Instructions to use hab-swe/KumaKuma-Qwen3.5-9B-R09 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use hab-swe/KumaKuma-Qwen3.5-9B-R09 with Transformers:

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

pipe = pipeline("text-generation", model="hab-swe/KumaKuma-Qwen3.5-9B-R09")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("hab-swe/KumaKuma-Qwen3.5-9B-R09")
model = AutoModelForMultimodalLM.from_pretrained("hab-swe/KumaKuma-Qwen3.5-9B-R09", 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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use hab-swe/KumaKuma-Qwen3.5-9B-R09 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "hab-swe/KumaKuma-Qwen3.5-9B-R09"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "hab-swe/KumaKuma-Qwen3.5-9B-R09",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/hab-swe/KumaKuma-Qwen3.5-9B-R09

SGLang

How to use hab-swe/KumaKuma-Qwen3.5-9B-R09 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 "hab-swe/KumaKuma-Qwen3.5-9B-R09" \
    --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": "hab-swe/KumaKuma-Qwen3.5-9B-R09",
		"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 "hab-swe/KumaKuma-Qwen3.5-9B-R09" \
        --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": "hab-swe/KumaKuma-Qwen3.5-9B-R09",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use hab-swe/KumaKuma-Qwen3.5-9B-R09 with Docker Model Runner:
```
docker model run hf.co/hab-swe/KumaKuma-Qwen3.5-9B-R09
```

You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

KumaKuma-Qwen3.5-9B-r09

Archived copy of the exported Hugging Face checkpoint from KumaKuma core round r09:

07-24-2026-1-kumakuma9b-core-r09-global_step_50-hf

Training provenance

Architecture: Qwen3_5ForConditionalGeneration
Precision: bfloat16
Round: 50 GRPO steps
Previous checkpoint: KumaKuma core r08, step 50
Curriculum: 3,024 deterministic multi-turn rows in 378 batches
Per-batch pool mix: 4 target-stage + 2 other-stage + 2 end-to-end
Target stage: S4
Source: 126 approved clean rows
Source SHA-256: 087b4a108a487686a8d3c67f0aab35d1b0191972922e65d1fd59bdaceb679b2f
Materialized train file SHA-256: da60027744d9df15067c29327cf73c225af06349ffb96bd7f47c792a1d183564

Export integrity

model.safetensors SHA-256:

fd13812def69bf7bba734b53ab82c83f018a24d05c0d9c2661c89e4fc3916b28

Evaluation note

The round completed all 50 finite training steps. Its seven-task gate measured overall 0.585914 and task score 0.489229, but the gate decision was hold because one track-regression check failed. This is an archived research checkpoint, not a promoted production release.

Downloads last month: 4

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for hab-swe/KumaKuma-Qwen3.5-9B-R09

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Finetuned

(991)

this model