composer
composer copied to clipboard
mosaicml
Supercharge Your Model Training
# What does this PR do? Fix the auto-microbatch. Before this change, composer added `sync_hook` to module.register_forward_hook and module.register_full_backward_hook. those hooks are triggered **AFTER** forward and backward of the original...
# What does this PR do? This decouples max_duration from t_max in LR schedulers. # What issue(s) does this change relate to? # Before submitting - [ ] Have you...
# What does this PR do? Bug where if user misspecified load_path and it did not exist, then `.metadata` file would not exist and then composer would think it was...
# What does this PR do? This PR removes/deprecates the ICL(Dataset|Metric) subclasses and migrates the relevant tests. This PR is not strictly necessary but would help prevent confusion about where...
Hi folks, I am using llm-foundry to train some LLMs, and trying to save checkpoints directly to network drive (AWS on-prem storage). The issue I am hitting looks like this:...
Hello, The `.fit()` method of the Trainer is [missing the `optimizers` parameters](https://github.com/mosaicml/composer/blob/4c5ba954e3007ce2af6eb3003efa9d76de38c959/composer/trainer/trainer.py#L1611C17-L1611C17) although it is [part of its documentation](https://github.com/mosaicml/composer/blob/4c5ba954e3007ce2af6eb3003efa9d76de38c959/composer/trainer/trainer.py#L1725). It would be nice to keep the optimizers and the schedulers...
# What does this PR do? This PR adds a callback that logs ICL outputs during eval. It modifies the custom metrics to keep track of incorrect model outputs. Each...
Hi, I was wondering regarding your code here. https://github.com/mosaicml/composer/blob/a7cad7c221ce8ad9697bde50db0b3f37f8b8025e/composer/datasets/in_context_learning_evaluation.py#L655 Why do you assume right padding (for InContextLearningMultipleChoiceTaskDataset problem, but also some others)? 1. Shouldn't the padding_side be derived from the...
# What does this PR do? This PR exposes `_convert_time` into the API. As discussed in the related issue, it has been converted into a State method. # What issue(s)...
# What does this PR do? This PR introduces the execution prediction task. It is an auxiliary task compatible with any code evaluation dataset that requires the model to inspect...