azherali commited on
Commit
3c7cb75
·
verified ·
1 Parent(s): 3861d42

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +149 -20
README.md CHANGED
@@ -19,27 +19,156 @@ should probably proofread and complete it, then remove this comment. -->
19
 
20
  # CodeDetect-UniXCoder-T2
21
 
22
- This model is a fine-tuned version of [microsoft/unixcoder-base](https://huggingface.co/microsoft/unixcoder-base) on the None dataset.
23
- It achieves the following results on the evaluation set:
24
- - Loss: 1.3991
25
- - Accuracy: 0.6569
26
- - F1: 0.6143
27
- - Precision: 0.6120
28
- - Recall: 0.6191
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
29
 
30
- ## Model description
31
-
32
- More information needed
33
-
34
- ## Intended uses & limitations
35
-
36
- More information needed
37
-
38
- ## Training and evaluation data
39
-
40
- More information needed
41
-
42
- ## Training procedure
43
 
44
  ### Training hyperparameters
45
 
 
19
 
20
  # CodeDetect-UniXCoder-T2
21
 
22
+ This model is a fine-tuned version of [microsoft/unixcoder-base](https://huggingface.co/microsoft/unixcoder-base)
23
+
24
+ ### 2. Load the Model
25
+
26
+ ```python
27
+ import torch
28
+ from transformers import AutoTokenizer, AutoModelForSequenceClassification
29
+
30
+ MODEL_NAME = "azherali/CodeDetect-UniXCoder-T2"
31
+ MAX_LENGTH = 512
32
+
33
+ # Select device
34
+ device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
35
+ print("Device:", device)
36
+
37
+ # Load tokenizer and model
38
+ tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
39
+
40
+ model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
41
+
42
+ # Configure class labels
43
+ model.config.id2label = {
44
+ "0": "human",
45
+ "1": "qwen",
46
+ "2": "phi",
47
+ "3": "meta-llama",
48
+ "4": "01-ai",
49
+ "5": "gemma",
50
+ "6": "ibm-granite",
51
+ "7": "bigcode",
52
+ "8": "deepseek",
53
+ "9": "gpt",
54
+ "10": "mistral"
55
+ }
56
+ model.config.label2id = {
57
+ "human": 0,
58
+ "qwen": 1,
59
+ "phi": 2,
60
+ "meta-llama": 3,
61
+ "01-ai": 4,
62
+ "gemma": 5,
63
+ "ibm-granite": 6,
64
+ "bigcode": 7,
65
+ "deepseek": 8,
66
+ "gpt": 9,
67
+ "mistral": 10
68
+ }
69
+
70
+ model.to(device)
71
+ model.eval()
72
+
73
+ print("Number of classes:", model.config.num_labels)
74
+ print("Label mapping:", model.config.id2label)
75
+ ```
76
+
77
+ ### 3. Prediction Function
78
+
79
+ The `predict_code()` function takes source code as input and returns the predicted label, class ID, and probability scores.
80
+
81
+ ```python
82
+ def predict_code(code: str) -> dict:
83
+ if not isinstance(code, str) or not code.strip():
84
+ raise ValueError("Please provide a non-empty code string.")
85
+
86
+ # Tokenize source code
87
+ inputs = tokenizer(
88
+ code,
89
+ truncation=True,
90
+ max_length=MAX_LENGTH,
91
+ padding=True,
92
+ return_tensors="pt",
93
+ )
94
+
95
+ # Move inputs to the selected device
96
+ inputs = {
97
+ key: value.to(device)
98
+ for key, value in inputs.items()
99
+ }
100
+
101
+ # Perform inference
102
+ with torch.inference_mode():
103
+ outputs = model(**inputs)
104
+ probabilities = torch.softmax(outputs.logits, dim=-1)[0]
105
+
106
+ # Get predicted class
107
+ predicted_id = int(probabilities.argmax().item())
108
+
109
+ labels = model.config.id2label or {}
110
+
111
+ predicted_label = labels.get(
112
+ predicted_id,
113
+ labels.get(str(predicted_id), f"class_{predicted_id}")
114
+ )
115
+
116
+ # Return prediction and class probabilities
117
+ return {
118
+ "predicted_class_id": predicted_id,
119
+ "predicted_label": predicted_label,
120
+ "class_probabilities": {
121
+ str(labels.get(i, labels.get(str(i), f"class_{i}"))):
122
+ round(float(probabilities[i].item()), 4)
123
+ for i in range(len(probabilities))
124
+ },
125
+ }
126
+ ```
127
+
128
+ ### 4. Example Inference
129
+
130
+ ```python
131
+ code = """
132
+ def add(a, b):
133
+ return a + b
134
+ """
135
+
136
+ result = predict_code(code)
137
+ print(result)
138
+ ```
139
+
140
+ ### 5. Example Output
141
+
142
+ The following output is illustrative. Actual predictions and scores depend on the input and model behavior.
143
+
144
+ ```python
145
+ {
146
+ "predicted_class_id": 9,
147
+ "predicted_label": "gpt",
148
+ "class_probabilities": {
149
+ "human": 0.0001,
150
+ "qwen": 0.4001,
151
+ "phi": 0.0014,
152
+ "meta-llama": 0.0405,
153
+ "01-ai": 0.0024,
154
+ "gemma": 0.0007,
155
+ "ibm-granite": 0.0034,
156
+ "bigcode": 0.0,
157
+ "deepseek": 0.0162,
158
+ "gpt": 0.5345,
159
+ "mistral": 0.0007,
160
+ },
161
+ }
162
+ ```
163
+
164
+ ### 6. Output Description
165
+
166
+ | Field | Description |
167
+ |---|---|
168
+ | `predicted_class_id` | Predicted class index: `0` or `1`. |
169
+ | `predicted_label` | Predicted class: `Human` or `AI`. |
170
+ | `class_probabilities` | Softmax scores for each class. |
171
 
 
 
 
 
 
 
 
 
 
 
 
 
 
172
 
173
  ### Training hyperparameters
174