Sable_v22_ctx32k-q4f32_1-MLC

MLC-compiled weights for the Dr. Sable browser demo. They are fetched at runtime by WebLLM and run entirely in the player's browser on WebGPU; nothing is sent to a server.

Play the demo: https://stickybitgames.itch.io/doctor-sable

developer Sticky Bit Games
publisher Thingamatronics LLC
base model Qwen2.5-1.5B-Instruct (Apache-2.0)
quantisation 4f32_1-MLC as compiled by MLC
context window 32,768 tokens
download ~880 MB
GPU memory ~3.3 GB (f32) / ~2.5 GB (f16)

This repo is the weights only. It is not a general-purpose assistant: the model is fine-tuned to play one character inside one game, and the game's engine -- not the model -- decides what is true in a run.

Desktop or laptop with a WebGPU-capable GPU. Phones and tablets cannot run it.

Downloads last month
16
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for Thingamatronics/Sable_v22_ctx32k-q4f32_1-MLC

Quantized
(374)
this model