Guillaume Lagrange comments

Results 243 comments of


                                            Guillaume Lagrange

STATUS_ENTRYPOINT_NOT_FOUND

Hmmm this typically happens when the linked libtorch version doesn’t match what the `tch` crate expects. There are a couple of issues on this over on there repo, especially for...

Add burn-peft: Parameter-Efficient Fine-Tuning (LoRA, DoRA, QLoRA, QDoRA)

Thanks for the contribution :slightly_smiling_face: /edit: just realized, I had marked this PR last week in my list since it included such a large area but I never commented. It...

Add burn-peft: Parameter-Efficient Fine-Tuning (LoRA, DoRA, QLoRA, QDoRA)

> totally reasonable thanks. Thanks for understanding!

Clarify `init_lr` in NoamLrSchedulerConfig (acts as scaling factor, not initial starting LR)

Sorry for the late reply, this one slipped through the notifications it seems! You're right, that's not really an accurate description. It is a scaling factor.

Clarify `init_lr` in NoamLrSchedulerConfig (acts as scaling factor, not initial starting LR)

Fixed in #3742

module 'pyarrow' has no attribute 'PyExtensionType'

Hmm that seems to be originating from the huggingface datasets library. `PyExtensionType` in `pyarrow` seems to have been removed in 2023. So perhaps your setup has some conflicting dependencies? Are...

module 'pyarrow' has no attribute 'PyExtensionType'

Ok thanks for confirming 🙂 It uses the huggingface dataset loader, which leverages the `datasets` python library to easily download the datasets from the HF hub. Somehow the python env...

Tensor operation for creating one-hot tensor from batched token sequences

Completed in #2413

Tensor operation for creating one-hot tensor from batched token sequences

Ahh you're right sorry! I'll leave this open.

[WIP] panic should be a Scalar instead!

Thanks for reporting! Linking a related issue: #2699 Support for scalar tensors in ONNX import has been added progressively but the importer is not complete. The effort is mostly community-driven...