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

Libraries

How to use beyoru/KAT-Coder-V2.5-Dev-VL-Flash with Transformers:

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

pipe = pipeline("image-text-to-text", model="beyoru/KAT-Coder-V2.5-Dev-VL-Flash")
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("beyoru/KAT-Coder-V2.5-Dev-VL-Flash")
model = AutoModelForMultimodalLM.from_pretrained("beyoru/KAT-Coder-V2.5-Dev-VL-Flash", 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 beyoru/KAT-Coder-V2.5-Dev-VL-Flash with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "beyoru/KAT-Coder-V2.5-Dev-VL-Flash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "beyoru/KAT-Coder-V2.5-Dev-VL-Flash",
		"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/beyoru/KAT-Coder-V2.5-Dev-VL-Flash

SGLang

How to use beyoru/KAT-Coder-V2.5-Dev-VL-Flash 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 "beyoru/KAT-Coder-V2.5-Dev-VL-Flash" \
    --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": "beyoru/KAT-Coder-V2.5-Dev-VL-Flash",
		"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 "beyoru/KAT-Coder-V2.5-Dev-VL-Flash" \
        --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": "beyoru/KAT-Coder-V2.5-Dev-VL-Flash",
		"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 beyoru/KAT-Coder-V2.5-Dev-VL-Flash with Docker Model Runner:
```
docker model run hf.co/beyoru/KAT-Coder-V2.5-Dev-VL-Flash
```

beyoru commited on Jul 28

Commit

337121a

verified ·

1 Parent(s): 4e51083

Add docker-compose.vllm.yaml: sample serving config

Browse files

Files changed (1) hide show

docker-compose.vllm.yaml +49 -0

docker-compose.vllm.yaml ADDED Viewed

	@@ -0,0 +1,49 @@

+# vLLM — cau hinh cho beyoru/KAT-Coder-V2.5-Dev-VL-Flash (NVFP4).
+#
+# ⚠️ HAI CANH BAO, doc truoc khi dung:
+#
+# 1. BAT BUOC GPU BLACKWELL (sm100/sm120): B200, B300, RTX PRO 6000 Blackwell,
+#    RTX 50-series, DGX Spark. NVFP4 chay weight 4-bit VA activation 4-bit tren
+#    FP4 tensor core. KHONG chay tren Hopper (H100/H200), Ada, Ampere.
+#    Tren nhung card do hay dung ban bf16: beyoru/KAT-Coder-V2.5-Dev-VL
+#
+# 2. CHUA KIEM CHUNG TREN PHAN CUNG THAT. Checkpoint duoc dung tren Hopper, noi
+#    NVFP4 khong chay duoc. Shape weight/scale, quantization_config va cau truc
+#    checkpoint da verify TINH; chua sinh mot token nao tren FP4 tensor core that.
+services:
+  kat-vl-flash:
+    image: vllm/vllm-openai:v0.26.0
+    container_name: kat-vl-flash
+    ipc: host
+    ports:
+      - "8000:8000"
+    volumes:
+      - /path/to/models:/models
+      - ${HOME}/.cache/huggingface:/root/.cache/huggingface
+    deploy:
+      resources:
+        reservations:
+          devices:
+            - driver: nvidia
+              device_ids: ["0"]
+              capabilities: [gpu]
+    command: >
+      --model /models/KAT-Coder-V2.5-Dev-VL-Flash
+      --served-model-name kat-vl-flash
+      --max-model-len 32768
+      --gpu-memory-utilization 0.90
+      --trust-remote-code
+# ⛔ DUNG ghim MoE backend. De vLLM tu chon. Tren Blackwell duong nhanh la
+#    CUTLASS / FlashInfer-TRTLLM / CuTe-DSL; ep Marlin se tut manh vi Marlin la
+#    kernel weight-only (W4A16), khong bao gio cham toi FP4 tensor core.
+#
+# Weight chi 22 GB (giam 68.7% so voi 70.2 GB bf16) nen vua thoai mai mot card
+# Blackwell don. KV cache da co scale fp8 hieu chuan san trong checkpoint.
+#
+# So do do chinh xac (doc duoc tu config.json):
+#   NVFP4 W4A4 : toan bo mlp.experts.*.{gate,up,down}_proj + shared_expert (group 16)
+#   FP8  W8A8  : self_attn.{q,k,v,o}_proj, linear_attn.{in_proj_qkv,in_proj_z,out_proj}, lm_head
+#   BF16       : router mlp.gate, shared_expert_gate, linear_attn.{in_proj_a,in_proj_b,norm},
+#                va TOAN BO vision tower (333 tensor)
+#   KV cache   : fp8 per-tensor static