Instructions to use Div97/orena-focus-procedure-w64-nw2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Div97/orena-focus-procedure-w64-nw2 with Transformers:

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

pipe = pipeline("image-text-to-text", model="Div97/orena-focus-procedure-w64-nw2")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Div97/orena-focus-procedure-w64-nw2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Div97/orena-focus-procedure-w64-nw2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Div97/orena-focus-procedure-w64-nw2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Div97/orena-focus-procedure-w64-nw2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/Div97/orena-focus-procedure-w64-nw2

SGLang

How to use Div97/orena-focus-procedure-w64-nw2 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 "Div97/orena-focus-procedure-w64-nw2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Div97/orena-focus-procedure-w64-nw2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "Div97/orena-focus-procedure-w64-nw2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Div97/orena-focus-procedure-w64-nw2",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Div97/orena-focus-procedure-w64-nw2 with Docker Model Runner:
```
docker model run hf.co/Div97/orena-focus-procedure-w64-nw2
```

DISCOVR-PROCEDURE W64→NW2

Two merged bfloat16 checkpoints used by DISCOVR-PROCEDURE, submitted to the ORena SAVE FOCUS 2026 PROCEDURE track by team Incision Impossible:

w64/: broad full-procedure temporal pointer;
nw2/: localized refinement and ordinary-question model.

Model hashes:

w64/model.safetensors 0647d7204fccbf708e5b15d8312d03062e86bea61310580b5eca97999c33fe4d
nw2/model.safetensors 3c032078c4e98a33bd7deb6b0f285ed45f0414d118fa1a086bd35d64546ba3e9

Source code: https://github.com/mdivyanshu97/orena-focus-procedure

Detailed documentation:

The detailed draft distinguishes the exact recovered W64 recipe from the NW2 fields that still require recovery from the peer H200 training machine.

Inference method

DISCOVR-PROCEDURE executes the two models sequentially:

W64 uses 128 full-procedure frames to locate a broad timestamp;
W64 is unloaded before NW2 is loaded;
NW2 refines temporal answers with 128 frames in a 1200-second window and 64 frames in a 100-second window;
ordinary questions use NW2 directly with 64 frames;
timestamped reappearance questions are rewritten as post-anchor presence queries, with four-chunk early-stop routing for needles.

Training summary

W64 is a language-only rank-16 LoRA adaptation warm-started on surgical scene-literacy questions and continued on 34,290 official FOCUS training rows from HeiCo and LapChole. NW2 is a separate language-only rank-16 continued adapter trained from the same SSG warm start on the historical capped_v1_64f.jsonl manifest. Its exact merged checkpoint and provenance are published, while its unavailable launch log and manifest-generation details are explicitly identified in the detailed method draft.

Intended use and limitations

These checkpoints are intended for non-commercial research and challenge reproduction. They are not medical devices and must not be used for clinical decision making. No patient videos or raw challenge annotations are distributed here.

License and data terms

The Qwen3-VL base model and released source code use Apache-2.0. Training also used SSG-VQA, whose repository specifies CC BY-NC-SA 4.0 for non-commercial scientific research, plus challenge datasets governed by their respective owners. The license: other metadata reflects these mixed terms; it does not replace any source-dataset license or access agreement.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for Div97/orena-focus-procedure-w64-nw2

Base model

Qwen/Qwen3-VL-4B-Instruct

Finetuned

(422)

this model