Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,73 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- alxxtexxr/indowebgen-dataset
|
| 5 |
+
language:
|
| 6 |
+
- id
|
| 7 |
+
---
|
| 8 |
+
# 🇮🇩🌐🤖 IndoWebGen: LLM for Automated (Bootstrap-Based) Website Generation Based-On Indonesian Instructions
|
| 9 |
+
Hugely inspired by [Web App Factory](https://huggingface.co/spaces/jbilcke-hf/webapp-factory-wizardcoder).
|
| 10 |
+
## Model Description:
|
| 11 |
+
- Base Model: [codellama/CodeLlama-7b-hf](https://huggingface.co/codellama/CodeLlama-7b-hf) [[1](https://arxiv.org/abs/2308.12950)]
|
| 12 |
+
- Finetuning Method: LoRA [[2](https://arxiv.org/abs/2106.09685)]
|
| 13 |
+
- Dataset: [alxxtexxr/indowebgen-dataset](https://huggingface.co/datasets/alxxtexxr/indowebgen-dataset)
|
| 14 |
+
|
| 15 |
+
## Finetuning Hyperparameters:
|
| 16 |
+
- Number of Epochs: 20
|
| 17 |
+
- Microbatch Size: 4
|
| 18 |
+
- Gradient Accumulation Step: 8
|
| 19 |
+
- LoRA Rank: 16
|
| 20 |
+
- LoRA Alpha: 32
|
| 21 |
+
|
| 22 |
+
## Inference:
|
| 23 |
+
Try the inference demo [here](https://indowebgen.alimtegar.my.id) or try running the inference code with the provided Google Colab notebook [here](https://colab.research.google.com/drive/1pqqLGcgRcUTBLCNeF0V6REi7INJ43IZb?usp=sharing). The inference code used is shown below:
|
| 24 |
+
```
|
| 25 |
+
# Install the required libraries
|
| 26 |
+
!pip install transformers bitsandbytes accelerate
|
| 27 |
+
|
| 28 |
+
# Import the neccessary modules
|
| 29 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 30 |
+
|
| 31 |
+
# Load the model and the tokenizer
|
| 32 |
+
model_id = 'alxxtexxr/indowebgen-7b'
|
| 33 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 34 |
+
model_id,
|
| 35 |
+
load_in_8bit=True,
|
| 36 |
+
# load_in_4bit=True, # for low memory
|
| 37 |
+
device_map='auto',
|
| 38 |
+
)
|
| 39 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
| 40 |
+
|
| 41 |
+
# Initialize the prompt
|
| 42 |
+
prompt_template = '''Berikut adalah instruksi pembuatan website beserta output-nya yang berupa kode HTML dari website yang dibuat:
|
| 43 |
+
|
| 44 |
+
### Instruksi:
|
| 45 |
+
{instruction}
|
| 46 |
+
|
| 47 |
+
### Output:
|
| 48 |
+
<html lang="id">'''
|
| 49 |
+
|
| 50 |
+
# INSERT YOUR OWN INDONESIAN INSTRUCTION BELOW
|
| 51 |
+
instruction = 'Buatlah website portfolio untuk Budi'
|
| 52 |
+
|
| 53 |
+
prompt = prompt_template.format(instruction=instruction)
|
| 54 |
+
|
| 55 |
+
# Generate the output
|
| 56 |
+
input_ids = tokenizer(prompt, return_tensors='pt').input_ids.to(model.device)
|
| 57 |
+
outputs = model.generate(
|
| 58 |
+
input_ids,
|
| 59 |
+
max_new_tokens=2400,
|
| 60 |
+
do_sample=True,
|
| 61 |
+
temperature=1.0,
|
| 62 |
+
top_k=3,
|
| 63 |
+
top_p=0.8,
|
| 64 |
+
repetition_penalty=1.1,
|
| 65 |
+
pad_token_id=tokenizer.unk_token_id,
|
| 66 |
+
)
|
| 67 |
+
print(tokenizer.batch_decode(outputs, skip_special_tokens=True)[0])
|
| 68 |
+
```
|
| 69 |
+
|
| 70 |
+
## Limitations
|
| 71 |
+
- The dataset used in training is limited to only 500 data, so the model performance may still not be optimal.
|
| 72 |
+
- The generated websites leverage Bootstrap for the styling and Font Awesome for the icons.
|
| 73 |
+
- The content of the generated websites is dummy (including the images), so the users need to further customize the websites.
|