Overwriting pytorch version 2.5.1 and fairseq2

[argonauts18]

Hello, I was wondering if anyone has attempted to train the base model in Colab or has faced a similar issue that I am having outside of Colab?

After installing pytorch 2.5.1 (cu121) and fairseq2 0.3.0rc1 (cu121) and run pip install -e ".[data,eval]", this command overwrites pytorch 2.5.1 --> 2.6.0 and fairseq2 0.3.0rc1 --> 0.4.5. According to the repo, LCM is dependent on fairseq2 0.3.0rc1. Was wondering if there was a workaround or if this is a Colab issue. Thank you!

[Puddings22]

im having another issue where i cant find fairseq2 0.3.0rc1

example:

ERROR: Could not find a version that satisfies the requirement fairseq2==0.3.0rc1 (from versions: 0.1.0, 0.1.1, 0.2.0, 0.2.1, 0.3.0, 0.4.0, 0.4.1, 0.4.2, 0.4.3, 0.4.4, 0.4.5)
ERROR: No matching distribution found for fairseq2==0.3.0rc1

how did you solve that one?

[tnixon]

@Puddings22 - it appears that 0.3.0rc1 is still available in FB's own package index:
facebookresearch/fairseq2#1134 (comment)

However, it would be nice to bring this project up to date with a more recent stable release of fairseq2...

[XiangningLin]

@tnixon Hello, the FB's own package does not work either now due to the lxml dependency error.

[xymou]

Any suggestion on this issue?

[CharlemagneBrain]

pip install fairseq2[arrow]==v0.3.0rc1 --pre --extra-index-url https://fair.pkg.atmeta.com/fairseq2/whl/rc/pt2.5.1/cu121 --upgrade