Instructions to use sais-org/MKB with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sais-org/MKB with Transformers:

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

pipe = pipeline("image-text-to-text", model="sais-org/MKB")
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("sais-org/MKB")
model = AutoModelForMultimodalLM.from_pretrained("sais-org/MKB", 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 sais-org/MKB with vLLM:

Install from pip and serve model

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

SGLang

How to use sais-org/MKB 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 "sais-org/MKB" \
    --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": "sais-org/MKB",
		"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 "sais-org/MKB" \
        --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": "sais-org/MKB",
		"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 sais-org/MKB with Docker Model Runner:
```
docker model run hf.co/sais-org/MKB
```

Descartes commited on 27 days ago

Commit

3fa9c43

verified ·

1 Parent(s): b0ee5cc

Clarify total parameter count (~11B)

Browse files

Files changed (1) hide show

README.md +11 -11

README.md CHANGED Viewed

@@ -33,11 +33,11 @@ extra_gated_description: >-
 **神珍 is a unified scientific multimodal foundation model** that
 supports scientific **understanding and generation** across Earth science,
-proteins, RNA, DNA, and small molecules within a single **8B** model. Native
-scientific encoders/decoders wrap a shared **Qwen3-VL-8B-Instruct** backbone, so
-heterogeneous scientific data (sequences, molecular graphs, gridded physical
-fields, medical images) are reasoned about and generated in one representation
-space — natural language in and out, no per-task fine-tuning.
 > 📜 **Technical report coming soon.**
@@ -67,13 +67,13 @@ space — natural language in and out, no per-task fine-tuning.
 ## Benchmarks
-**神珍** (**8B**) vs **Biology-Instructions** (Llama-3.1-**8B**, text-token,
-no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
-**Bold** = best; <u>underline</u> = second-best.
 ### Biological sequence understanding
-| Task | Metric | 神珍 (8B) | Biology-Instructions (8B) | Intern-S1-Pro (~1T) |
 |:-----|:------:|:----------------:|:-------------------------:|:-------------------:|
 | DNA · Epigenetic marks (EMP) | MCC | **71.99** | 3.64 | <u>14.02</u> |
 | DNA · Promoter det. 300bp (PD300) | MCC | **91.17** | 58.18 | <u>82.65</u> |
@@ -95,7 +95,7 @@ no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
 ### Molecule understanding (SMolInstruct)
-| Task | Metric | 神珍 (8B) | LlaSMol |
 |:-----|:------:|:----------------:|:-------:|
 | BBBP | Acc | **96.95** | 74.60 |
 | HIV | Acc | **97.00** | 96.70 |
@@ -106,7 +106,7 @@ no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
 ### Earth-science forecasting — vs ECMWF HRES (day-10, global ERA5 0.25°)
-| Variable | Metric | 神珍 (8B) | ECMWF HRES (NWP) |
 |:---------|:------:|:----------------:|:----------------:|
 | Z500 | RMSE ↓ | **≈740** | ≈810 |
 | T2M | RMSE ↓ (K) | **≈2.65** | ≈2.90 |

 **神珍 is a unified scientific multimodal foundation model** that
 supports scientific **understanding and generation** across Earth science,
+proteins, RNA, DNA, and small molecules. Native scientific encoders/decoders
+wrap a shared **Qwen3-VL-8B-Instruct** backbone (about **11B** parameters in
+total), so heterogeneous scientific data (sequences, molecular graphs, gridded
+physical fields, medical images) are reasoned about and generated in one
+representation space — natural language in and out, no per-task fine-tuning.
 > 📜 **Technical report coming soon.**
 ## Benchmarks
+**神珍** (8B backbone, ~11B total) vs **Biology-Instructions** (Llama-3.1-**8B**,
+text-token, no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific
+model). **Bold** = best; <u>underline</u> = second-best.
 ### Biological sequence understanding
+| Task | Metric | 神珍 (~11B) | Biology-Instructions (8B) | Intern-S1-Pro (~1T) |
 |:-----|:------:|:----------------:|:-------------------------:|:-------------------:|
 | DNA · Epigenetic marks (EMP) | MCC | **71.99** | 3.64 | <u>14.02</u> |
 | DNA · Promoter det. 300bp (PD300) | MCC | **91.17** | 58.18 | <u>82.65</u> |
 ### Molecule understanding (SMolInstruct)
+| Task | Metric | 神珍 (~11B) | LlaSMol |
 |:-----|:------:|:----------------:|:-------:|
 | BBBP | Acc | **96.95** | 74.60 |
 | HIV | Acc | **97.00** | 96.70 |
 ### Earth-science forecasting — vs ECMWF HRES (day-10, global ERA5 0.25°)
+| Variable | Metric | 神珍 (~11B) | ECMWF HRES (NWP) |
 |:---------|:------:|:----------------:|:----------------:|
 | Z500 | RMSE ↓ | **≈740** | ≈810 |
 | T2M | RMSE ↓ (K) | **≈2.65** | ≈2.90 |