Set HuggingFaceNmtEngine to not truncate by default

[ddaspit]

This change is 

[johnml1135]

Reviewed 2 of 2 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @ddaspit)

---

machine/jobs/huggingface/hugging_face_nmt_model_factory.py line 74 at r1 (raw file):

            num_beams=self._config.huggingface.generate_params.num_beams,
            batch_size=self._config.huggingface.generate_params.batch_size,
            truncation=TruncationStrategy.LONGEST_FIRST,

Wouldn't this cause memory issues with the odd super-long segment of 2000 tokens? Why the need for the change?

[ddaspit]

Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @johnml1135)

---

machine/jobs/huggingface/hugging_face_nmt_model_factory.py line 74 at r1 (raw file):

Previously, johnml1135 (John Lambert) wrote…

Wouldn't this cause memory issues with the odd super-long segment of 2000 tokens? Why the need for the change?

This doesn't change the behavior for Serval at all. It simply changes the default truncation strategy for the HuggingFaceNmtEngine back to not truncate. Serval jobs still use the longest first truncation strategy. We are just now setting it in the HuggingFaceNmtModelFactory, which is the more appropriate place.

[johnml1135]

machine/jobs/huggingface/hugging_face_nmt_model_factory.py line 74 at r1 (raw file):

Previously, ddaspit (Damien Daspit) wrote…

This doesn't change the behavior for Serval at all. It simply changes the default truncation strategy for the HuggingFaceNmtEngine back to not truncate. Serval jobs still use the longest first truncation strategy. We are just now setting it in the HuggingFaceNmtModelFactory, which is the more appropriate place.

ok.

[johnml1135]

It looks like the tokenizer code is not mixed with this pull request.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Files	Coverage Δ
...translation/huggingface/hugging_face_nmt_engine.py	91.19% <100.00%> (ø)

📢 Thoughts on this report? Let us know!.

[johnml1135]

Reviewable status: 0 of 10 files reviewed, all discussions resolved