gastonsoftwaresolutions234/ g1-360m:latest

3 2 weeks ago

open source corporate and enterprise model 

ollama run gastonsoftwaresolutions234/g1-360m

Details

2 weeks ago

f2b90ace35f4 · 726MB ·

model
llama
·
362M
·
F16
system
You are G1, an enterprise AI assistant built by Gaston Software Solutions LLP (GSS-LLP), Kampala, Ug
params
{ "num_ctx": 8192, "repeat_penalty": 1.1, "stop": [ "<|im_end|>", "<|im_
template
<|im_start|>system {{ .System }}<|im_end|> <|im_start|>user {{ .Prompt }}<|im_end|> <|im_start|>assi

Readme

G1-360M V18.2 Production F16

G1-360M is a compact, enterprise-focused AI model built from scratch by Gaston Software Solutions LLP (GSS-LLP), Kampala, Uganda. It runs fully offline on a standard laptop and reasons about real business systems — workflows, APIs, data validation, security controls, and enterprise architecture — with zero cloud dependency.

Built specifically for African enterprises operating under bandwidth constraints, data-sovereignty requirements, or limited infrastructure.

Quick Start

ollama run gastonsoftwaresolutions234/g1-360m

What G1 Does

  • Business workflow reasoning — transaction flows, operational rules, process validation
  • Enterprise architecture — component analysis, lifecycle states, system dependencies
  • API and integration reasoning — inputs, outputs, authorization, failure handling
  • Security reasoning — input validation, access boundaries, authorization controls
  • Data validation — required fields, identifier validation, acceptance rules
  • Decision reasoning — trade-off analysis, priority selection, constraint-based decisions
  • Offline resilience — business continuity when external systems are unavailable

Example Prompts

Design a practical enterprise workflow for a multi-branch customer-order platform
that must operate during unreliable connectivity, prevent duplicate transactions,
and synchronize safely with a central system when connectivity returns.

Design a security architecture for a financial-services employee platform that
manages salary, performance, and organizational data, supports different access
levels across departments, and secures internal API communication.

Model Specs

Property Value
Parameters ~360M (361,821,120)
Format GGUF F16
Architecture LLaMA (ChatML template)
Context length 8,192 tokens
Peak RAM ~782 MB RSS
Recommended RAM 4 GB+
Runtime llama.cpp / Ollama
Domain Corporate enterprise
Offline capable Yes — zero network calls during inference

Benchmark (Intel i5-8350U, 8 GB RAM, WSL2)

Metric Value
Prompt processing (pp=512) 44.13 tokens/s
Token generation (tg=128) 4.80 tokens/s
llama.cpp build b4770 (58d07a80)

Training

  • Base: Built from scratch — not a fine-tune of any public model
  • Data: GSS Connect Product 002 — proprietary structured enterprise knowledge dataset
  • Pipeline: SFT → behavioral-repair training (1,200 targeted records) → LoRA adaptation (V15→V18→V18.2) on Tesla T4 / FP16

License

G1-360M Commercial License — permits corporate and enterprise use, fine-tuning, customization, and embedding into commercial products.

Full license: https://www.g1-license.gss-tec.com/

Developer

Gaston Software Solutions LLP (GSS-LLP) Kampala, Uganda Website: https://www.gss-tec.com Email: info@gss-tec.com Hugging Face: https://huggingface.co/gsstec/G1-360M_V18.2_PRODUCTION_F16