(meta)
| 72 | logits_out = torch.empty((v, h), dtype=torch.float32, device=device) |
| 73 | |
| 74 | def grid(meta): |
| 75 | gv = triton.cdiv(v, meta["BLOCK_SIZE_V"]) |
| 76 | gh = triton.cdiv(h, meta["BLOCK_SIZE_H"]) |
| 77 | return (min(num_sms, gv * gh),) |
| 78 | |
| 79 | kernel_kwargs = dict( |
| 80 | weights_ptr=weights, |
nothing calls this directly
no outgoing calls
no test coverage detected