Kant-Qwen-1.5B (LoRA)

Qwen2.5-1.5B fine-tuned .

Training

• Dataset: tarnava/kant_qa (3873 examples)
• Base: Qwen/Qwen2.5-1.5B
• LoRA: r=64, 3 epochs
• Final loss: 0.21

Usage

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen2.5-1.5B", device_map="auto")
model = PeftModel.from_pretrained(model, "modular-ai/qwen")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-1.5B")

def ask_kant(q):
    prompt = f"### Instruction: You are Immanuel Kant.\n\n### Input: {q}\n\n### Response:"
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    output = model.generate(**inputs, max_new_tokens=300)
    return tokenizer.decode(output[0]).split("### Response:")[-1].strip()

print(ask_kant("What is freedom?"))