Instructions to use natnitaract/tinfield-1-nvfp4-probe with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use natnitaract/tinfield-1-nvfp4-probe with Transformers:

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

pipe = pipeline("image-text-to-text", model="natnitaract/tinfield-1-nvfp4-probe")
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("natnitaract/tinfield-1-nvfp4-probe")
model = AutoModelForMultimodalLM.from_pretrained("natnitaract/tinfield-1-nvfp4-probe", 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 natnitaract/tinfield-1-nvfp4-probe with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "natnitaract/tinfield-1-nvfp4-probe"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "natnitaract/tinfield-1-nvfp4-probe",
		"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/natnitaract/tinfield-1-nvfp4-probe

SGLang

How to use natnitaract/tinfield-1-nvfp4-probe 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 "natnitaract/tinfield-1-nvfp4-probe" \
    --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": "natnitaract/tinfield-1-nvfp4-probe",
		"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 "natnitaract/tinfield-1-nvfp4-probe" \
        --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": "natnitaract/tinfield-1-nvfp4-probe",
		"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 natnitaract/tinfield-1-nvfp4-probe with Docker Model Runner:
```
docker model run hf.co/natnitaract/tinfield-1-nvfp4-probe
```

Tinfield-1 NVFP4

Public research conversion of badtheorylabs/Tinfield-1, derived from Qwen/Qwen3.8-Flash-Next. The source checkpoint was pinned to f7c714a97b71342db9e2fc04009d2704c656c80c.

Checkpoint format

This Hugging Face repository contains the complete original export, including all 10 safetensors shards:

Routed experts: NVIDIA ModelOpt NVFP4 W4A4, block size 16.
PLE embeddings, vision, MTP and other non-expert weights: retained in BF16.
Architecture: Qwen4ExpForConditionalGeneration / qwen4_exp.
The original configuration, tokenizer, processors and tensor index accompany the shards.

The ple-int8-e579 Kaggle variant is a separate quantized PLE derivative. This Hugging Face checkpoint contains BF16 PLE. Load all shards and the configuration from this repository together.

Conversion and runtime

Conversion used NVIDIA Model-Optimizer recipe commit 87c9f8cf83021957d1a1a575c90c9a4eaaf7ef0c, ModelOpt 0.46.0, Transformers 5.17.0 and PyTorch 2.13.0+cu130. This export has no additional SFT or RL training.

Use a compatible Qwen4Exp/NVFP4 serving runtime. The recorded functional configuration used vLLM 0.30.0 on one NVIDIA B200, BF16 PLE offloaded to CPU, and approximately 377 GB host RAM. This is a measured configuration; lower-memory hardware requires separate validation. ModelOpt reported that Transformers 5.17.0 was outside its tested version range.

Evaluation record

The historical 48-episode first-crack probe recorded 20 cracks (41.67%), versus 26/72 (36.11%) for its historical stock baseline. The observed +5.56 percentage points did not meet the preregistered +10-point gate: NO_GO. Serving runtimes differed, so this comparison does not isolate a causal model improvement. See evaluation-summary.json for the aggregate result and its limits.

These results apply to the original BF16 PLE export. Scores from later Kaggle runs with INT8 PLE apply to that separate checkpoint/runtime/harness combination.

Release contents and integrity

The 2026-10-05 publication cleanup retains all model shards and serving assets byte-for-byte. release-manifest.json records their sizes and hashes. Internal calibration examples, per-game probe records, machine diagnostics and conversion scripts have been archived outside this public release. The model card retains the aggregate historical evaluation result.

License and credits

Qwen Community License 1.0. Credits to badtheorylabs for Tinfield-1, Qwen for the base model, and NVIDIA Model Optimizer for the quantization tooling.

Downloads last month: 63

Safetensors

Model size

117B params

Tensor type

BF16

I64

Model tree for natnitaract/tinfield-1-nvfp4-probe

Base model

Qwen/Qwen3.8-Flash-Next

Finetuned

badtheorylabs/Tinfield-1

Quantized

(7)

this model