File size: 2,106 Bytes
b49f7d7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 | ---
language:
- en
license: apache-2.0
library_name: transformers
tags:
- mixture-of-experts
- MoE
- coding
- 4-bit
- nf4
- bitsandbytes
- quantized
- LFM2
- Qwen
base_model:
- Akahsizrr/fuse-1-Lite
- LiquidAI/LFM2.5-2.6B
- Qwen/Qwen3.6-35B-A3B
pipeline_tag: text-generation
inference: false
---
# fuse-1 Lite — 4-bit Quantized (NF4)
> **4-bit NF4 quantized version of [fuse-1 Lite](https://huggingface.co/Akahsizrr/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
```python
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
| Precision | VRAM | GPU |
|-----------|------|-----|
| **4-bit (this model)** | **3.36 GB** | T4, RTX 3060, M2 Pro |
| 8-bit | 6.00 GB | T4, L4, RTX 3060 |
| bfloat16 | ~12 GB | L4, A10G, RTX 4090 |
See the [main model card](https://huggingface.co/Akahsizrr/fuse-1-Lite) for full architecture details, training info, and technical report.
|