nanoGPT FP8 compilation failed

[yizhuoz004]

debug.txt

Error:

(t5799)error: result at index 0 has type 'tensor<1x?x768xf8E4M3FN>', but decomposition has type 'tensor<?x?x?xf8E4M3FN>'

[christopherbate]

This one is a simple fix, I'll ensure it gets sync'd up here tomorrow

[pranavm-nvidia]

I have a draft PR to enable float8, but still seeing the same error:

    (t4723)error: result at index 0 has type 'tensor<1x?x768xf8E4M3FN>', but decomposition has type 'tensor<?x?x?xf8E4M3FN>'

    This error occured while trying to compile the following FlatIR expression:
          |
          | t4723: [rank=(3), shape=((-1, -1, -1)), dtype=(float8), loc=(gpu:0)] = ConvertOp(t_inter1106)
          | 


    Note: This originated from the following expression:

    --> /tripy/tripy/frontend/module/linear.py:136 in __call__()
          |
      136 |                 q_x = quantize(x, self.input_scale, self.quant_dtype)
          |                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^