Text Generation
Transformers
English
Polish
jfp-aligned
jfp
deterministic
multi-agent
governance
auditable-ai
protocol
viki
tool-use
qwen2
How to use from
SGLangUse 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 "jarohullowicki/Jjfp-core-v1" \
--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": "jarohullowicki/Jjfp-core-v1",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'Quick Links
JFP-Core-v1 — Jaro Flash Protocol Aligned Model
What is this?
JFP-Core-v1 is the first model configuration aligned with the Jaro Flash Protocol (JFP) v16E.0.0 — a deterministic multi-agent AI execution framework developed by Jarosław Kuchta.
This is not a standard language model. It is a protocol-governed AI engine designed for auditable, traceable, and reproducible AI applications.
Key Properties
- Deterministic execution — consistent outputs for structured inputs
- JFP constitutional layer — built-in rules that cannot be overridden by user prompts
- Reduced hallucination — refusal protocol replaces confabulation
- Audit trail ready — every output is traceable and explainable
- Anti-drift design — behavior stays within defined protocol boundaries
- VOQL compatible — native support for VIKI Operational Query Language
Intended Use
- Auditable enterprise AI pipelines
- Multi-agent orchestration systems
- Compliance-sensitive applications (legal, medical, financial)
- JFP-standard compatible tooling
Not Intended For
- General-purpose chat
- Creative/open-ended generation without protocol constraints
- Applications requiring unpredictable or exploratory outputs
Base Model
Built as a fine-tune configuration for Qwen/Qwen2.5-1.5B-Instruct.
Lightweight local agent designed for continuous learning via JFP-BC constitution.
Tools
This model includes a native tool specification. See tools.json in this repository.
Supported tools:
code_execution— sandboxed Python/bash executionfile_operations— read/write within JFP boundariesvoql_query— VIKI Operational Query Language queriesagent_dispatch— sub-agent communication within VIKI ecosystemaudit_log— immutable action loggingapi_call— external REST API calls (whitelist only)jfp_validate— schema validation against JFP standard
Limitations
- Requires JFP-compliant input format for optimal performance
- Not designed for open-ended creative tasks
- Commercial use requires separate licensing agreement
Protocol
Built on JFP v16E.0.0 — part of the VIKI ecosystem. Author: Jarosław Kuchta | GitHub
License
cc-by-nc-4.0 — Free for non-commercial use. Commercial licensing: contact the author.
Citation
@misc{kuchta2026jfp,
author = {Jarosław Kuchta},
title = {Jaro Flash Protocol (JFP) v16E.0.0},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/jarohullowicki/Jjfp-core-v1}
}
- Downloads last month
- 10
Install from pip and serve model
# Install SGLang from pip: pip install sglang# Start the SGLang server: python3 -m sglang.launch_server \ --model-path "jarohullowicki/Jjfp-core-v1" \ --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": "jarohullowicki/Jjfp-core-v1", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'