smoky1496/SpawnPoint-Coder-7B-GGUF

SpawnPoint-Coder-7B GGUF

GGUF builds of SpawnPoint-Coder-7B for Ollama, llama.cpp and LM Studio: run Spawn Point on a Mac, an AMD GPU or a plain CPU. A 7B model that generates Godot 4 games: you describe a game, it writes the design as strict JSON, and Spawn Point compiles it into a playable Godot 4 project.

file	size	use
SpawnPoint-Coder-7B-Q4_K_M.gguf	~4.7 GB	default: Ollama, laptops, CPU
SpawnPoint-Coder-7B-Q8_0.gguf	~8.1 GB	closest to the full model, more memory

Run it with Spawn Point

pip install spawnpoint-studio
spawnpoint llm up            # without an NVIDIA GPU this pulls the Q4_K_M build into Ollama
spawnpoint create "a ninja platformer where you collect stars"

Or directly: ollama run hf.co/smoky1496/SpawnPoint-Coder-7B-GGUF:Q4_K_M. Spawn Point uses Ollama's OpenAI-compatible API with JSON-schema structured output, so answers always match its schemas.

v5 adds side-scrolling shooters, reads Vietnamese directly and answers in a small per-genre format. Through Ollama, the Q4_K_M build gets 73% of 30 human-written game ideas fully right (English and Vietnamese) and 88% of the 16 sealed prompts about shooters and the v4 fixes. It needs Spawn Point 0.3 or later (older versions: use the v4 tag of this repo).

Limitations

Tuned for the 2D game kinds Spawn Point builds (platformers, top-down arenas and survivors, bubble shooters, farms, tower defense, side-scrolling shooters) with its component library; other genres and free-form coding are no better than a general 7B model. On CPU, expect tens of seconds per design.

License

Apache 2.0, see LICENSE and NOTICE.