aimgo commited on
Commit
7552ad5
·
verified ·
1 Parent(s): 84944d7

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +32 -0
README.md CHANGED
@@ -34,6 +34,38 @@ The model is intended to be used on segments of **250** characters. Anything els
34
 
35
  Emendator Reconstruction: "Diligam te Domine fortitudo mea Dominus firmamentum meum refugium meum liberator meus"
36
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37
  If you use this in your work, please cite:
38
  ```
39
  @misc{mccarthy2026Emendator,
 
34
 
35
  Emendator Reconstruction: "Diligam te Domine fortitudo mea Dominus firmamentum meum refugium meum liberator meus"
36
 
37
+ To use Emendator, you can load it via the Transformers library:
38
+
39
+ ```python
40
+
41
+ from transformers import T5ForConditionalGeneration, AutoTokenizer
42
+
43
+ model_path = 'aimgo/Emendator'
44
+ tokenizer = AutoTokenizer.from_pretrained(model_path)
45
+
46
+ model = T5ForConditionalGeneration.from_pretrained(model_path, torch_dtype=torch.bfloat16).to(device)
47
+
48
+ model.eval()
49
+
50
+ enc = tokenizer(texts, return_tensors="pt", padding=True, truncation=True, max_length=256).to(self.device)
51
+
52
+ max_input_len = enc["input_ids"].shape[1]
53
+
54
+ with torch.cuda.amp.autocast(dtype=torch.bfloat16, enabled=True):
55
+ outputs = model.generate(
56
+ enc["input_ids"],
57
+ attention_mask=enc["attention_mask"],
58
+ max_new_tokens=max_input_len + 32,
59
+ num_beams=4,
60
+ do_sample=False,
61
+ early_stopping=True,
62
+ repetition_penalty=1.15,
63
+ use_cache=True,
64
+ )
65
+
66
+ outputs = tokenizer.batch_decode(outputs, skip_special_tokens=True)
67
+ ```
68
+
69
  If you use this in your work, please cite:
70
  ```
71
  @misc{mccarthy2026Emendator,