composer

composer copied to clipboard

self.labels not set in eval_forward() for huggingfacemodel if self.use_logits=False

1

** Environment ** ``` Collecting system information... --------------------------------- System Environment Report Created: 2024-09-20 23:30:14 UTC --------------------------------- PyTorch information ------------------- PyTorch version: 2.3.1+cu121 Is debug build: False CUDA used to build...

andydelworth

Finetune Composer model with Streaming dataset example

1

Could you please provide the script of using streaming data to complete finetune task if possible. I tried to do it but the result showed that the the model did...

zhouhao001832050

# What does this PR do? # What issue(s) does this change relate to? # Before submitting - [ ] Have you read the [contributor guidelines](https://github.com/mosaicml/composer/blob/main/CONTRIBUTING.md)? - [ ] Is...

eitanturok

** To reproduce Steps to reproduce the behavior: 1. Go to https://docs.mosaicml.com/projects/composer/en/stable/trainer/performance_tutorials/analyzing_traces.html#trainer-events ## Expected behavior There's an image of a trace there. ## Additional context Instead I see this (the...

GMNGeoffrey

Optionally use flash-attn's CE loss for metrics

1

# What does this PR do? Resubmission of https://github.com/mosaicml/composer/pull/3394 -- using FA's CE Loss results in lower peak reserved memory usage and higher throughput. We are not adding flash attention...

snarayan21

Unable to script model

5

Currently running mosaic composer in a databricks environment but experiencing different behavior of `torch.jit.script` after importing composer depending on the DBR installed. ### Code to reproduce ``` import torch class...

callmekris

Torch script export not working with hugging face model

1

When trying to export a model that was built using the hugging face class and trained with composer to torchscript a line of code in forward function in the `hf_from_composer_checkpoint`...

HassanFahmy

JackZ-db

# What does this PR do? Add eval after all event to the mosaicmllogger to pass to Mapi. This should only trigger if there is some executing evaluator: https://github.com/mosaicml/composer/blob/55f0b7d1880caaf218a56212e86f174bbc463d12/composer/trainer/trainer.py#L2700 Adding...

ethanma-db

Bump PEFT Version

1

# What does this PR do? Bumps peft to min 0.12.0, since it contains a crucial memory-saving fix when saving adapters. # What issue(s) does this change relate to? #...

snarayan21