@tqchen, So this makes me wonder -- what are the exact reasons that we need to maintain the relay abstraction until upto the graph runtime ? As @matt-arm mentions, I quite like the idea of making fuse-ops a TIR (the improved one with blocks) pass because currently its forward guessing the semantics of the operator to know whether its fusible by categorizing them through obvious features (e.g., fusion of element-wise ops). Moreover, this links back to the old question of mine related to the new TIR proposal, would we benefit by having the whole graph / basic blocks in TIR after relay optimizations ?
--- [Visit Topic](https://discuss.tvm.apache.org/t/rfc-cascade-scheduling/8119/7) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.apache.org/email/unsubscribe/44de744da20c01ab8cbbde0c86f6e7fa0d6b96b14d0e893113802d4f6224e02e).
