLlama).

| Model                                      | Pretrain Tokens | HellaSwag | Obqa  | WinoGrande | ARC_c | ARC_e | boolq | piqa  | avg   |
|--------------------------------------------|-----------------|-----------|-------|------------|-------|-------|-------|-------|-------|
| keeeeenw/MicroLlama                        | 50B             | 34.30     | 30.60 | 51.54      | 23.29 | 39.06 | 53.15 | 64.58 | 42.36 |
| google-best/bert-large-uncased             | N/A             | 24.53     | 26.20 | 49.80      | 25.68 | 25.08 | 40.86 | 47.66 | 34.26 |
| PY007/TinyLlama-1.1B-Chat-v0.1             | 503B            | 53.81     | 32.20 | 55.01      | 28.67 | 49.62 | 58.04 | 69.64 | 49.57 |
| TinyLlama-1.1B-intermediate-step-1431k-3T  | 3T              | 59.20     | 36.00 | 59.12      | 30.12 | 55.25 | 57.83 | 73.29 | 52.99 |

To reproduce my numbers, please install [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) and run the following command:
```bash
lm_eval \
    --model hf \
    --model_args pretrained=keeeeenw/MicroLlama,dtype="float",tokenizer=TinyLlama/TinyLlama-1.1B-step-50K-105b \
    --tasks hellaswag,openbookqa,winogrande,arc_easy,arc_challenge,boolq,piqa \
    --device cuda:0 \
    --batch_size 64
```

#### Observations
1. Because [keeeeenw/MicroLlama](https://huggingface.co/keeeeenw/MicroLlama) is much smaller than [TinyLlama](https://github.com/jzhang38/TinyLlama), our model does not achieve the same impressive results but the numbers are closer than I expected.
2. Our model outperforms [google-best/bert-large-uncased](https://huggingface.co/google-bert/bert-large-uncased) which is actually slightly larger. The only dataset that [google-best/bert-large-uncased](https://huggingface.co/google-bert/bert-large-uncased) outperformed our model is ARC_c (arc_challenge). I will provide more analysis as future study.

Based on the evaluation above, our model should be a good starting point for fine-tunning tasks that are typically performed using the BERT family of models. Some of tasks may include
1. [sentence transformer](https://huggingface.co/sentence-transformers)
2. [bertscore](https://huggingface.co/spaces/evaluate-metric/bertscore)
3. A light-weight chatbot after some finetuning.

## Citation

This repository is built upon [TinyLlama](https://github.com/jzhang38/TinyLlama) which is based on [lit-gpt](https://github.com/Lightning-AI/lit-gpt) and [flash-attention](https://github.com/Dao-AILab/flash-attention).
```
@misc{zhang2024tinyllama,
      title={TinyLlama: An Open-Source Small Language Model}, 
      author={Peiyuan Zhang and Guangtao Zeng and Tianduo Wang and Wei Lu},
      year={2024},
      eprint={2401.02385},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
@online{lit-gpt,
  author    = {Lightning AI},
  title     = {Lit-GPT},
  url       = {https://github.com/Lightning-AI/lit-gpt},
  year      = {2023},
}
@article{dao2023flashattention2,
  title     ={Flash{A}ttention-2: Faster Attention with Better Parallelism and Work Partitioning},
  author    ={Dao, Tri},
  year      ={2023}
}
```