Spaces:
Sleeping
Sleeping
Update model/utils.py
Browse files- model/utils.py +30 -7
model/utils.py
CHANGED
|
@@ -76,21 +76,44 @@ def maybe_masked_mean(t: float["b n d"], mask: bool["b n"] = None) -> float["b d
|
|
| 76 |
|
| 77 |
return num / den.clamp(min=1.0)
|
| 78 |
|
| 79 |
-
|
| 80 |
-
# simple utf-8 tokenizer, since paper went character based
|
| 81 |
def list_str_to_tensor(text: list[str], padding_value=-1) -> int["b nt"]: # noqa: F722
|
| 82 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 83 |
text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)
|
| 84 |
return text
|
| 85 |
|
| 86 |
-
|
| 87 |
-
# char tokenizer, based on custom dataset's extracted .txt file
|
| 88 |
def list_str_to_idx(
|
| 89 |
text: list[str] | list[list[str]],
|
| 90 |
-
|
| 91 |
padding_value=-1,
|
| 92 |
) -> int["b nt"]: # noqa: F722
|
| 93 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 94 |
text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)
|
| 95 |
return text
|
| 96 |
|
|
|
|
| 76 |
|
| 77 |
return num / den.clamp(min=1.0)
|
| 78 |
|
|
|
|
|
|
|
| 79 |
def list_str_to_tensor(text: list[str], padding_value=-1) -> int["b nt"]: # noqa: F722
|
| 80 |
+
# Split each string into words
|
| 81 |
+
list_words = [t.split() for t in text]
|
| 82 |
+
|
| 83 |
+
# Convert words to tensors (assuming words are already in byte format)
|
| 84 |
+
list_tensors = [torch.tensor([*bytes(" ".join(words), "UTF-8")]) for words in list_words] # ByT5 style
|
| 85 |
text = pad_sequence(list_tensors, padding_value=padding_value, batch_first=True)
|
| 86 |
return text
|
| 87 |
|
|
|
|
|
|
|
| 88 |
def list_str_to_idx(
|
| 89 |
text: list[str] | list[list[str]],
|
| 90 |
+
vocab_map: dict[str, int], # {word: idx}
|
| 91 |
padding_value=-1,
|
| 92 |
) -> int["b nt"]: # noqa: F722
|
| 93 |
+
# Split each string into words if not already split
|
| 94 |
+
if isinstance(text[0], str):
|
| 95 |
+
list_words = []
|
| 96 |
+
for t in text:
|
| 97 |
+
# Split the text by triple spaces
|
| 98 |
+
parts = t.split(" ")
|
| 99 |
+
words = []
|
| 100 |
+
for i, part in enumerate(parts):
|
| 101 |
+
# Split each part into words (by single spaces)
|
| 102 |
+
words.extend(part.split())
|
| 103 |
+
# Add a space token if there are more parts (i.e., triple spaces were present)
|
| 104 |
+
if i < len(parts) - 1:
|
| 105 |
+
words.append(" ") # Add a space token
|
| 106 |
+
list_words.append(words)
|
| 107 |
+
else:
|
| 108 |
+
list_words = text
|
| 109 |
+
|
| 110 |
+
# Convert words to their corresponding indices using vocab_map
|
| 111 |
+
list_idx_tensors = [
|
| 112 |
+
torch.tensor([vocab_map.get(word, 0) for word in words]) # Use 0 for unknown words
|
| 113 |
+
for words in list_words
|
| 114 |
+
]
|
| 115 |
+
|
| 116 |
+
# Pad the sequences
|
| 117 |
text = pad_sequence(list_idx_tensors, padding_value=padding_value, batch_first=True)
|
| 118 |
return text
|
| 119 |
|