Cleanup: Remove legacy files

example.ipynb

@@ -1,41 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "07604227",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "9882fd75",
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "fsdp",
-   "language": "python",
-   "name": "python3"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.12.5"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 5
-}

utils/masked_data_modeling_loss.py

@@ -1,24 +0,0 @@
-import torch
-from einops import rearrange
-'''
-    Simple class to do all MLM sort of loss operations in one place
-'''
-class MaskedDataLossWithSoftmax(torch.nn.Module):
-    def __init__(self, ignore: int=-100, reduction: str='mean', weight=None):
-        super(MaskedDataLossWithSoftmax, self).__init__()
-        self.loss = torch.nn.CrossEntropyLoss(ignore_index=-100, 
-                    reduction=reduction, 
-                    weight=weight)
-
-    def __call__(self, logits: torch.Tensor, 
-                        labels: torch.Tensor
-                        )-> torch.Tensor:
-        """
-            Logits: [batch_size, seq_len, vocab_size]; without softmax applied
-            Labels should have -100 for all indices that are not part of masked tokens
-
-        """        
-        logits = rearrange(logits, 'b s v -> b v s')
-        loss = self.loss(logits, labels)
-        
-        return loss

utils/yaml_util.py

@@ -1,24 +0,0 @@
-import yaml
-class MyLoader(yaml.SafeLoader):
-    # returns
-    def construct_mapping(self, *args, **kwargs):
-        super().add_constructor(None, construct_undefined)
-        # when loading we want to skip keys that require construction,
-        mapping = super().construct_mapping(*args, **kwargs)
-
-        return mapping
-import typing
-class Tagged(typing.NamedTuple):
-    tag: str
-    value: object
-
-def construct_undefined(self, node):
-    if isinstance(node, yaml.nodes.ScalarNode):
-        value = self.construct_scalar(node)
-    elif isinstance(node, yaml.nodes.SequenceNode):
-        value = self.construct_sequence(node)
-    elif isinstance(node, yaml.nodes.MappingNode):
-        value = self.construct_mapping(node)
-    else:
-        assert False, f"unexpected node: {node!r}"
-    return Tagged(node.tag, value)