Update README.md
Browse files
README.md
CHANGED
@@ -57,13 +57,56 @@ The model's predictions may be biased or overly reliant on the training dataset,
|
|
57 |
Use the code below to get started with the model.
|
58 |
|
59 |
```python
|
60 |
-
|
61 |
-
|
62 |
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
67 |
```
|
68 |
|
69 |
## Training Details
|
|
|
57 |
Use the code below to get started with the model.
|
58 |
|
59 |
```python
|
60 |
+
import torch
|
61 |
+
from transformers import AutoTokenizer, AutoModelForQuestionAnswering
|
62 |
|
63 |
+
# Load the model and tokenizer
|
64 |
+
model_name = "Sadat07/bert_squad"
|
65 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
66 |
+
model = AutoModelForQuestionAnswering.from_pretrained(model_name)
|
67 |
+
|
68 |
+
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
69 |
+
model.to(device)
|
70 |
+
|
71 |
+
|
72 |
+
context = """
|
73 |
+
The person who invented light was
|
74 |
+
Thomas Edison.He was born in 1879.
|
75 |
+
"""
|
76 |
+
question = "When did Thomas Edison invent?"
|
77 |
+
|
78 |
+
|
79 |
+
inputs = tokenizer(question, context, return_tensors="pt", truncation=True, max_length=512)
|
80 |
+
input_ids = inputs["input_ids"].to(device)
|
81 |
+
attention_mask = inputs["attention_mask"].to(device)
|
82 |
+
|
83 |
+
|
84 |
+
print("Tokenized Input:", tokenizer.decode(input_ids[0]))
|
85 |
+
|
86 |
+
# Perform inference
|
87 |
+
with torch.no_grad():
|
88 |
+
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
|
89 |
+
start_scores = outputs.start_logits
|
90 |
+
end_scores = outputs.end_logits
|
91 |
+
|
92 |
+
# Logits
|
93 |
+
print("Start logits:", start_scores)
|
94 |
+
print("End logits:", end_scores)
|
95 |
+
|
96 |
+
# Get start and end indices
|
97 |
+
start_idx = torch.argmax(start_scores)
|
98 |
+
end_idx = torch.argmax(end_scores) + 1
|
99 |
+
|
100 |
+
# Decode the answer
|
101 |
+
if start_idx >= end_idx:
|
102 |
+
print("Model did not predict a valid answer. Please check context and question.")
|
103 |
+
else:
|
104 |
+
answer = tokenizer.convert_tokens_to_string(
|
105 |
+
tokenizer.convert_ids_to_tokens(input_ids[0][start_idx:end_idx])
|
106 |
+
)
|
107 |
+
print(f"Question: {question}")
|
108 |
+
print(f"Answer: {answer}")
|
109 |
+
|
110 |
```
|
111 |
|
112 |
## Training Details
|