Instructions to use techwithsergiu/Qwen3.5-2B-bnb-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use techwithsergiu/Qwen3.5-2B-bnb-4bit with Transformers:

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

pipe = pipeline("image-text-to-text", model="techwithsergiu/Qwen3.5-2B-bnb-4bit")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("techwithsergiu/Qwen3.5-2B-bnb-4bit")
model = AutoModelForMultimodalLM.from_pretrained("techwithsergiu/Qwen3.5-2B-bnb-4bit")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use techwithsergiu/Qwen3.5-2B-bnb-4bit with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "techwithsergiu/Qwen3.5-2B-bnb-4bit"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "techwithsergiu/Qwen3.5-2B-bnb-4bit",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/techwithsergiu/Qwen3.5-2B-bnb-4bit

SGLang

How to use techwithsergiu/Qwen3.5-2B-bnb-4bit 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 "techwithsergiu/Qwen3.5-2B-bnb-4bit" \
    --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": "techwithsergiu/Qwen3.5-2B-bnb-4bit",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "techwithsergiu/Qwen3.5-2B-bnb-4bit" \
        --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": "techwithsergiu/Qwen3.5-2B-bnb-4bit",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use techwithsergiu/Qwen3.5-2B-bnb-4bit with Docker Model Runner:
```
docker model run hf.co/techwithsergiu/Qwen3.5-2B-bnb-4bit
```

techwithsergiu commited on Mar 8

Commit

7ff1e3b

verified ·

1 Parent(s): 54c9614

upd Model family in readme

Browse files

Files changed (4) hide show

.gitattributes +1 -0
README.md +10 -10
diagrams/diagram_01.png +0 -0
diagrams/diagram_02.png +2 -2

.gitattributes CHANGED Viewed

@@ -35,3 +35,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
 diagrams/diagram_02.png filter=lfs diff=lfs merge=lfs -text

 *tfevents* filter=lfs diff=lfs merge=lfs -text
 tokenizer.json filter=lfs diff=lfs merge=lfs -text
 diagrams/diagram_02.png filter=lfs diff=lfs merge=lfs -text
+diagrams/diagram_01.png filter=lfs diff=lfs merge=lfs -text

README.md CHANGED Viewed

@@ -19,7 +19,7 @@ Retains the full visual tower — this is a **VLM-capable** model (image + text
 Primary use-case: Unsloth LoRA fine-tuning when you need image understanding in the
 fine-tuned result.
-> If you only need text fine-tuning, use the smaller and cleaner
 > [techwithsergiu/Qwen3.5-text-2B-bnb-4bit](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B-bnb-4bit)
 > instead — same backbone, visual tower removed, lighter VRAM footprint.
@@ -33,16 +33,16 @@ fine-tuned result.
 ![](diagrams/diagram_01.png)
-| Model | Format | Visual |
 |---|---|---|
-| [Qwen/Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) | f16 | ✅ |
-| **techwithsergiu/Qwen3.5-2B-bnb-4bit** | BNB NF4 | ✅ |
-| [techwithsergiu/Qwen3.5-text-2B](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B) | bf16 | ❌ |
-| [techwithsergiu/Qwen3.5-text-2B-bnb-4bit](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B-bnb-4bit) | BNB NF4 | ❌ |
-The BNB-quantized models are roughly 40% of the original f16 size. The visual tower is
-kept at bf16 and is a fixed ~0.65 GB overhead shared across all model sizes — it accounts
-for a larger fraction of the total in smaller models (0.8B) than in larger ones (9B).
 ## Fine-tuning

 Primary use-case: Unsloth LoRA fine-tuning when you need image understanding in the
 fine-tuned result.
+> If you only need text fine-tuning, use
 > [techwithsergiu/Qwen3.5-text-2B-bnb-4bit](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B-bnb-4bit)
 > instead — same backbone, visual tower removed, lighter VRAM footprint.
 ![](diagrams/diagram_01.png)
+| Model | Type | Base model |
 |---|---|---|
+| [Qwen/Qwen3.5-2B](https://huggingface.co/Qwen/Qwen3.5-2B) | f16 · VLM · source | — |
+| **[techwithsergiu/Qwen3.5-2B-bnb-4bit](https://huggingface.co/techwithsergiu/Qwen3.5-2B-bnb-4bit)** | BNB NF4 · VLM | Qwen/Qwen3.5-2B |
+| [techwithsergiu/Qwen3.5-text-2B](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B) | bf16 · text-only | Qwen/Qwen3.5-2B |
+| [techwithsergiu/Qwen3.5-text-2B-bnb-4bit](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B-bnb-4bit) | BNB NF4 · text-only | Qwen3.5-text-2B |
+| [techwithsergiu/Qwen3.5-text-2B-GGUF](https://huggingface.co/techwithsergiu/Qwen3.5-text-2B-GGUF) | GGUF quants | Qwen3.5-text-2B |
+The visual tower is a fixed ~0.65 GB bf16 overhead shared across all model sizes.
+BNB-quantized models are roughly 40% of the original f16 size (exact ratio varies by size).
 ## Fine-tuning

diagrams/diagram_01.png CHANGED Viewed

Git LFS Details

SHA256: 5ca69d979ead26a6cbc558a7a884a2511ffe130c074e0de19c7a9c0925e56598
Pointer size: 131 Bytes
Size of remote file: 126 kB

diagrams/diagram_02.png CHANGED Viewed

Git LFS Details

SHA256: 6a5b3bf2522f461adeae89d9f66d065330a9e6868d0f472061e6941595f229cd
Pointer size: 131 Bytes
Size of remote file: 333 kB

Git LFS Details

SHA256: c83120e74d873a26508dd455b254fbdc4f0ba9a41a5b1377692cd7141c4d2f7f
Pointer size: 131 Bytes
Size of remote file: 338 kB