File size: 7,095 Bytes
43b66f1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229

"""
Hugging Face model interface for code generation fine-tuning.
"""
import streamlit as st
import pandas as pd
import torch
from transformers import (
    AutoTokenizer, 
    AutoModelForSeq2SeqLM, 
    Trainer, 
    TrainingArguments,
    DataCollatorForSeq2Seq,
)
from datasets import Dataset
import numpy as np
import time
import os
from pathlib import Path
import uuid
import json

@st.cache_resource(show_spinner=False)
def load_model_and_tokenizer(model_name):
    """
    Load a pre-trained model and tokenizer from Hugging Face.
    
    Args:
        model_name: Name of the model on Hugging Face (e.g., 'Salesforce/codet5-base')
        
    Returns:
        Tuple of (tokenizer, model)
    """
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    return tokenizer, model

def preprocess_code_dataset(dataset_df, tokenizer, max_input_length=256, max_target_length=256, task_prefix=""):
    """
    Preprocess the code dataset for fine-tuning.
    
    Args:
        dataset_df: Pandas DataFrame with 'input' and 'target' columns
        tokenizer: HuggingFace tokenizer
        max_input_length: Maximum length for input sequences
        max_target_length: Maximum length for target sequences
        task_prefix: Prefix to add to inputs (e.g., "translate code to comment: ")
        
    Returns:
        HuggingFace Dataset ready for training
    """
    def preprocess_function(examples):
        inputs = [task_prefix + text for text in examples["input"]]
        targets = examples["target"]
        
        model_inputs = tokenizer(inputs, max_length=max_input_length, truncation=True, padding="max_length")
        
        # Set up the tokenizer for targets
        with tokenizer.as_target_tokenizer():
            labels = tokenizer(targets, max_length=max_target_length, truncation=True, padding="max_length")
            
        model_inputs["labels"] = labels["input_ids"]
        return model_inputs
    
    # Convert DataFrame to HuggingFace Dataset
    hf_dataset = Dataset.from_pandas(dataset_df)
    
    # Split dataset into train and validation
    splits = hf_dataset.train_test_split(test_size=0.1)
    train_dataset = splits["train"]
    eval_dataset = splits["test"]
    
    # Apply preprocessing
    train_dataset = train_dataset.map(
        preprocess_function,
        batched=True,
        remove_columns=["input", "target"]
    )
    eval_dataset = eval_dataset.map(
        preprocess_function,
        batched=True,
        remove_columns=["input", "target"]
    )
    
    return train_dataset, eval_dataset

def setup_trainer(model, tokenizer, train_dataset, eval_dataset, output_dir, training_args):
    """
    Set up the Trainer for fine-tuning.
    
    Args:
        model: HuggingFace model
        tokenizer: HuggingFace tokenizer
        train_dataset: Preprocessed training dataset
        eval_dataset: Preprocessed evaluation dataset
        output_dir: Directory to save model and checkpoints
        training_args: Dictionary of training arguments
        
    Returns:
        HuggingFace Trainer
    """
    # Define training arguments
    args = TrainingArguments(
        output_dir=output_dir,
        per_device_train_batch_size=training_args.get("batch_size", 8),
        per_device_eval_batch_size=training_args.get("batch_size", 8),
        learning_rate=training_args.get("learning_rate", 5e-5),
        num_train_epochs=training_args.get("epochs", 3),
        weight_decay=training_args.get("weight_decay", 0.01),
        evaluation_strategy="epoch",
        save_strategy="epoch",
        load_best_model_at_end=True,
        push_to_hub=False,
        gradient_accumulation_steps=training_args.get("gradient_accumulation", 1),
        warmup_steps=training_args.get("warmup_steps", 100),
        logging_dir=os.path.join(output_dir, "logs"),
        logging_steps=10,
    )
    
    # Data collator
    data_collator = DataCollatorForSeq2Seq(
        tokenizer,
        model=model,
        label_pad_token_id=tokenizer.pad_token_id,
        pad_to_multiple_of=8
    )
    
    # Initialize Trainer
    trainer = Trainer(
        model=model,
        args=args,
        train_dataset=train_dataset,
        eval_dataset=eval_dataset,
        tokenizer=tokenizer,
        data_collator=data_collator,
    )
    
    return trainer

def generate_code_comment(model, tokenizer, code, max_length=100, task_prefix="translate code to comment: "):
    """
    Generate a comment for a given code snippet.
    
    Args:
        model: Fine-tuned model
        tokenizer: Tokenizer
        code: Input code snippet
        max_length: Maximum length of the generated comment
        task_prefix: Prefix to add to the input
        
    Returns:
        Generated comment as string
    """
    inputs = tokenizer(task_prefix + code, return_tensors="pt", padding=True, truncation=True)
    
    # Move inputs to the same device as model
    device = model.device
    inputs = {k: v.to(device) for k, v in inputs.items()}
    
    # Generate
    output_ids = model.generate(
        inputs["input_ids"], 
        max_length=max_length,
        num_beams=4,
        early_stopping=True
    )
    
    comment = tokenizer.decode(output_ids[0], skip_special_tokens=True)
    return comment

def generate_code_from_comment(model, tokenizer, comment, max_length=200, task_prefix="translate comment to code: "):
    """
    Generate code from a given comment/description.
    
    Args:
        model: Fine-tuned model
        tokenizer: Tokenizer
        comment: Input comment or description
        max_length: Maximum length of the generated code
        task_prefix: Prefix to add to the input
        
    Returns:
        Generated code as string
    """
    inputs = tokenizer(task_prefix + comment, return_tensors="pt", padding=True, truncation=True)
    
    # Move inputs to the same device as model
    device = model.device
    inputs = {k: v.to(device) for k, v in inputs.items()}
    
    # Generate
    output_ids = model.generate(
        inputs["input_ids"], 
        max_length=max_length,
        num_beams=4,
        early_stopping=True
    )
    
    code = tokenizer.decode(output_ids[0], skip_special_tokens=True)
    return code

def save_training_config(output_dir, config):
    """
    Save training configuration to a JSON file.
    
    Args:
        output_dir: Directory to save the configuration
        config: Dictionary with training configuration
    """
    config_path = os.path.join(output_dir, "training_config.json")
    with open(config_path, "w") as f:
        json.dump(config, f, indent=2)

def load_training_config(model_dir):
    """
    Load training configuration from a JSON file.
    
    Args:
        model_dir: Directory with the saved model
        
    Returns:
        Dictionary with training configuration
    """
    config_path = os.path.join(model_dir, "training_config.json")
    if os.path.exists(config_path):
        with open(config_path, "r") as f:
            return json.load(f)
    return {}