Marina Barannikov

Add distributed training with `accelerate`

2

## What this does Enable launching multi-GPU training with HuggingFace `accelerate` and updates evaluation to handle mixed precision with `python -m accelerate.commands.launch` [https://huggingface.co/docs/accelerate/index](https://huggingface.co/docs/accelerate/index) You can now launch scripts on multiple...

## What this does Implements the possibility to train on multiple datasets and creates a `dataset_index` token that conditions a FiLM layer applied to the inputs. That layer is applied...

## What this does Update dataset tag to : 'first dataset name...' when `dataset_repo_id` is a list. A WandB tag needs to be `str` and no more than 64 characters...

## What this does Adds the option to finetune a policy on a new device (MacBook MPS). This way the weights are loaded from the safetensors file. The YAML config...