Q-bert commited on
Commit
7bf75d4
1 Parent(s): e0196a9

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +26 -0
app.py CHANGED
@@ -139,6 +139,30 @@ def gradio_train_stock_model(stock_symbol, start_date, end_date, feature_range_m
139
  )
140
  return result
141
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
142
  iface = gr.Interface(
143
  fn=gradio_train_stock_model,
144
  inputs=[
@@ -152,6 +176,8 @@ iface = gr.Interface(
152
  gr.Slider(minimum=1, maximum=64, step=1, label="Batch Size", value=16),
153
  gr.Slider(minimum=1e-5, maximum=1e-1, step=1e-5, label="Learning Rate", value=2e-4)
154
  ],
 
 
155
  outputs="text",
156
  )
157
 
 
139
  )
140
  return result
141
 
142
+ title = "StockLlama-TrainOnAnyStock"
143
+ description = """
144
+ ## StockLlama
145
+ ![The Logo](https://private-user-images.githubusercontent.com/119312866/361069298-11d12a8f-63b8-42ce-b66c-d77924831e90.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjQ2MDYwMTcsIm5iZiI6MTcyNDYwNTcxNywicGF0aCI6Ii8xMTkzMTI4NjYvMzYxMDY5Mjk4LTExZDEyYThmLTYzYjgtNDJjZS1iNjZjLWQ3NzkyNDgzMWU5MC5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwODI1JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDgyNVQxNzA4MzdaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT05ZmUyOWQ4Nzc5YjU0YmZlNGYyMjRmZGY4OWRhYTk5MWZjZGRkMGIzZDQ1YjAwZmQwM2YyY2RkNTcyZmE2ZjgwJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.Oz-_THt_8gGhVod5cCURKaeepzvTGXqGeLi_MkRm09g)
146
+ ### Description
147
+ StockLlama is a time series forecasting pre-trained model based on Llama, enhanced with custom embeddings for improved accuracy.
148
+ ### How It Works
149
+ **Data Collection:** The model retrieves historical stock price data using the yfinance library. Users specify the stock symbol, date range, and other parameters through a Gradio interface.
150
+
151
+ **Data Preprocessing:** The collected stock prices are scaled to a specified range using a custom Scaler class. The data is then divided into sequences of a defined length, with each sequence serving as input to the model and the next data point as the target.
152
+
153
+ **Model Architecture:** StockLlama is a modified version of the Llama model, specifically tailored for time series forecasting. The model is enhanced with custom embeddings and fine-tuned using a LoRA (Low-Rank Adaptation) configuration, allowing for efficient training on the specific stock data.
154
+
155
+ **Training Process:** The training is managed using the Hugging Face Trainer class. The model learns to predict the next data point in the sequence, optimizing its weights over multiple epochs. The training process can be monitored via Weights & Biases integration.
156
+
157
+ **Deployment:** After training, the model is pushed to the Hugging Face Hub, making it accessible for future predictions. The scaler used for data normalization is also saved and uploaded, ensuring that new data can be correctly transformed and predictions can be accurately descaled.
158
+
159
+ ### Contributing
160
+ Contributions to this project are welcome! If you find any issues or want to add new features, feel free to open an issue or submit a pull request.
161
+ ### License
162
+ This project is licensed under the [Apache 2.0 License](https://opensource.org/license/apache-2-0).
163
+ ### Credits
164
+ The StockLlama model used in this project is based on the work by [Talha Rüzgar Akkuş](https://www.linkedin.com/in/talha-r%C3%BCzgar-akku%C5%9F-1b5457264/).
165
+ """
166
  iface = gr.Interface(
167
  fn=gradio_train_stock_model,
168
  inputs=[
 
176
  gr.Slider(minimum=1, maximum=64, step=1, label="Batch Size", value=16),
177
  gr.Slider(minimum=1e-5, maximum=1e-1, step=1e-5, label="Learning Rate", value=2e-4)
178
  ],
179
+ description=description,
180
+ title=title,
181
  outputs="text",
182
  )
183