Add support for tt.dot using 3d tensors - #13
Conversation
|
I'd be more inclined to wait for the original authors to continue their work here rather than just proactively fork all PRs. |
Maybe we’ll have to wait until the heat death of the universe. 😂 I thought it’s fine to just fork it over directly, just go ahead and review it as usual. I’ll follow up on any issues if they come up. |
|
such as this pr: triton-lang/triton#8766 |
|
I mean no disrespect to the original author. I’ve also acknowledged the source of the fork in the comments. |
|
I personally think as a common courtesy, it's better to let the original authors complete their original work. Also licensing and legal matters are complicated. The original authors agreed to Microsoft's CLA -- not Meta's CLA. So I'm not really sure if we can just directly copy the work over here. To be safe on the legal safe and courteous to the original authors, would you mind closing these PRs and asking them if they have any interests in carrying these over themselves? |
|
I don't know to contact them, and I don’t really have the time to pursue that. I brought these PRs over because they seemed helpful to the |
fork from microsoft/triton-shared#352
Really simple change to emit linalg.batch_matmul instead of linalg.matmul if the input to tt.dot is a 3d tensor.