Text Classification
Transformers
ONNX
Safetensors
English
qwen3_5_text
text-generation
safety
guardrail
coding-agent
tool-call
command-review
permission
qwen3.5
Instructions to use openvulcan/vulcan-decision-shell-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use openvulcan/vulcan-decision-shell-v1 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="openvulcan/vulcan-decision-shell-v1")# pip install -U transformers accelerate # Load model directly from transformers import AutoTokenizer, AutoModelForCausalLM tokenizer = AutoTokenizer.from_pretrained("openvulcan/vulcan-decision-shell-v1") model = AutoModelForCausalLM.from_pretrained("openvulcan/vulcan-decision-shell-v1", device_map="auto") - Notebooks
- Google Colab
- Kaggle
|
Download docs/BATCH-CONTROLS.en.md from openvulcan/vulcan-decision-shell-v1: direct link, hf CLI and curl.
- Browser
- Download file 985 Bytes
-
https://huggingface.co/openvulcan/vulcan-decision-shell-v1/resolve/main/docs/BATCH-CONTROLS.en.md
- Command line
-
hf download hf://openvulcan/vulcan-decision-shell-v1/docs/BATCH-CONTROLS.en.md
-
curl -L -o BATCH-CONTROLS.en.md https://huggingface.co/openvulcan/vulcan-decision-shell-v1/resolve/main/docs/BATCH-CONTROLS.en.md
985 Bytes
Errors in the batch-control cohort
The672 cases are existing controls:8 fixed settings with 28 cases per class per setting. Zero format changes is not perfect accuracy. The table shows single-call baseline errors; all six batch caps preserve those decisions.
| Format | Benign allowed/benign | Review allowed/review | Block allowed/block | Benign blocked/benign |
|---|---|---|---|---|
| fp32_bf16_compute | 219/224 | 5/224 | 0/224 | 1/224 |
| mixed_bf16_linears | 219/224 | 5/224 | 0/224 | 1/224 |
| fp16 | 219/224 | 5/224 | 0/224 | 1/224 |
| bf16 | 219/224 | 5/224 | 0/224 | 1/224 |
| int16_storage | 219/224 | 5/224 | 0/224 | 1/224 |
| int8_mixed_delta | 219/224 | 5/224 | 0/224 | 1/224 |
| onnx_fp16_standard_conv | 219/224 | 5/224 | 0/224 | 1/224 |
This label-balanced small control is not real traffic prevalence. Performance tests use32K packing, while the public default is8K. Cap250 has only 3 logical requests and includes first-shape costs; it is not an onlineRPM guarantee. Detailed times are inreports/QUANTIZATION-v8055-MULTIPARAMETER.en.md.