6 Downloads Updated 1 month ago
ollama run coraje187/Genesis-AI-1.5B-v2
ollama launch claude --model coraje187/Genesis-AI-1.5B-v2
ollama launch opencode --model coraje187/Genesis-AI-1.5B-v2
ollama launch hermes --model coraje187/Genesis-AI-1.5B-v2
ollama launch openclaw --model coraje187/Genesis-AI-1.5B-v2
Genesis-AI-1.5B-v2 is a highly specialized, locally-hosted LLM fine-tuned to act as the autonomous reasoning engine for the Genesis Engine AI (a Python-based PC game memory scanner and manipulation tool).
This V2 model has been aggressively fine-tuned on a custom dataset of 10,000 highly contextual game-hacking scenarios. It excels at translating natural language requests (e.g., “I just took damage” or “give me max gold”) into structured JSON commands that control the Genesis Engine’s C-level memory scanner.
Model Highlights Task-Specific Fine-Tuning: Trained on 10,000 examples mapping human gaming intent directly to memory-scanning paradigms (Exact Value, Increased/Decreased, Auto-Actions, Memory Writes). Structured Output: Strictly adheres to outputting raw JSON objects matching the Genesis Engine’s action, type, value, and do_write schema without generating conversational “slop” or markdown wrappers that break parsers. Lightweight & Fast: At 1.5 Billion parameters (Q4_K_M quantization), this model is specifically designed to run seamlessly in the background alongside resource-heavy modern PC games without causing frame drops or stutters. Offline Privacy: Everything runs 100% locally on your machine, ensuring no telemetry or game state data is sent to external cloud APIs. Use Cases This model is specifically designed to be used in tandem with the Genesis Engine AI desktop application. When hooked up to the engine, it can autonomously:
Determine whether a value should be scanned as an Integer, Float, or Double. Deduce whether to perform a “First Scan” or a “Next Scan” filtering operation based on gameplay context. Identify when a memory value should be frozen or overwritten (e.g., “give me unlimited health” triggers an exact-value write and memory freeze). Technical Details Base Model: Qwen2.5-1.5B-Instruct Parameters: 1.5B Quantization: Q4_K_M (Optimized for low VRAM usage) Training Method: LoRA/QLoRA (SFTTrainer) Dataset Size: 10,000 Custom Scenarios NOTE
This model is heavily specialized for JSON command generation within the Genesis Engine ecosystem and is not intended for general conversational chat or creative writing.
The Engine will be released very soon!