--- tags: - gemma-4 - peft - lora - finn-organization license: gemma --- # e1 Fine-tuned from `google/gemma-4-E4B-it-qat-q4_0-unquantized`. ## Description e1 is an AI model created by Finn Organization. It is fine-tuned on a diverse dataset covering mathematics, physics, computer science, biology, chemistry, philosophy, linguistics, and more. ## Usage ```python from unsloth import FastLanguageModel from peft import PeftModel model, processor = FastLanguageModel.from_pretrained( "google/gemma-4-E4B-it-qat-q4_0-unquantized", max_seq_length=2048, load_in_4bit=True, ) model = PeftModel.from_pretrained(model, "Finn-Technologies/e1") messages = [ {"role": "system", "content": "You are e1, an AI model created by Finn Organization."}, {"role": "user", "content": "Your prompt here"}, ] text = processor.tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=True) outputs = model.generate(**processor(text, return_tensors="pt"), max_new_tokens=200) print(processor.decode(outputs[0])) ``` ## Training - **Base model**: Gemma 4 E4B QAT (4-bit quantized) - **Method**: QLoRA (r=16, target_modules=[q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj]) - **Train loss**: 1.09 - **Eval loss**: 1.02