Fetching metadata from the HF Docker repository...
-
Add direct-load BF16 repeat comparison
-
Add bounded larger-family CUDA quantizer controls
-
Load 8B directly inside allocated ZeroGPU call
-
Switch larger-family control to root-loaded Qwen3 8B
-
Disable unused KV cache for embedding controls
-
Expose larger-family benchmark traceback
-
Isolate GTE startup packing from completed 0.6B model
-
Move GTE 1.5B CUDA placement to ZeroGPU startup
-
Add mounted 1.5B and 8B BF16 cross-runtime controls
-
Lazy-load CUDA quantizer configuration inside GPU call
-
Add bounded bitsandbytes INT8 and NF4 controls
-
Add bounded ZeroGPU BF16 CUDA control
