Instructions to use Demondiablo/Qwen3-4B-Instruct-2507-8bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Demondiablo/Qwen3-4B-Instruct-2507-8bit with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Demondiablo/Qwen3-4B-Instruct-2507-8bit")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Demondiablo/Qwen3-4B-Instruct-2507-8bit")
model = AutoModelForCausalLM.from_pretrained("Demondiablo/Qwen3-4B-Instruct-2507-8bit", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Demondiablo/Qwen3-4B-Instruct-2507-8bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Demondiablo/Qwen3-4B-Instruct-2507-8bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Demondiablo/Qwen3-4B-Instruct-2507-8bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Demondiablo/Qwen3-4B-Instruct-2507-8bit

SGLang

How to use Demondiablo/Qwen3-4B-Instruct-2507-8bit with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Demondiablo/Qwen3-4B-Instruct-2507-8bit" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Demondiablo/Qwen3-4B-Instruct-2507-8bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Demondiablo/Qwen3-4B-Instruct-2507-8bit" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Demondiablo/Qwen3-4B-Instruct-2507-8bit",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Demondiablo/Qwen3-4B-Instruct-2507-8bit with Docker Model Runner:
```
docker model run hf.co/Demondiablo/Qwen3-4B-Instruct-2507-8bit
```

Qwen3-4B-Instruct-2507-8bit

8-bit quantized version of Qwen/Qwen3-4B-Instruct-2507 using bitsandbytes LLM.int8(). Runs on a single consumer GPU with ~5-6 GB VRAM.

Quantization Details

Field	Value
Quantization method	bitsandbytes
Precision	8-bit weights (`load_in_8bit: true`)
`llm_int8_threshold`	6.0
Base model	Qwen/Qwen3-4B-Instruct-2507
File size (`model.safetensors`)	4.42 GB
Estimated VRAM	~5.5 GB

Model Specs (from `config.json`)

Architecture: Qwen3ForCausalLM (qwen3)
Parameters: 4B
Hidden size: 2560, Layers: 36, Heads: 32, KV heads: 8
Context length: 262,144 tokens
Vocab size: 151,936
Base dtype: bfloat16
Tie word embeddings: true

How to Use

Transformers

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Demondiablo/Qwen3-4B-Instruct-2507-8bit"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    load_in_8bit=True,
    device_map="auto",
    torch_dtype=torch.float16,
)

messages = [{"role": "user", "content": "Hello, who are you?"}]
text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer(text, return_tensors="pt").to(model.device)

outputs = model.generate(**inputs, max_new_tokens=256, do_sample=True, temperature=0.7)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

This model uses bitsandbytes 8-bit quantization. Install it with pip install bitsandbytes.

vLLM

vLLM does not support bitsandbytes 8-bit directly. Use the AWQ 4-bit version for vLLM, or use Transformers / llama.cpp for this 8-bit build.

License

Apache 2.0. See the original model card for full details.

Downloads last month: 55

Safetensors

Model size

4B params

Tensor type

F32

BF16

Model tree for Demondiablo/Qwen3-4B-Instruct-2507-8bit

Base model

Qwen/Qwen3-4B-Instruct-2507

Quantized

(377)

this model

Qwen3-4B-Instruct-2507-8bit

Quantization Details

Model Specs (from config.json)

How to Use

Transformers

vLLM

License

Model tree for Demondiablo/Qwen3-4B-Instruct-2507-8bit

Model Specs (from `config.json`)