Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

[inductor] KeyError: 'BLOCK_K' when Enabling Triton BMM

See original GitHub issue

When I set torchinductor.config.triton.use_bmm = True, I get the following exception:

  File "/opt/conda/lib/python3.7/site-packages/torchinductor/triton_ops/batched_matmul.py", line 274, in bmm_out
    _kernel[grid](a, b, c, M, N, K, K, 1, N, 1, N, 1, GROUP_M=8, ACC_TYPE=ACC_TYPE)
  File "/opt/conda/lib/python3.7/site-packages/triton/runtime/jit.py", line 106, in launcher
    return self.run(*args, grid=grid, **kwargs)
  File "/opt/conda/lib/python3.7/site-packages/triton/runtime/autotuner.py", line 199, in run
    kwargs[v] = heur({**dict(zip(self.arg_names, args)), **kwargs})
  File "/opt/conda/lib/python3.7/site-packages/torchinductor/triton_ops/batched_matmul.py", line 15, in <lambda>
    "EVEN_K": lambda args: args["K"] % (args["BLOCK_K"] * args["SPLIT_K"]) == 0,
KeyError: 'BLOCK_K'

You can reproduce by running benchmarks/microbenchmarks/inductor_bmm.py with the latest version of triton.

I’m really not sure what the canonical triton version is at this point, but I’m reporting it here regardless.

Issue Analytics

State:
Created a year ago
Comments:9 (4 by maintainers)

Top GitHub Comments

1reaction

desertfirecommented, Oct 11, 2022

We need to add tests for this (and other similar ops).

0reactions

netw0rkf10wcommented, Dec 13, 2022

@jansel Thanks!

Top Results From Across the Web

Untitled

Enabling modding for pc, Y793h dell, Podkarpacka troja cennik, Pmpr concurso 2013 noticias ... Ijaminecraft one command block tnt, Hopeanuoli ja weed pelit, ......