Howard Liberty

Original [issue 322](https://code.google.com/p/anymemo/issues/detail?id=322) created by helloworld1 on 2014-08-11T18:29:50.000Z: AnyMemo should be able to handle file opening from other apps like Google Drive and Gmail. Now apps just complain the file...

[BUG] Install on AMD ROCm system but fails to build on CUDA dependencies

10

**Describe the bug** When I try to install deepspeed 0.8.3 on AMD GPU, it is able to detect hip but it fails to build because it cannot find cuda headers.`...

gdx-setup: html build error: the method setProperty(Object, String, double) is undefined for the type JsUtils

6

Please ensure you have given all the following requested information in your report. #### Issue details Use gdx-setup on the master (2023-11-20, even with this commit https://github.com/libgdx/libgdx/commit/fe99b6a16a4146aab76a728f607d7a3bb49ece6f) to set up...

Unable to merge checkpoint use `accelerate merge-weights`

4

### System Info ```Shell - `Accelerate` version: 0.31.0 - Platform: Linux-5.15.138.1-4.cm2-x86_64-with-glibc2.35 - `accelerate` bash location: /home/jobuser/.local/bin/accelerate - Python version: 3.10.2 - Numpy version: 1.24.3 - PyTorch version (GPU?): 2.3.0.0+git97ff6cf (True)...

On version `trl==0.9.4`, SFTTrainer only support `Dataset`. However it prevent streaming large dataset. This change added `IterableDataset` so large dataset can be streamed. The key is that ``` dataset =...

When using `datasets.IterableDataset` instead of `datasets.Dataset`, trl trainer will crash ``` dataset = datasets.IterableDataset.from_generator(get_training_data(custom_args.data_path), features=datasets.Features({"prompt": datasets.Value("string")})) dataset_train, dataset_eval = dataset, dataset trainer = trl.SFTTrainer( model=model, tokenizer=tokenizer, train_dataset=dataset_train, eval_dataset=dataset_eval, dataset_text_field="prompt", max_seq_length=custom_args.max_seq_length,...