Simon Salmon
commited on
Commit
•
1741106
1
Parent(s):
5d95541
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,44 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
This can be used to paraphrase. I recommend using the code I have attached below. You can generate it without using LogProbs, but you are likely to be best served by manually examining the most likely outputs.
|
2 |
+
|
3 |
+
```
|
4 |
+
from transformers import AutoTokenizer, AutoModelWithLMHead
|
5 |
+
tokenizer = AutoTokenizer.from_pretrained("gpt2")
|
6 |
+
model = AutoModelWithLMHead.from_pretrained("BigSalmon/ParaphraseParentheses2.0")
|
7 |
+
```
|
8 |
+
|
9 |
+
Example Prompt:
|
10 |
+
```
|
11 |
+
the nba is [mask] [mask] viewership.
|
12 |
+
the nba is ( facing / witnessing / confronted with / suffering from / grappling with ) ( lost / tanking ) viewership...
|
13 |
+
|
14 |
+
ai is certain to [mask] the third industrial revolution.
|
15 |
+
ai is certain to ( breed / catalyze / inaugurate / catalyze / usher in / call forth / turn loose / lend its name to ) the third industrial revolution.
|
16 |
+
|
17 |
+
the modern-day knicks are a disgrace to [mask].
|
18 |
+
the modern-day knicks are a disgrace to the franchise's ( rich legacy / tradition of excellence / uniquely distinguished record ).
|
19 |
+
|
20 |
+
HuggingFace is [mask].
|
21 |
+
HuggingFace is ( an amazing company /
|
22 |
+
```
|
23 |
+
|
24 |
+
```
|
25 |
+
import torch
|
26 |
+
prompt = "Insert Your Prompt Here. It is Best To Have a Few Examples Before Like The Example Prompt Shows."
|
27 |
+
text = tokenizer.encode(prompt)
|
28 |
+
myinput, past_key_values = torch.tensor([text]), None
|
29 |
+
myinput = myinput
|
30 |
+
myinput= myinput.to(device)
|
31 |
+
logits, past_key_values = model(myinput, past_key_values = past_key_values, return_dict=False)
|
32 |
+
logits = logits[0,-1]
|
33 |
+
probabilities = torch.nn.functional.softmax(logits)
|
34 |
+
best_logits, best_indices = logits.topk(500)
|
35 |
+
best_words = [tokenizer.decode([idx.item()]) for idx in best_indices]
|
36 |
+
text.append(best_indices[0].item())
|
37 |
+
best_probabilities = probabilities[best_indices].tolist()
|
38 |
+
words = []
|
39 |
+
for i in range(500):
|
40 |
+
m = ([best_words[i]])
|
41 |
+
m = str(m)
|
42 |
+
m = m.replace("[' ", "").replace("']", "")
|
43 |
+
print(m)
|
44 |
+
```
|