metadata
base_model: unsloth/gemma-4-E2B-it
tags:
- text-generation-inference
- transformers
- unsloth
- gemma4
- trl
license: apache-2.0
language:
- en
Run in kaggle
# =========================================================
# Install dependencies (Kaggle usually already has some)
# =========================================================
!pip install -q transformers peft accelerate bitsandbytes
# =========================================================
# Imports
# =========================================================
import torch
from transformers import AutoProcessor, AutoModelForCausalLM
from peft import PeftModel
# =========================================================
# Config
# =========================================================
BASE_MODEL = "google/gemma-4-E2B-it"
LORA_MODEL = "rahul7star/gemma_4_lora"
# =========================================================
# Load processor
# =========================================================
processor = AutoProcessor.from_pretrained(BASE_MODEL)
# =========================================================
# Load base model
# =========================================================
model = AutoModelForCausalLM.from_pretrained(
BASE_MODEL,
torch_dtype=torch.float16, # safer for Kaggle GPU
device_map="auto"
)
# =========================================================
# Load LoRA adapter on top of base model
# =========================================================
model = PeftModel.from_pretrained(model, LORA_MODEL)
# optional: merge LoRA for faster inference
model = model.merge_and_unload()
print("Model + LoRA loaded successfully 🚀")
# =========================================================
# Inference function
# =========================================================
def generate_response(user_input):
messages = [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": user_input},
]
text = processor.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True,
enable_thinking=False
)
inputs = processor(text=text, return_tensors="pt").to(model.device)
input_len = inputs["input_ids"].shape[-1]
with torch.no_grad():
outputs = model.generate(
**inputs,
max_new_tokens=512,
temperature=0.7,
top_p=0.9
)
response = processor.decode(
outputs[0][input_len:],
skip_special_tokens=True
)
return response
# =========================================================
# Test
# =========================================================
print(generate_response("Write a short joke about saving RAM."))
Uploaded model
- Developed by: rahul7star
- License: apache-2.0
- Finetuned from model : unsloth/gemma-4-E2B-it
This gemma4 model was trained 2x faster with Unsloth
