Fix small-LLM B200 dockerfile

[mmarcinkiewicz]

No description provided.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[ShriyaRishab]

@suachong - can you please review this?

[ShriyaRishab]

@suachong - is this the error you're still getting? -

0000-steps/0 [default6]:[rank6]:   File "/usr/local/lib/python3.12/dist-packages/triton/backends/nvidia/compiler.py", line 297, in make_cubin

0000-steps/0 [default6]:[rank6]:     raise RuntimeError(f'Internal Triton PTX codegen error: \n{log}')

0000-steps/0 [default6]:[rank6]: RuntimeError: Internal Triton PTX codegen error:

0000-steps/0 [default6]:[rank6]: ptxas fatal   : Value 'sm_100' is not defined for option 'gpu-name'

0000-steps/0 [default6]:

0000-steps/0 [default6]:

0000-steps/0 [default6]:

0000-steps/0 [default6]:[rank6]: Set TORCHDYNAMO_VERBOSE=1 for the internal stack trace (please do this especially if you're reporting a bug to PyTorch). For even more developer context, set TORCH_LOGS="+dynamo"

@mmarcinkiewicz - can you please help with this one?

NVIDIA-SMI 580.65.06
Driver Version: 580.65.06
CUDA Version: 13.0

[suachong]

@mmarcinkiewicz is the Dockerfile for B200 working? I'm still running into the same error that shriya posted.

[mmarcinkiewicz]

I've rebuilt the image and I can repro now, which is weird because it used to work.

[mmarcinkiewicz]

@suachong please try now

[suachong]

Tested with training run to convergence.

[hiwotadese]

@suachong is this working now and can we merge this?

[suachong]

[AMD Official Use Only - AMD Internal Distribution Only] Yeah it is working. I’ve already approved the changes yesterday. I believe Michal said he wanted to push more changes but the Dockerfile he provided now works. From: Hiwot Tadese Kassa ***@***.***> Sent: Thursday, September 11, 2025 11:47 AM To: mlcommons/training ***@***.***> Cc: Chong, Su Ann ***@***.***>; Mention ***@***.***> Subject: Re: [mlcommons/training] Fix small-LLM B200 dockerfile (PR #826) Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding. [Image removed by sender.]hiwotadese left a comment (mlcommons/training#826)<#826 (comment)> @suachong<https://github.com/suachong> is this working now and can we merge this? — Reply to this email directly, view it on GitHub<#826 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BJADJOHFKOVSLYN6JQTCEIT3SGKOFAVCNFSM6AAAAACFEZUJ7CVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEOBRGUZDQMZZGQ>. You are receiving this because you were mentioned.Message ID: ***@***.***>