We are thrilled to announce the release of version **1.1.2** of the `shortterm-memory` package. This update focuses on optimizing performance, enhancing memory management, and improving overall efficiency.
---
**Main Updates**
1. **Advanced Memory Management Enhancements**
- **GPU Acceleration:**
Added automatic GPU support using PyTorch, significantly improving summarization performance when a CUDA device is available.
- **Batch Processing for Memory Compression:**
Introduced batch processing (default batch size: 5) for more efficient handling of conversation history during summarization, reducing latency.
- **Optimized Memory Counter:**
Enhanced the memory counter to use token-based counting instead of word-based, improving accuracy and performance.
---
2. **Efficient Conversation History Handling**
- **Optimized Update Memory Method:**
The `update_memory` method has been refactored for better performance, including dynamic compression and automatic trimming of older conversations.
- **Dynamic Compression Threshold:**
Compression is now triggered based on the token count rather than word count, providing more precise control over memory usage.
---
3. **Integration with Transformers**
- **Enhanced Summarization:**
Improved summarization using BART with optimized parameters for faster and more concise summaries.
- **Efficient Tokenization:**
Direct use of tokenization for both counting and summarization, reducing redundant operations.
---
4. **Error Handling and Logging**
- **Improved Logging:**
Enhanced logging features to provide detailed insights into memory operations, including compression and trimming events.
- **Error Handling:**
Added robust error handling to manage issues during summarization and memory updates gracefully.
---
**Usage Example**
python
import torch
from transformers import pipeline
import logging
from shortterm_memory.ChatbotMemory import ChatbotMemory
Initialize Chatbot Memory
chat_memory = ChatbotMemory()
Example of updating the memory with user input and bot response
user_input = "Hello, how are you?"
bot_response = "I'm fine, thank you! How about you?"
chat_memory.update_memory(user_input, bot_response)
Retrieve the updated conversation history
history = chat_memory.get_memory()
print(history)
**How to Upgrade**
To upgrade to this latest version, run the following command:
bash
pip install --upgrade shortterm-memory