gemma-4-E2B-Claude-4.6-Opus-Reasoning-Distilled

Gemma-4-E2B-it Fine-tuned on Claude Opus 4.6 Conversations

This is a fine-tuned version of google/gemma-4-E2B-it trained on ~9,600 conversation examples generated from Claude Opus 4.6.

Model Details

• Base Model: google/gemma-4-E2B-it (2.3B effective parameters)
• Training Method: QLoRA (r=8, alpha=16)
• Dataset: 9,633 synthetic conversations generated
• Epochs: 3
• Sequence Length: 128 tokens
• Fine-tuned Layers: All linear layers (q_proj, v_proj, k_proj, o_proj, gate_proj, up_proj, down_proj)

Training Hardware

• GPU: NVIDIA RTX 5070 (12GB VRAM)
• Training Time: ~10.5 hours total (3 epochs)

Usage

from transformers import AutoProcessor, AutoModelForCausalLM

processor = AutoProcessor.from_pretrained("LyJonathan/gemma-4-e2b-claude-distilled")
model = AutoModelForCausalLM.from_pretrained("LyJonathan/gemma-4-e2b-claude-distilled", device_map="cuda", dtype=torch.float16)

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Write a short joke about saving RAM."},
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = processor(text=text, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256)
print(processor.decode(outputs[0], skip_special_tokens=True))