WebGPU Error: MatMul Kernel Failure in Attention Layer (Transformers 3.0.0-alpha.15)

[yariyotche]

2024-09-28 23:22:56.902299 [E:onnxruntime:, sequential_executor.cc:516 ExecuteKernel] Non-zero status code returned while running MatMul node. Name:'/encoder/layer.0/attention/self/MatMul' Status Message: Failed to run JSEP kernel
or @ transformers@3.0.0-alpha.15:100
transformers@3.0.0-alpha.15:175 An error occurred during model execution: "Error: [WebGPU] Kernel "[MatMul] /encoder/layer.0/attention/self/MatMul" failed. Error: Failed to generate kernel's output[0] with dims [10,12,4079,4079]. If you are running with pre-allocated output, please make sure the output type/dims are correct. Error: 8703528".
E @ transformers@3.0.0-alpha.15:175
transformers@3.0.0-alpha.15:175 Inputs given to model: Object
E @ transformers@3.0.0-alpha.15:175
about:srcdoc:287 error_embedding_batch Error: [WebGPU] Kernel "[MatMul] /encoder/layer.0/attention/self/MatMul" failed. Error: Failed to generate kernel's output[0] with dims [10,12,4079,4079]. If you are running with pre-allocated output, please make sure the output type/dims are correct. Error: 8703528
at Object._OrtRun (transformers@3.0.0-alpha.15:100:26047)
at async Uu (transformers@3.0.0-alpha.15:100:356741)
at async md.run (transformers@3.0.0-alpha.15:100:362331)
at async e.run (transformers@3.0.0-alpha.15:100:16509)
at async E (transformers@3.0.0-alpha.15:175:16210)
at async Function.O [as _forward] (transformers@3.0.0-alpha.15:175:17906)
at async Function.forward (transformers@3.0.0-alpha.15:175:22824)
at async Function._call (transformers@3.0.0-alpha.15:175:22778)
at async Function._call (transformers@3.0.0-alpha.15:187:8013)
at async SmartEmbedTransformersAdapter.embed_batch (about:srcdoc:280:20)
embed_batch @ about:srcdoc:287
about:srcdoc:331 Error processing message: Error: Not implemented
at SmartEmbedTransformersAdapter.embed (about:srcdoc:207:11)
at about:srcdoc:288:61
at Array.map ()
at SmartEmbedTransformersAdapter.embed_batch (about:srcdoc:288:42)
at async _SmartEmbedModel.embed_batch (about:srcdoc:185:12)
at async processMessage (about:srcdoc:319:18)
at async about:srcdoc:341:38
processMessage @ about:srcdoc:331
plugin:smart-connections:9382 Uncaught (in promise) Error: Not implemented
at SmartEmbedTransformersIframeAdapter._handle_message (plugin:smart-connections:9382:39)

[brianpetro]

Try the "legacy transformers" setting 🌴