John Giorgi

Convert all corpora to Standoff and add them to the `standoff` directory.

## 🚀 Feature [Seq2Seq tasks](https://github.com/Lightning-AI/lightning-flash/blob/master/flash/text/seq2seq/core/model.py) tasks (and tasks that inherit from it like [`SummarizationTask`](https://github.com/Lightning-AI/lightning-flash/blob/master/flash/text/seq2seq/summarization/model.py)) only allow a user to specify a couple of arguments to `model.generate` https://github.com/Lightning-AI/lightning-flash/blob/651e85851509fd04f723caedfef8d487d77df4e0/flash/text/seq2seq/core/model.py#L139-L144 however, the [`generate`](https://huggingface.co/docs/transformers/v4.26.0/en/main_classes/text_generation#transformers.GenerationMixin.generate)...

Looks like python3.6 is no longer supported by GitHub actions. Drop it from the build config and add python3.9 in its place.

Unable to use BLEURT in offline mode

6

### Describe the bug Trying to use [BLEURT](https://github.com/huggingface/datasets/tree/main/metrics/bleurt) in offline mode fails. The script and model weights are cached to disk fine (when in _online_ mode). In _offline_ mode, it...

Can unsloth models be compiled by TensorRT-LLM?

1

Hi! I am wondering if there are any modifications made by `FastLanguageModel ` that would cause problems when compiling a model with [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM)? I suppose I can try myself (and...

I am currently working on the same dataset, so I went looking for SOTA performance and I am fairly confident your model achieves it. I added the macro-f1 metric to...

Try training on BioRED (https://ftp.ncbi.nlm.nih.gov/pub/lu/BioRED/)

Ideally, the [unit test for our model](https://github.com/JohnGiorgi/seq2rel/blob/d91ccaa7df1cc6345ec04f6aaf4e2401de9397e2/tests/models/test_copynet_seq2rel.py#L22) would check that the model can memorize a single training example. This is technically possible by providing the argument `metric_to_check` to [`ensure_model_can_train_save_and_load`](https://docs.allennlp.org/main/api/common/testing/model_test_case/#ensure_model_can_train_save_and_load). However,...

Right now, we have a deserialized output format that looks like: ```python [ { "ADE": [ (("fenoprofen", "DRUG"), ("pure red cell aplasia", "EFFECT")) ] } ] ``` it would be...