Instructions to use jakejharris/jspark3 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jakejharris/jspark3 with Transformers:

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

pipe = pipeline("image-text-to-text", model="jakejharris/jspark3")
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("jakejharris/jspark3")
model = AutoModelForMultimodalLM.from_pretrained("jakejharris/jspark3", 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 jakejharris/jspark3 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jakejharris/jspark3"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jakejharris/jspark3",
		"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/jakejharris/jspark3

SGLang

How to use jakejharris/jspark3 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 "jakejharris/jspark3" \
    --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": "jakejharris/jspark3",
		"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 "jakejharris/jspark3" \
        --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": "jakejharris/jspark3",
		"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 jakejharris/jspark3 with Docker Model Runner:
```
docker model run hf.co/jakejharris/jspark3
```

jspark3 / THIRD_PARTY_NOTICES.md

jakejharris

Publish JSpark3 v1 card and verified metadata

e9cbbaf verified 16 days ago

preview code

Raw

History Blame Contribute Delete

2.22 kB

Third-party acknowledgements

This repository is an independent research implementation. It does not wholesale-vendor the projects below, but its design and intended experiments build on their work. The BTX writer is the disclosed derived exception.

Robert J. Aumann and Lloyd S. Shapley for the Aumann-Shapley value.
Joshua Hill and NVIDIA Model Optimizer PR #2183 for Aumann-Shapley quantization sensitivity and the associated coverage/additivity analysis.
Albert Tseng, Qingyao Sun, David Hou, Christopher De Sa, and the QTIP/QuIP# authors for trellis quantization and incoherence-processing foundations.
turboderp and ExLlamaV3 contributors for EXL3/Trellis quantization and its encoder/runtime ecosystem.
The Qwen Team for Qwen3 and Qwen/Qwen3-30B-A3B-Base (Apache-2.0).
Z.ai for GLM-5.2, the source model in the preserved prior-control lineage.
malaiwah for the GLM-5.2 MTP-78 overlay and calibration capture, and Josh Cartu for the associated MTP-78 recipe and rank-sliced runtime work. These credits identify the named predecessor's provenance; the Qwen experiment does not incorporate the MTP-78 runtime or draft layer.
The MC-MoE, HIGGS, PQI, GuidedQuant, MoEQuant, EAC-MoE, and VSRAQ authors for the mixed-precision, end-loss, router-aware, and route-shift research identified precisely in docs/REFERENCES.md.
Luke Alonso and Local Inference Lab contributors for B12X, the vLLM fork, and local runtime research. The official BTX writer ports atom assembly from B12X btx_synth.py at the pinned commit; that module is marked Apache-2.0 and the complete upstream license is preserved in THIRD_PARTY_LICENSES/B12X-APACHE-2.0.txt.
NVIDIA Model Optimizer, vLLM, Hugging Face Transformers, safetensors, and huggingface_hub contributors.
Google DeepMind for Gemma 4, used as the planned portability model.

Any future vendored adapter must retain the upstream license and notice in the same commit that introduces the code.

These acknowledgements identify intellectual and software lineage; they do not imply endorsement of this repository by any named person or project. See docs/REFERENCES.md for the method-to-component mapping and primary links.