You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
How do you install transformers_engine==1.13?. I have seen this solved the problem KeyError: 'recipe'. It appears to me during test of evo2 (ArcInstitute/evo2#35 (comment)).
My system is the next
tranformer_engine:
2.0.0+754d2a0
CUDA:
12.4
CUDNN:
90100
Devide:
Quadro RTX 6000
torch:
2.5.1+cu124
When I use pip wheels it says there is no version compatible, not sure why. I was also wonderin if the error comes from some compatibility as well as seems that it might ask for fp8 conpatibilty. But anyways, if you have some advice on how to install earlier version would be great.
Best wishes,
Cesar
The text was updated successfully, but these errors were encountered:
Hi @ksivaman , Thanks for your response. I used pip install transformers_engine[pytorch]==1.13 or pip install transformers_engine==1.13 but none ot them worked. I also went to source, get the whole package and I did pip install . , but that installed version 2.0.0.
The first command is correct. Did you have a TE already installed from source when you trying to install the wheel? There is a known bug with older wheels that has now been resolved (see comment). Could you uninstall transformer-engine and then try to reinstall using transformers_engine[pytorch]==1.13.
Hi,
How do you install transformers_engine==1.13?. I have seen this solved the problem KeyError: 'recipe'. It appears to me during test of evo2 (ArcInstitute/evo2#35 (comment)).
My system is the next
tranformer_engine:
2.0.0+754d2a0
CUDA:
12.4
CUDNN:
90100
Devide:
Quadro RTX 6000
torch:
2.5.1+cu124
When I use pip wheels it says there is no version compatible, not sure why. I was also wonderin if the error comes from some compatibility as well as seems that it might ask for fp8 conpatibilty. But anyways, if you have some advice on how to install earlier version would be great.
Best wishes,
Cesar
The text was updated successfully, but these errors were encountered: