fuse-1 Lite — 4-bit Quantized (NF4)

4-bit NF4 quantized version of fuse-1 Lite. 3.36 GB VRAM — runs on T4, RTX 3060, and consumer GPUs.

This is the bitsandbytes 4-bit quantized version of fuse-1 Lite, using NF4 (NormalFloat 4-bit) quantization with double quantization for maximum compression.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig
import torch

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)

model = AutoModelForCausalLM.from_pretrained(
    "Akahsizrr/fuse-1-Lite-4bit",
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained("Akahsizrr/fuse-1-Lite-4bit")

messages = [{"role": "user", "content": "Write a Python function to check if a number is prime."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=512, do_sample=True, temperature=0.1)

print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True))

VRAM Requirements

See the main model card for full architecture details, training info, and technical report.