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", device_map="auto")
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
```

some benchmarks

#13

by owao - opened Jul 11

Discussion

owao

Jul 11

Qwen3.6-27B-MoQ-4.5bpw vs ThinkingCap-Qwen3.6-27B-Q4_K_M

I like this test I found somewhere in the past on reddit, that's testing multiple abilities at the same time.
The quants are not absolute apple to apple but that's all I can run.

PROMPT = "Reverse this string: '.DefaultCellStyle'"
GROUND_TRUTH = "elytSlleCtluafeD."
RUNS_PER_MODEL = 50
max_tokens: 6000

==================================================
Qwen3.6-27B-MTP-MoQ-4.5_110k
==================================================

All 50 runs already completed
Results: 17/50 correct (34.0%)
Tokens: sum=133493, avg=2669.9, min=1345, max=6000

==================================================
ThinkingCap-Qwen3.6-27B-Q4_K_M_100k
==================================================

All 50 runs already completed
Results: 18/50 correct (36.0%)
Tokens: sum=69059, avg=1381.2, min=487, max=6000

Not a large number of runs, but seems in line with the promise :)
Great job!

owao

Jul 13

Oh here is the python script for who is interested I forgot to include it: https://paste.sh/HO480lkt#5PIGj6qtxijHiOk8Wj7rWZJl

klasocki

BottleCapAI org Jul 14

Thanks so much for contributing the scores, it's amazing when the community validates the results!

armin5000

Jul 15

Wow, thank you very much for your work klasocki.

==================================================

Evaluating model: ThinkingCap-Qwen3.6-27B-MTP-Q6_K-128k

==================================================

Running 50 evaluation run(s)...

Prompt: Reverse this string: '.DefaultCellStyle'

Ground truth: elytSlleCtluafeD.

Results will be saved to results\ThinkingCap-Qwen3.6-27B-MTP-Q6_K-128k.json

Run 1: [FAIL] (extracted: None)

Run 2: [FAIL] (extracted: None)

Run 3: [OK] (extracted: elytSlleCtluafeD.)

Run 4: [FAIL] (extracted: None)

Run 5: [OK] (extracted: elytSlleCtluafeD.)

Run 6: [OK] (extracted: elytSlleCtluafeD.)

Run 7: [OK] (extracted: elytSlleCtluafeD.)

Run 8: [OK] (extracted: elytSlleCtluafeD.)

Run 9: [OK] (extracted: elytSlleCtluafeD.)

Run 10: [OK] (extracted: elytSlleCtluafeD.)

Run 11: [FAIL] (extracted: None)

Run 12: [OK] (extracted: elytSlleCtluafeD.)

Run 13: [OK] (extracted: elytSlleCtluafeD.)

Run 14: [FAIL] (extracted: None)

Run 15: [OK] (extracted: elytSlleCtluafeD.)

Run 16: [OK] (extracted: elytSlleCtluafeD.)

Run 17: [FAIL] (extracted: None)

Run 18: [OK] (extracted: elytSlleCtluafeD.)

Run 19: [FAIL] (extracted: None)

Run 20: [FAIL] (extracted: None)

Run 21: [FAIL] (extracted: None)

Run 22: [FAIL] (extracted: None)

Run 23: [FAIL] (extracted: None)

Run 24: [OK] (extracted: elytSlleCtluafeD.)

Run 25: [FAIL] (extracted: None)

Run 26: [OK] (extracted: elytSlleCtluafeD.)

Run 27: [OK] (extracted: elytSlleCtluafeD.)

Run 28: [FAIL] (extracted: None)

Run 29: [FAIL] (extracted: None)

Run 30: [OK] (extracted: elytSlleCtluafeD.)

Run 31: [FAIL] (extracted: None)

Run 32: [OK] (extracted: elytSlleCtluafeD.)

Run 33: [FAIL] (extracted: None)

Run 34: [OK] (extracted: elytSlleCtluafeD.)

Run 35: [OK] (extracted: elytSlleCtluafeD.)

Run 36: [OK] (extracted: elytSlleCtluafeD.)

Run 37: [OK] (extracted: elytSlleCtluafeD.)

Run 38: [OK] (extracted: elytSlleCtluafeD.)

Run 39: [OK] (extracted: elytSlleCtluafeD.)

Run 40: [OK] (extracted: elytSlleCtluafeD.)

Run 41: [OK] (extracted: elytSlleCtluafeD.)

Run 42: [FAIL] (extracted: None)

Run 43: [OK] (extracted: elytSlleCtluafeD.)

Run 44: [OK] (extracted: elytSlleCtluafeD.)

Run 45: [OK] (extracted: elytSlleCtluafeD.)

Run 46: [FAIL] (extracted: None)

Run 47: [FAIL] (extracted: None)

Run 48: [OK] (extracted: elytSlleCtluafeD.)

Run 49: [FAIL] (extracted: None)

Run 50: [OK] (extracted: elytSlleCtluafeD.)

==================================================

Results: 30/50 correct (60.0%)

Tokens: sum=49503, avg=990.1, min=498, max=6000

owao

Jul 16

Hey @armin5000 50 runs is again not really enough to fully rely on them, but I'm a bit surprised by your results, I let it ran a bit longer, for 200 runs, but can't go past 32% success rate. Even if we say the confidence interval for your 50 runs is something like +/-15%, it still seems quite diverging from my results!
If you don't mind I'd like to ask what parameters you ran this with?
Here were mine, for this last test:

--temp 1
--top-p 0.95
--top-k 20
--min-p 0
--presence-penalty 0
--repeat-penalty 1
--flash-attn on
--spec-type draft-mtp
--spec-draft-n-max 2

==================================================
Evaluating model: ThinkingCap-Qwen3.6-27B-Q4_K_M_100k
==================================================
Results: 65/200 correct (32.5%)
Tokens: sum=280031, avg=1400.2, min=266, max=6000

armin5000

Jul 17

hi, I use the Q6_K model.

==================================================
Model: ThinkingCap-Qwen3.6-27B-MTP-Q6_128k

Total runs: 200
Correct: 101/200 (50.5%)
Failed: 0
Tokens: sum=190883, avg=954.4, min=433, max=6000

"%LLAMA_EXE%" ^
-m "%MODEL%" ^
--mmproj "%MMPROJ%" ^
--alias "Qwen3.6:27B-MTP" ^
-c 128608 ^
-ngl 999 ^
--flash-attn on ^
--host 127.0.0.1 ^
--port %PORT% ^
--parallel 1 ^
--jinja ^
--cache-type-k q8_0 ^
--cache-type-v q8_0 ^
--spec-type draft-mtp ^
--spec-draft-n-max 4 ^
--image-min-tokens 1024 ^
--temp 0.2 ^
--top-k 20 ^
--top-p 0.9 ^
--min-p 0.0 ^
--presence-penalty 0.0 ^
--repeat-penalty 1.05 ^
--n-predict 8192 ^
--reasoning on

owao

Jul 17

interesting, thank you!

kalle07

26 days ago

•

edited 26 days ago

fails on lm-studio bionic agentic coding.
long working procedure ~1hour
(Perhaps I check this again on other tasks)

finaly back using:
Qwopus3.6-27B-Coder-Compat-MTP-Q4_K_M.gguf
or original
Qwen3.6-27B-UD-MTP-Q4_K_XL.gguf

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

some benchmarks

==================================================Model: ThinkingCap-Qwen3.6-27B-MTP-Q6_128k

==================================================
Model: ThinkingCap-Qwen3.6-27B-MTP-Q6_128k