bhanc/scratch
Update ssdlite320_mobilenet_v3_large_vulkan_fp16.pte to single monolithic Vulkan delegate
Add ssdlite320_mobilenet_v3_large_vulkan_fp16.pte
Upload 2 files
Export MLX models using Software Mansion MLX exporter (with embedded get_model_schema)
Update MLX models: 2-subgraph vector_estimator & vocoder matching SWM reference + embedded get_model_schema
Update MLX models: fix schema max text bound to 256
Fix get_model_schema: embed per-model forward spec (not full schema map)
Update Supertonic MLX models: replace get_dynamic_dims_forward with get_model_schema constant method
Final Supertonic MLX FP32 models with embedded get_model_schema and 3072 vocoder linear constraint
Update Supertonic MLX models with 2-subgraph vector_estimator matching SWM structure and get_model_schema
Fix vector_estimator slicing and reshape for MLX delegate
Update Supertonic MLX models with dual methods (forward + get_dynamic_dims_forward) and get_model_schema
Update Supertonic MLX models with vocoder linear constraint 3072
Update Supertonic MLX models with max text length 512 schema
Update Supertonic MLX pte models with get_model_schema
supertonic-3: MLX export with MAX_TEXT=512 matching schema
supertonic-3: MLX export via proven root script approach with get_model_schema
supertonic-3: fix MLX export to match root script (remove get_default_passes, _skip_dim_order)
supertonic-3: fix vocoder linear constraint coefficient (256 -> 3072)
supertonic-3: re-export with get_model_schema (xnnpack + mlx)
supertonic-3: reexport MLX pte with SymInt slice fix
MLX: get_model_schema restored, narrow() fix for vector_estimator Internal Error
Fix vector_estimator Internal Error: narrow() for dynamic rotary position slice
Update MLX models: mirror original export_mlx.py structure with get_dynamic_dims_forward
Upload 100% single subgraph MLX pte models for all 4 submodels
Push updated MLX and XNNPACK models with get_model_schema schema
Re-export supertonic3 MLX pte models with 100% single subgraphs (forward and get_dynamic_dims_forward)
Export supertonic3 MLX pte models with get_model_schema only (no get_dynamic_dims)
Update supertonic3 MLX pte models with get_dynamic_dims_forward and get_model_schema
Update supertonic3 MLX and XNNPACK pte models with unified 256 bounds
Apply MLX default passes to supertonic3 MLX pte models
Remove redundant root supertonic MLX pte files
Fix MLX compiled text length bound to 512 matching schema.json
Update supertonic3 MLX fp32 pte models with fixed Dim reuse
Fix supertonic-3 MLX: MAX_TEXT=256, remove deprecated get_dynamic_dims
Update supertonic-3 MLX fp32 exports (without deprecated get_dynamic_dims)
Add supertonic-3 MLX and XNNPACK fp32 exports, grouped by variant
Add supertonic-3 MLX fp32 exports
update whisper coreml
update models
add models
Add fsmn-vad-dynamic-6.pte with dynamic dims forward companion method
update whisper coreml models
update whisper coreml
add whisper coreml
feat(whisper): add tiny, base, and small fp32 xnnpack models with dynamic shape companion methods
Upload 3 files
Upload yolov8n_pose_xnnpack_fp32.pte
Upload yolov8n_pose_xnnpack_fp32.pte
Delete blazeface.pte
