Gemma-4_x_Claude_4.6_Opus

Uploaded finetuned model

• Developed by: vaibhav81331
• License: apache-2.0
• Finetuned from model : unsloth/gemma-4-E4B-it

---

base_model: unsloth/gemma-4-E4B-it tags:

• text-generation
• transformers
• unsloth
• gemma
• lora license: apache-2.0 language:
• en datasets:
• nohurry/Opus-4.6-Reasoning-3000x-filtered

---

🧠 Gemma-4 Finetuned on Opus 4.6 Reasoning

A high-quality reasoning-focused fine-tuned model built on Gemma-4 using LoRA and optimized with Unsloth ⚡

---

🚀 Model Overview

• Base Model: unsloth/gemma-4-E4B-it
• Finetuning Method: LoRA
• Framework: Unsloth + Hugging Face TRL
• Dataset: nohurry/Opus-4.6-Reasoning-3000x-filtered
• Language: English

---

✨ Key Features

✅ Strong reasoning capabilities ✅ Fast inference (optimized via Unsloth) ✅ Lightweight (LoRA adapters) ✅ Suitable for:

• Complex Q&A
• Logical reasoning
• Structured problem solving

---

📊 Training Details

• Epochs: ~1 (based on training progress)
• Learning Rate: 1.85e-4
• Loss (final): ~0.92 and decreasing
• GPU: High utilization (~98%) on RXT Pro 6000 on traning
• Stability: No gradient explosion / stable convergence

---

🧪 Example Usage

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "vaibhav81331/Gemma-4_x_Claude_4.6_Opus"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id)

prompt = "Explain why the sky is blue in simple terms."
inputs = tokenizer(prompt, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))

---

⚡ Using with LoRA (Recommended)

If this repo contains LoRA adapters:

from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained("unsloth/gemma-4-E4B-it")
model = PeftModel.from_pretrained(base_model, "vaibhav81331/Gemma-4_x_Claude_4.6_Opus")

---

📁 Repository Structure

.
├── adapter_model.safetensors
├── config.json
├── tokenizer.json
├── README.md

---

⚠️ Important Notes

• This is a fine-tuned derivative of Gemma-4
• You must comply with the original model license
• Performance depends on prompt quality and task domain

---

🧠 Limitations

• May hallucinate on out-of-domain queries
• Not guaranteed to be factually accurate
• Limited to training dataset distribution

---

🙌 Credits

• Built using Unsloth
• Training powered by Hugging Face TRL
• Dataset by nohurry

---

📜 License

Apache 2.0 (check base model license for additional constraints)

---

⭐ Support

If you found this useful:

• ⭐ Star the repo
• 🤗 Like the model on Hugging Face
• 🍴 Fork and improve!

---

This gemma4 model was trained 2x faster with Unsloth and Huggingface's TRL library.

<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>