update env example to use Groq instead of HuggingFace

Switched the primary LLM provider to Groq with clear comments
on which model to pick, what each variable does, and when the
HF vars are actually needed (only for deployment to HF Spaces).

.env.example

@@ -1,19 +1,36 @@
-# Copy this to .env and fill in your values
-# Never commit .env to git
+# Copy this file to .env and fill in your values.
+# Never commit .env to git — it's already in .gitignore.
 
-# HuggingFace
-HF_TOKEN=hf_your_token_here
-API_BASE_URL=https://api-inference.huggingface.co/models
-MODEL_NAME=mistralai/Mistral-7B-Instruct-v0.3
+# ------------------------------------------------------------------
+# Groq (required — Person 3's LLM pipeline uses this)
+# Get your key at: https://console.groq.com → API Keys
+# ------------------------------------------------------------------
+GROQ_API_KEY=gsk_your_key_here
 
-# Anthropic (used by P3's attack classifier as fallback)
-ANTHROPIC_API_KEY=sk-ant-your_key_here
+# Which Groq model to use.
+# Fast + free options: llama-3.1-8b-instant, mixtral-8x7b-32768
+# Smarter but slower: llama-3.3-70b-versatile
+MODEL_NAME=llama-3.1-8b-instant
 
-# Server
+# ------------------------------------------------------------------
+# Server settings
+# ------------------------------------------------------------------
+
+# Maximum number of attack turns per episode
 MAX_TURNS=10
+
+# Set to true to enable FastAPI debug mode and verbose logging
 DEBUG=false
-DEFAULT_LLM_PROVIDER=huggingface
 
-# Timeouts
+# How long to wait for a single Groq API call (seconds)
 LLM_TIMEOUT=30
+
+# How many times to retry a failed Groq call before giving up
 LLM_MAX_RETRIES=3
+
+# ------------------------------------------------------------------
+# HuggingFace (only needed if deploying to HF Spaces)
+# The inference.py attacker script uses this to call the HF API
+# ------------------------------------------------------------------
+HF_TOKEN=hf_your_token_here
+API_BASE_URL=https://api-inference.huggingface.co/models