Deploy ARDY ZeroGPU Blender motion API

README.md

@@ -14,6 +14,8 @@ preload_from_hub:
   - unsloth/llama-3-8b-Instruct
   - McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp
   - McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp-supervised
+  - nvidia/ARDY-Core-RP-20FPS-Horizon40
+  - nvidia/ARDY-G1-RP-25FPS-Horizon52
 ---
 
 # ARDY Motion Playground

app.py

@@ -3,6 +3,11 @@ import os
 os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
 # Load models locally (no external text-encoder service) and prefer the HF cache.
 os.environ.setdefault("TEXT_ENCODER_MODE", "local")
+# README `preload_from_hub` is mounted read-only at runtime. Keep Xet's mutable
+# chunk cache and logs in /tmp, while resolving preloaded model snapshots from
+# the normal Hub cache before attempting any network download.
+os.environ.setdefault("HF_XET_CACHE", "/tmp/huggingface-xet")
+os.environ.setdefault("LOCAL_CACHE", "true")
 
 import base64
 import gzip