Spaces:
Running
Running
Commit History
-
Clean up debug diagnostics now that WebGPU works end-to-end
Verified signature2cf4acc -
Use INT4 encoder (MatMulNBits — WebGPU-supported); ORT 1.23 keeps webgpu fp16 fix for decoders
Verified signatureb8c2d24 -
Per-step diagnostics: pinpoint which ORT call crashes
Verified signature7be4ffd -
Surface full error info on transcribe failure
Verified signature50bb779 -
Upgrade onnxruntime-web 1.20 → 1.23 (webgpu bundle); revert WASM forcing — 1.23 fixes fp16 transformer NaNs
Verified signaturee6b3306 -
Force decoders to WASM (WebGPU fp16 in ort-web 1.20 returns all-NaN on this transformer)
Verified signatured62ebc2 -
Diagnostics: logit type/len/health/first-8/top-5 with Number() coercion
Verified signature29d0bb3 -
Await tensor.getData() for WebGPU outputs (audio_embeds + logits) so data is actually copied back to CPU
Verified signature1320aec -
fp16: use canonical u16 bit-pattern viewed as Float16Array; diagnostic top-5 dump
Verified signatureed37d13 -
Encoder: prefer static INT8 (QLinearConv); INT4 fallback. Recovers 92.7%-class quality in browser
Verified signature9e54b79 -
Float16Array for fp16 tensors; per-session WebGPU fallback (decoders stay on webgpu)
Verified signatureaf482b4 -
Encoder: try INT8 first, auto-fallback to INT4 if ConvInteger unsupported in browser
Verified signaturea821180 -
Use ort.webgpu bundle + auto-fallback to wasm if webgpu init fails
Verified signature96f08e1 -
Bump cache key to invalidate RTN weights (GPTQ ship)
Verified signature36417e8
-
Use INT8 encoder + INT4 decoder (91.9% accuracy); force-English prompt default
Verified signature61dfe9b -
Switch Space to static SDK: pure browser inference via onnxruntime-web
Verified signaturea4c397e