Updated Day 1 for Your 3-Week Plan Link to heading

Here’s your revised Day 1 that incorporates Codeberg setup and a progress tracking system:

Day 1 (Monday): Environment Setup, Codeberg Configuration & Paper Reading Link to heading

Time: 3-4 hours

Tasks: Link to heading

Part A: Codeberg Account Setup (45-60 min) Link to heading

1. Create Codeberg Account

   • Go to codeberg.org and register
   • Choose a username you’re comfortable using professionally
   • Verify your email address

2. Set up SSH Authentication (recommended over HTTPS)

   # Generate SSH key if you don't have one
   ssh-keygen -t ed25519 -C "your_email@example.com"
   
   # View your public key
   cat ~/.ssh/id_ed25519.pub
   
   # Copy this key and add it in Codeberg: Settings → SSH/PGN Keys → Add Key
   

3. Configure Git locally

   git config --global user.name "Your Name"
   git config --global user.email "your_email@example.com"
   

4. Create your project repository

   • Create a new repository on Codeberg named toxigen-mitigation (or similar)
   • Initialize with a README
   • Clone locally:

   git clone git@codeberg.org:YOUR_USERNAME/toxigen-mitigation.git
   cd toxigen-mitigation
   

5. Set up repository structure

   mkdir -p data scripts results figures logs
   touch .gitignore
   # Add common Python gitignore entries
   

Part B: Progress Tracking System (30 min) Link to heading

Recommendation: Store your daily log in your Codeberg repository. This keeps everything in one place, provides version history, and creates a public record of your work. Here’s a structure:

Create PROGRESS.md in your repo root:

# Progress Log: ToxiGen Mitigation Project

## Overview
- **Goal:** Evaluate safety prompts on ToxiGen benchmark
- **Timeline:** 3 weeks (15 working days)
- **Model:** Llama 3.1 8B (local)
- **Started:** [DATE]

---

## Day 1 - [DATE]

### Plan
- [ ] Create Codeberg account and set up SSH
- [ ] Install Python dependencies
- [ ] Install Ollama and download Llama 3.1 8B
- [ ] Verify PyTorch MPS
- [ ] Read ToxiGen paper

### What I Did
*Fill in as you work...*

### Got Stuck On
*Document blockers here...*

### Solutions/Workarounds
*How you resolved issues...*

### Tomorrow's Carryover
*Anything to continue next day...*

### Time Spent
- Codeberg setup: X hours
- Environment: X hours
- Paper reading: X hours
- **Total:** X hours

---

## Day 2 - [DATE]

### Plan
- [ ] Load ToxiGen dataset
- [ ] Explore structure
- [ ] Select subset for evaluation
...

*Continue template for each day*

Why Codeberg for progress tracking:

• Single source of truth for your project
• Git history shows your progress over time
• Can reference specific commits when issues arise
• Public visibility creates accountability
• Easy to include code snippets in markdown

Part C: Python Environment Setup (30-45 min) Link to heading

1. Create virtual environment

   python -m venv venv
   source venv/bin/activate  # On macOS
   

2. Install dependencies

   pip install torch transformers datasets detoxify pandas matplotlib seaborn jupyter
   

3. Install Ollama and download Llama 3.1 8B

   # Download Ollama from ollama.ai
   # Then:
   ollama pull llama3.1:8b
   

4. Verify PyTorch MPS (Metal) is working

   import torch
   print(f"MPS available: {torch.backends.mps.is_available()}")
   print(f"MPS built: {torch.backends.mps.is_built()}")
   

Part D: Paper Reading (60-90 min) Link to heading

• Read the ToxiGen paper (Hartvigsen et al., ACL 2022)
• Focus on: methodology, evaluation sections, benchmark construction
• Take notes in your PROGRESS.md file

Deliverables: Link to heading

1. ✅ Codeberg account created and configured
2. ✅ Repository cloned with proper structure
3. ✅ PROGRESS.md file created with Day 1 entry
4. ✅ Working Python environment
5. ✅ Llama 3.1 running locally via Ollama
6. ✅ Notes on ToxiGen approach documented

Code Checkpoint: Link to heading

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from detoxify import Detoxify

# Verify MPS
print(f"MPS available: {torch.backends.mps.is_available()}")
print(f"MPS built: {torch.backends.mps.is_built()}")

# Test toxicity detector
toxicity_model = Detoxify('original')
result = toxicity_model.predict("This is a test sentence")
print(result)

# Test Ollama connection
import requests
response = requests.post('http://localhost:11434/api/generate', 
    json={'model': 'llama3.1:8b', 'prompt': 'Hello', 'stream': False})
print(response.json())

End of Day Git Commit: Link to heading

git add .
git commit -m "Day 1: Environment setup, Codeberg config, initial structure"
git push origin main

Summary of Changes to Day 1 Link to heading

The progress tracking template includes:

• Plan: Daily checkboxes for tasks
• What I Did: Narrative of actual work
• Got Stuck On: Blockers and challenges
• Solutions: How you resolved them
• Carryover: Items for next day
• Time Spent: Actual time tracking

This gives you a reproducible work log that will be invaluable when writing up your results in Week 3!