| # TinyWorld Space runtime (Modal-backed). Heavy inference runs on Modal and is | |
| # called over HTTP, so the Space stays light and builds fast. | |
| # | |
| # To run inference IN the Space on ZeroGPU instead (needs HF PRO + ZeroGPU hardware), | |
| # use requirements-zerogpu.txt and set the Space variable TINYWORLD_INFER=local. | |
| gradio>=6.18.0 | |
| httpx | |
| numpy | |
| soundfile | |
| cohere | |