Instructions to use Kwaipilot/KAT-Coder-V2.5-Dev with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Kwaipilot/KAT-Coder-V2.5-Dev with Transformers:

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

pipe = pipeline("text-generation", model="Kwaipilot/KAT-Coder-V2.5-Dev")
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("Kwaipilot/KAT-Coder-V2.5-Dev")
model = AutoModelForMultimodalLM.from_pretrained("Kwaipilot/KAT-Coder-V2.5-Dev", 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]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Kwaipilot/KAT-Coder-V2.5-Dev with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Kwaipilot/KAT-Coder-V2.5-Dev"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Kwaipilot/KAT-Coder-V2.5-Dev",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Kwaipilot/KAT-Coder-V2.5-Dev

SGLang

How to use Kwaipilot/KAT-Coder-V2.5-Dev 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 "Kwaipilot/KAT-Coder-V2.5-Dev" \
    --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": "Kwaipilot/KAT-Coder-V2.5-Dev",
		"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 "Kwaipilot/KAT-Coder-V2.5-Dev" \
        --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": "Kwaipilot/KAT-Coder-V2.5-Dev",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Kwaipilot/KAT-Coder-V2.5-Dev with Docker Model Runner:
```
docker model run hf.co/Kwaipilot/KAT-Coder-V2.5-Dev
```

Still Waiting for Vision 👀 > Stopped Waiting. Transplanted One.

#14

by KennyVo - opened Jul 29

Discussion

KennyVo

Jul 29

•

edited about 2 hours ago

KAT-Coder-V2.5-Dev has basically become my default coding model.

The only thing I still miss is vision support. Screenshots, browser renders, UI mockups, and visual debugging are a huge part of modern development, and that's probably the biggest gap KAT still has.

I know I'm not the first person to bring this up, but consider this another vote for vision. 😄

Thanks for building such a great open-weight model.

Update Sep 09 2026

Vision is here 👀

The wait is over — I've uploaded a working vision-enabled version of KAT-Coder-V2.5-Dev:

🔗 KennyVo/KAT-Coder-V2.5-Dev-OrnithVision

What happened

The original KAT-Coder-V2.5-Dev is a pure language model — no vision tower, no MTP layers. The AWQ-INT4 quantized variant from cyankiwi added a visual tower, but all 166 bias tensors in it were zeroed out, making the vision encoder non-functional.

I fixed this by transplanting the working visual tower from Ornith-1.5-35B-A3B (same Qwen3_VisionTransformer architecture) into the KAT language backbone. 333 visual tensors replaced, all 123,493 language model tensors preserved.

Quick stats

Visual tower: 27-layer ViT, hidden_size=1152, 16 heads, patch_size=16
Language backbone: KAT-Coder-V2.5-Dev (35B MoE, 3B active) — unchanged
Quantization: AWQ INT4 (group size 32)
Total size: ~23 GB, 5 shards

Usage

from transformers import AutoModelForCausalLM, AutoProcessor

model = AutoModelForCausalLM.from_pretrained(
    "KennyVo/KAT-Coder-V2.5-Dev-OrnithVision",
    torch_dtype="auto", device_map="auto"
)
processor = AutoProcessor.from_pretrained("KennyVo/KAT-Coder-V2.5-Dev-OrnithVision")

vllm serve KennyVo/KAT-Coder-V2.5-Dev-OrnithVision --dtype float16

Reproduce it yourself

See REPRODUCE.md for step-by-step instructions, or grab the transplant script from the repo.

Fun fact: the entire combining study, Python source code, forensic analysis, and this model card were generated by KAT-Coder-V2.5-Dev itself — a model doing its own repair.

Credits: Kwaipilot (language backbone), ornith-ai (vision tower), cyankiwi (quantization).

x-polyglot-x

Jul 29

I was just wondering about this. I went searching for an mmproj file in vain. I would also love to have vision added!

zanghao

Jul 29

try this out ?

beyoru/KAT-Coder-V2.5-Dev-VL
beyoru/KAT-Coder-V2.5-Dev-VL-Flash

x-polyglot-x

Jul 29

try this out ?

beyoru/KAT-Coder-V2.5-Dev-VL
beyoru/KAT-Coder-V2.5-Dev-VL-Flash

It seems the vision component is subpar:

Vision capability — READ THIS
The vision tower was never co-trained with this checkpoint's language model, which was heavily RL-tuned for agentic coding. Measured on the bf16 parent (this build inherits the behaviour, and 4-bit experts can only make it worse):

probe	bf16 parent
Dominant colour (4 solid-colour images)	4 / 4
Shape (circle / square / triangle)	3 / 3
Text/number reading (OCR)	0 / 4 — CAT→CCT, 42→48, HELLO→Hiro
✅ Coarse visual questions: colour, shape, layout, rough scene gist.
❌ Not usable for: reading code from screenshots, OCR, UI labels, document or chart understanding — anything where one wrong character changes the answer.
If your workload is text-only, the vision tower costs you 0.9 GB and nothing else; the language model is what this checkpoint is for.

aphom

about 1 month ago

•

edited about 1 month ago

Used base model's BF16 mmproj and it works perfectly.
https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF/blob/main/mmproj-BF16.gguf

I have only tested it with llamacpp with "--image-min-tokens 2048".

x-polyglot-x

28 days ago

That worked perfectly - thank you!

otacon6530

4 days ago

I agree the base model's vision mmproj works fine.

KennyVo

about 3 hours ago

•

edited about 2 hours ago

Pushed to 1st post

KennyVo changed discussion title from Still Waiting for Vision 👀 to Still Waiting for Vision 👀 > Stopped Waiting. Transplanted One. about 2 hours ago

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