Commit
·
8de1c3a
1
Parent(s):
1c26bab
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Python T5 base model
|
2 |
+
|
3 |
+
Pre-trained model on CodeSearchNet Python dataset using a span-masking objective. The training objective and model were introduced in this paper and first released in this repository. PyT5 model used git-t5 framework build on top of JAX/Flax to pre-train the model on a TPU v3-8 node.
|
4 |
+
|
5 |
+
# How to use
|
6 |
+
|
7 |
+
You can use this model to denoise span-masked sequences. Note, that you'll need to add some boilerplate code for adding the noise to your sequences.
|
8 |
+
|
9 |
+
Add the following code for encoding an input text:
|
10 |
+
```python
|
11 |
+
from typing import Dict, Optional, Tuple
|
12 |
+
|
13 |
+
import numpy as np
|
14 |
+
import torch
|
15 |
+
from transformers import PreTrainedTokenizerBase
|
16 |
+
|
17 |
+
from git_t5.data import DataCollatorForT5MLM
|
18 |
+
|
19 |
+
|
20 |
+
def encode(
|
21 |
+
tokenizer: PreTrainedTokenizerBase,
|
22 |
+
text: str,
|
23 |
+
noise_density: float = 0.15,
|
24 |
+
mean_noise_span_length: float = 3.0,
|
25 |
+
extra_tokens_per_span_inputs: int = 1,
|
26 |
+
extra_tokens_per_span_targets: int = 1,
|
27 |
+
seed: Optional[int] = None,
|
28 |
+
) -> Tuple[Dict[str, torch.Tensor], int]:
|
29 |
+
def compute_lengths(tokens_length: int) -> Tuple[int, int]:
|
30 |
+
num_noise_tokens = int(round(tokens_length * noise_density))
|
31 |
+
num_nonnoise_tokens = tokens_length - num_noise_tokens
|
32 |
+
num_noise_spans = int(round(num_noise_tokens / mean_noise_span_length))
|
33 |
+
# inputs contain all nonnoise tokens, sentinels for all noise spans
|
34 |
+
# and one EOS token.
|
35 |
+
return (
|
36 |
+
num_nonnoise_tokens + num_noise_spans * extra_tokens_per_span_inputs + 1,
|
37 |
+
num_noise_tokens + num_noise_spans * extra_tokens_per_span_targets + 1,
|
38 |
+
)
|
39 |
+
|
40 |
+
encoding = tokenizer(
|
41 |
+
text,
|
42 |
+
truncation=False,
|
43 |
+
return_attention_mask=False,
|
44 |
+
return_length=True,
|
45 |
+
)
|
46 |
+
|
47 |
+
input_length = encoding.pop("length")
|
48 |
+
input_length = input_length[0]
|
49 |
+
input_length, target_length = compute_lengths(input_length)
|
50 |
+
|
51 |
+
np.random.seed(seed)
|
52 |
+
|
53 |
+
data_collator = DataCollatorForT5MLM(
|
54 |
+
tokenizer=tokenizer,
|
55 |
+
noise_density=noise_density,
|
56 |
+
mean_noise_span_length=mean_noise_span_length,
|
57 |
+
input_length=input_length,
|
58 |
+
target_length=target_length,
|
59 |
+
eos_token_id=tokenizer.eos_token_id,
|
60 |
+
pad_token_id=tokenizer.pad_token_id,
|
61 |
+
decoder_start_token_id=tokenizer.pad_token_id,
|
62 |
+
sentinel_token_id=tokenizer.convert_tokens_to_ids("<extra_id_0>"),
|
63 |
+
)
|
64 |
+
|
65 |
+
batch = data_collator([encoding]) # type: ignore
|
66 |
+
batch = {key: torch.tensor(val) for key, val in batch.items()}
|
67 |
+
|
68 |
+
return batch, target_length
|
69 |
+
```
|
70 |
+
|
71 |
+
Next, download the model and tokenizer:
|
72 |
+
```python
|
73 |
+
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer,
|
74 |
+
|
75 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("formermagic/pyt5-base")
|
76 |
+
|
77 |
+
tokenizer = AutoTokenizer.from_pretrained("formermagic/pyt5-base")
|
78 |
+
```
|
79 |
+
|
80 |
+
Finally, encode your input and generate the output sequence:
|
81 |
+
```python
|
82 |
+
text = """
|
83 |
+
def alias(self, annotationtype, set, fallback=False):
|
84 |
+
if inspect.isclass(annotationtype): annotationtype = annotationtype.ANNOTATIONTYPE
|
85 |
+
if annotationtype in self.set_alias and set in self.set_alias[annotationtype]:
|
86 |
+
return self.set_alias[annotationtype][set]
|
87 |
+
elif fallback:
|
88 |
+
return set
|
89 |
+
else:
|
90 |
+
raise KeyError("No alias for set " + set)
|
91 |
+
"""
|
92 |
+
|
93 |
+
batch, max_length = encode(tokenizer, text, seed=22)
|
94 |
+
outputs = model.generate(batch["input_ids"], max_length=max_length, num_beams=1)
|
95 |
+
print(tokenizer.batch_decode(outputs[..., 1:]))
|
96 |
+
print(tokenizer.batch_decode(batch["labels"]))
|
97 |
+
```
|
98 |
+
|
99 |
+
You should see the following output:
|
100 |
+
```shell
|
101 |
+
['<extra_id_0>, fallback=<extra_id_1> inspect<extra_id_2>.set_alias<extra_id_3> return self.set<extra_id_4>) def fallback']
|
102 |
+
['<extra_id_0>, fallback=<extra_id_1> inspect<extra_id_2>.set_alias<extra_id_3> return self.set<extra_id_4>) </s></s>']
|
103 |
+
```
|
104 |
+
|
105 |
+
As you can see, the predicted result is very close to the target sequence.
|