Instructions to use bottlecapai/ThinkingCap-Qwen3.6-27B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use bottlecapai/ThinkingCap-Qwen3.6-27B with Transformers:

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

pipe = pipeline("image-text-to-text", model="bottlecapai/ThinkingCap-Qwen3.6-27B")
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("bottlecapai/ThinkingCap-Qwen3.6-27B")
model = AutoModelForMultimodalLM.from_pretrained("bottlecapai/ThinkingCap-Qwen3.6-27B")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use bottlecapai/ThinkingCap-Qwen3.6-27B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "bottlecapai/ThinkingCap-Qwen3.6-27B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "bottlecapai/ThinkingCap-Qwen3.6-27B",
		"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/bottlecapai/ThinkingCap-Qwen3.6-27B

SGLang

How to use bottlecapai/ThinkingCap-Qwen3.6-27B 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 "bottlecapai/ThinkingCap-Qwen3.6-27B" \
    --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": "bottlecapai/ThinkingCap-Qwen3.6-27B",
		"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 "bottlecapai/ThinkingCap-Qwen3.6-27B" \
        --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": "bottlecapai/ThinkingCap-Qwen3.6-27B",
		"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 bottlecapai/ThinkingCap-Qwen3.6-27B with Docker Model Runner:
```
docker model run hf.co/bottlecapai/ThinkingCap-Qwen3.6-27B
```

MTP support?

by volodXYZ - opened 11 days ago

Discussion

volodXYZ

11 days ago

Any chance that you will re-train the MTP head to match the updated weights? Cheers.

klasocki

BottleCapAI org 10 days ago

@volodXYZ yes, it's on the top of our todo list 😊

Adamiros

BottleCapAI org 10 days ago

Please also give a try to the current MTP weights, they've worked decently for us so far.

vcerny

4 days ago

Would you consider training DSpark speculator model too?

klasocki

BottleCapAI org 1 day ago

@volodXYZ we validated the MTP weights and they should work as good in terms of acceptance rate and provided speedup as with the original Qwen, so I will slowly close the issue if you are ok with it :)

@vcerny for now we probably will not train additional speculators as there are a few other efforts we want to focus on, but we might consider it if it continues to be in high demand. Do you happen to have some numbers for how much better is DSpark than vanilla MTP for the original Qwen3.6-27B?

wano

1 day ago

It’s much better. Both DSpark and DFlash can significantly improve inference speed. Compared with standard decoding, DFlash can deliver several times higher performance, while DSpark typically provides an additional 10% improvement on top of DFlash.

https://z-lab.ai/projects/dflash/

vcerny

1 day ago

I can confirm MTP-4 works fine with the model.

I don't think there is enough data to evaluate MTP vs DSpark for qwen3.6 27B , but I made ChatGPT to estimate it based on results from other Qwen models and other results found online. Take it with a grain of solt...

klasocki

BottleCapAI org 1 day ago

•

edited 1 day ago

@wano @vcerny I ran a quick test on a few prompts, the original DFlash predictor trained for Qwen still works with ThinkingCap - it has close to 2x the tok/s of the vanilla MTP, average acceptance length of 7.32 (compared to ~3.6 of classic MTP) and 5x speedup over no speculative decoding. Compared to 7.36 avg accepted len that it gets with the original Qwen, the tok/s is almost the same and ours should still be much faster with the reduced thinking.

For now I'd recommend to use that, and I'll add it to our backlog to retrain the predictor properly - but can't promise when that would happen given that the benefits seem not as significant 😀

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment