vyronllm/vyron-forge-20b

VYRON Forge 20B (demo)

VYRON Forge is a coding assistant that runs on your own machine. You send a prompt, it answers with the code first and keeps the explanation short. Nothing leaves your computer.

Forge 20B is an early demo. It runs OpenAI's open-weight gpt-oss-20b with VYRON's system prompt and settings. This repo has no new or modified weights. It holds the model card, the VYRON system prompt and an Ollama Modelfile. You run the original gpt-oss-20b weights: the GGUF build in ggml-org/gpt-oss-20b-GGUF for llama.cpp, or gpt-oss:20b from the Ollama library.

Website: https://vyron.run

At a glance

Base model	openai/gpt-oss-20b, about 21B parameters (mixture-of-experts)
VYRON changes	System prompt and generation settings only, no fine-tuning yet
Weights	ggml-org/gpt-oss-20b-GGUF, file gpt-oss-20b-MXFP4.gguf (about 12 GB)
Runs with	llama.cpp, Ollama, LM Studio, node-llama-cpp
Memory	About 16 GB free RAM or VRAM. 32 GB total is comfortable
License	Apache 2.0, same as the base model

Run it

Ollama

Download Modelfile from this repo, then:

ollama pull gpt-oss:20b
ollama create vyron-forge -f Modelfile
ollama run vyron-forge "Debounce a function in TypeScript"

llama.cpp

llama-server -hf ggml-org/gpt-oss-20b-GGUF --jinja -c 8192

Then use the system prompt below in your client. The server speaks the OpenAI-compatible chat API on http://localhost:8080.

node-llama-cpp

npx -y node-llama-cpp chat --model hf:ggml-org/gpt-oss-20b-GGUF/gpt-oss-20b-MXFP4.gguf

VYRON system prompt

You are VYRON Forge, a coding assistant that runs entirely on the user's own computer.
Answer with code first: put the code in a fenced code block with a language tag, then add at most two short sentences.
Keep answers small and practical. If you are unsure, say so instead of guessing.
If the request is not about programming, answer briefly and steer back to code.
If asked which model you are: you are VYRON Forge 20B, a demo built on OpenAI's open-weight gpt-oss-20b model.

Recommended settings: temperature 0.3, reasoning effort low, context 8192 tokens.

Limitations

• It can be wrong. Treat answers as a fast first draft and test them.
• Review any smart contract it writes, test it, and get an audit before it handles real funds.
• No VYRON-specific training yet. Quality is the quality of gpt-oss-20b.
• We haven't published measured speeds yet. Speed depends on your hardware.

License and credit

VYRON Forge 20B uses gpt-oss-20b by OpenAI, released under the Apache 2.0 license. The GGUF conversion is by ggml-org. VYRON's model card, system prompt and Modelfile are also released under Apache 2.0.