Added new demo

autonomous_rag/README.md

@@ -1,4 +1,4 @@
-### 🤖 AutoRAG: Autonomous RAG with GPT-4o and Vector Database
+## 🤖 AutoRAG: Autonomous RAG with GPT-4o and Vector Database
 This Streamlit application implements an Autonomous Retrieval-Augmented Generation (RAG) system using OpenAI's GPT-4o model and PgVector database. It allows users to upload PDF documents, add them to a knowledge base, and query the AI assistant with context from both the knowledge base and web searches.
 Features
 

chat_with_gmail/README.md

@@ -2,13 +2,13 @@
 
 LLM app with RAG to chat with Gmail in just 30 lines of Python Code. The app uses Retrieval Augmented Generation (RAG) to provide accurate answers to questions based on the content of your Gmail Inbox.
 
-## Features
+### Features
 
 - Connect to your Gmail Inbox
 - Ask questions about the content of your emails
 - Get accurate answers using RAG and the selected LLM
 
-## Installation
+### Installation
 
 1. Clone the repository
 

llama-3-2-finetuning/README.md

@@ -0,0 +1,85 @@
+## 🦙 Finetune Llama 3.2 in 30 Lines of Python
+
+This script demonstrates how to finetune the Llama 3.2 model using the [Unsloth](https://unsloth.ai/) library, which makes the process easy and fast. You can run this example to finetune Llama 3.1 1B and 3B models for free in Google Colab.
+
+### Features
+
+- Finetunes Llama 3.2 model using the Unsloth library
+- Implements Low-Rank Adaptation (LoRA) for efficient finetuning
+- Uses the FineTome-100k dataset for training
+- Configurable for different model sizes (1B and 3B)
+
+### Installation
+
+1. Clone the repository:
+
+```bash
+git clone https://github.com/your-username/llama-3-2-finetuning.git
+cd llama-3-2-finetuning
+```
+
+2. Install the required dependencies:
+
+```bash
+pip install -r requirements.txt
+```
+
+## Usage
+
+1. Open the script in Google Colab or your preferred Python environment.
+
+2. Run the script to start the finetuning process:
+
+```bash
+# Run the entire script
+python finetune_llama3.2.py
+```
+
+3. The finetuned model will be saved in the "finetuned_model" directory.
+
+## How it Works
+
+1. **Model Loading**: The script loads the Llama 3.2 3B Instruct model using Unsloth's FastLanguageModel.
+
+2. **LoRA Setup**: Low-Rank Adaptation is applied to specific layers of the model for efficient finetuning.
+
+3. **Data Preparation**: The FineTome-100k dataset is loaded and preprocessed using a chat template.
+
+4. **Training Configuration**: The script sets up the SFTTrainer with specific training arguments.
+
+5. **Finetuning**: The model is finetuned on the prepared dataset.
+
+6. **Model Saving**: The finetuned model is saved to disk.
+
+## Configuration
+
+You can modify the following parameters in the script:
+
+- `model_name`: Change to "unsloth/Llama-3.1-1B-Instruct" for the 1B model
+- `max_seq_length`: Adjust the maximum sequence length
+- `r`: LoRA rank
+- Training hyperparameters in `TrainingArguments`
+
+## Customization
+
+- To use a different dataset, replace the `load_dataset` function call with your desired dataset.
+- Adjust the `target_modules` in the LoRA setup to finetune different layers of the model.
+- Modify the chat template in `get_chat_template` if you're using a different conversational format.
+
+## Running on Google Colab
+
+1. Open a new Google Colab notebook.
+2. Copy the entire script into a code cell.
+3. Add a cell at the beginning to install the required libraries:
+
+```
+!pip install torch transformers datasets trl unsloth
+```
+
+4. Run the cells to start the finetuning process.
+
+## Notes
+
+- This script is optimized for running on Google Colab's free tier, which provides access to GPUs.
+- The finetuning process may take some time, depending on the model size and the available computational resources.
+- Make sure you have enough storage space in your Colab instance to save the finetuned model.

llama-3-2-finetuning/finetune_llama3.2.py

@@ -0,0 +1,60 @@
+import torch
+from unsloth import FastLanguageModel
+from datasets import load_dataset
+from trl import SFTTrainer
+from transformers import TrainingArguments
+from unsloth.chat_templates import get_chat_template, standardize_sharegpt
+
+# Load model and tokenizer
+model, tokenizer = FastLanguageModel.from_pretrained(
+    model_name="unsloth/Llama-3.2-3B-Instruct",
+    max_seq_length=2048, load_in_4bit=True,
+)
+
+# Add LoRA adapters
+model = FastLanguageModel.get_peft_model(
+    model, r=16,
+    target_modules=[
+        "q_proj", "k_proj", "v_proj", "o_proj",
+        "gate_proj", "up_proj", "down_proj"
+    ],
+)
+
+# Set up chat template and prepare dataset
+tokenizer = get_chat_template(tokenizer, chat_template="llama-3.1")
+dataset = load_dataset("mlabonne/FineTome-100k", split="train")
+dataset = standardize_sharegpt(dataset)
+dataset = dataset.map(
+    lambda examples: {
+        "text": [
+            tokenizer.apply_chat_template(convo, tokenize=False)
+            for convo in examples["conversations"]
+        ]
+    },
+    batched=True
+)
+
+# Set up trainer
+trainer = SFTTrainer(
+    model=model,
+    train_dataset=dataset,
+    dataset_text_field="text",
+    max_seq_length=2048,
+    args=TrainingArguments(
+        per_device_train_batch_size=2,
+        gradient_accumulation_steps=4,
+        warmup_steps=5,
+        max_steps=60,
+        learning_rate=2e-4,
+        fp16=not torch.cuda.is_bf16_supported(),
+        bf16=torch.cuda.is_bf16_supported(),
+        logging_steps=1,
+        output_dir="outputs",
+    ),
+)
+
+# Train the model
+trainer.train()
+
+# Save the finetuned model
+model.save_pretrained("finetuned_model")

llama-3-2-finetuning/requirements.txt

@@ -0,0 +1,5 @@
+torch
+unsloth
+transformers 
+datasets 
+trl