SlungusLoaf/pigeon-harmony-gemma-2b

Pigeon Harmony - Gemma 2B Version

Google's Gemma model - Fast and reliable!

Copy ALL of this into a NEW Colab notebook

Step 1: Install libraries

!pip install transformers torch accelerate -q

Step 2: Import

from transformers import AutoTokenizer, AutoModelForCausalLM import torch

print("🐦 Chargement de Pigeon Harmony avec Gemma...")

Step 3: Load Gemma 2B (Google's model!)

model_name = "google/gemma-2b-it" # 'it' = instruction-tuned for chat

tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModelForCausalLM.from_pretrained( model_name, torch_dtype=torch.float16, device_map="auto" )

print("✅ Pigeon Harmony avec Gemma est prêt!")

Step 4: Chat function with Gemma's format

def chat(message): # Gemma's chat format prompt = f"""user {message} model Je suis Pigeon Harmony, un assistant IA qui parle français québécois. """

inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(
    **inputs,
    max_new_tokens=200,
    temperature=0.8,
    top_p=0.9,
    top_k=50,
    do_sample=True,
    repetition_penalty=1.2,  # Prevents loops!
    no_repeat_ngram_size=3,   # No repeating phrases!
    pad_token_id=tokenizer.eos_token_id,
    eos_token_id=tokenizer.eos_token_id
)

response = tokenizer.decode(outputs[0], skip_special_tokens=True)

# Extract just the model's response
if "<start_of_turn>model" in response:
    response = response.split("<start_of_turn>model")[-1].strip()

# Remove the system prompt if it appears
if "Je suis Pigeon Harmony" in response:
    parts = response.split("Je suis Pigeon Harmony, un assistant IA qui parle français québécois.")
    if len(parts) > 1:
        response = parts[1].strip()

return response

Step 5: Test it!

print("\n🐦 Testing Pigeon Harmony avec Gemma:\n")

print("Toi: Salut! Comment ça va?") response = chat("Salut! Comment ça va?") print(f"Pigeon: {response}\n")

print("Toi: C'est quoi ton nom?") response = chat("C'est quoi ton nom?") print(f"Pigeon: {response}\n")

print("Toi: Parle-moi de la poutine") response = chat("Parle-moi de la poutine") print(f"Pigeon: {response}\n")

Step 6: Interactive chat!

print("\n💬 Mode interactif (tape 'bye' pour quitter):\n")

conversation_history = []

while True: user_input = input("\nToi: ") if user_input.lower() in ['bye', 'quit', 'exit', 'salut', 'tchao']: print("🐦 À la prochaine! Coucou!") break

response = chat(user_input)
print(f"\nPigeon: {response}")

conversation_history.append({"user": user_input, "pigeon": response})

print(f"\n✨ Tu as eu {len(conversation_history)} conversations avec Pigeon Harmony!")