Support Grouped Query Attention (GQA) and update flash attention options#41
Merged
Annotations
10 errors
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/ops.go#L776
impossible type assertion: target.(*Function)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/flash.go#L504
options.Scale undefined (type *compute.ScaledDotProductAttentionConfig has no field or method Scale)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/flash.go#L477
options.Mask undefined (type *compute.ScaledDotProductAttentionConfig has no field or method Mask)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/flash.go#L405
options.Scale undefined (type *compute.ScaledDotProductAttentionConfig has no field or method Scale)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/flash.go#L382
options.Mask undefined (type *compute.ScaledDotProductAttentionConfig has no field or method Mask)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/flash.go#L66
cfg.Causal undefined (type *compute.ScaledDotProductAttentionConfig has no field or method Causal)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/builder.go#L96
cannot use f (variable of type *Function) as compute.Function value in return statement: *Function does not implement compute.Function (wrong type for method FusedScaledDotProductAttention)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/builder.go#L83
cannot use b.mainFn (variable of type *Function) as compute.Function value in return statement: *Function does not implement compute.Function (wrong type for method FusedScaledDotProductAttention)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/function.go#L86
cannot use closure (variable of type *Function) as compute.Function value in return statement: *Function does not implement compute.Function (wrong type for method FusedScaledDotProductAttention)
|
|
Test with AutoInstall CPU PJRT plugin:
compute/xla/function.go#L36
cannot use (*Function)(nil) (value of type *Function) as compute.Function value in variable declaration: *Function does not implement compute.Function (wrong type for method FusedScaledDotProductAttention)
|
background
wait
wait-all
cancel
parallel
Loading