gemma4-e4b-toolcall-v02-lora

Gemma 4 E4B — Tool-Calling LoRA Adapter (v0.2)

QLoRA adapter for tool-calling, designed to be applied on top of google/gemma-4-E4B-it. Achieves 94.4% accuracy on 1000 diverse tool-calling queries.

For full details, see the merged model card.

Usage

from transformers import AutoProcessor, AutoModelForMultimodalLM
from peft import PeftModel
import torch

# Load base + adapter
base = AutoModelForMultimodalLM.from_pretrained(
    "google/gemma-4-E4B-it",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)
model = PeftModel.from_pretrained(base, "roshangrewal/gemma4-e4b-toolcall-v02-lora")
model.eval()

processor = AutoProcessor.from_pretrained("google/gemma-4-E4B-it")

# Define tools and query
tools = [{"type": "function", "function": {
    "name": "get_weather",
    "description": "Get weather for a city",
    "parameters": {"type": "object", "properties": {"city": {"type": "string"}}, "required": ["city"]}
}}]
messages = [{"role": "user", "content": "What's the weather in Mumbai?"}]

text = processor.apply_chat_template(messages, tools=tools, tokenize=False, add_generation_prompt=True)
inputs = processor(text=text, return_tensors="pt").to(model.device)

output = model.generate(**inputs, max_new_tokens=200, do_sample=False)
print(processor.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=False))
# <|tool_call>call:get_weather{city:<|"|>Mumbai<|"|>}<tool_call|>

Adapter Details

BFCL Leaderboard (Official)

Evaluation (1000 queries)

Links

• Merged model: roshangrewal/gemma4-e4b-toolcall-v02
• Base model: google/gemma-4-E4B-it