File size: 16,073 Bytes
6c482f9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
"""
Agent Tuning Module for Agent Tuning Optimization Framework

This module provides functionality for efficiently tuning large language models
into specialized agents using a combination of positive examples, negative examples,
and synthetically generated interaction trajectories.
"""

import os
import torch
import numpy as np
from typing import List, Dict, Any, Union, Optional, Tuple
from tqdm import tqdm
from transformers import (
    Trainer, TrainingArguments, 
    DataCollatorForLanguageModeling,
    AutoModelForCausalLM, AutoTokenizer
)
from datasets import Dataset

from data.trajectory_data import Trajectory, TrajectoryDataset
from models.llm_interface import LLMInterface

class AgentTuner:
    """Base class for agent tuning methods."""
    
    def __init__(self, name: str):
        """
        Initialize the agent tuner.
        
        Args:
            name: Name of the tuning method
        """
        self.name = name
    
    def tune(
        self, 
        model_name: str,
        trajectories: List[Trajectory],
        **kwargs
    ) -> Tuple[Any, Dict[str, Any]]:
        """
        Tune a model into a specialized agent.
        
        Args:
            model_name: Name of the base model
            trajectories: List of training trajectories
            **kwargs: Additional tuning parameters
            
        Returns:
            Tuple of (tuned_model, training_metrics)
        """
        raise NotImplementedError("Subclasses must implement this method")
    
    def save_model(self, model: Any, path: str) -> None:
        """
        Save the tuned model.
        
        Args:
            model: Tuned model
            path: Path to save the model
        """
        raise NotImplementedError("Subclasses must implement this method")
    
    def load_model(self, path: str) -> Any:
        """
        Load a tuned model.
        
        Args:
            path: Path to the model
            
        Returns:
            Loaded model
        """
        raise NotImplementedError("Subclasses must implement this method")


class SupervisedFineTuner(AgentTuner):
    """Tune agents using supervised fine-tuning."""
    
    def __init__(self):
        """Initialize the supervised fine-tuner."""
        super().__init__("supervised_fine_tuning")
    
    def tune(
        self, 
        model_name: str,
        trajectories: List[Trajectory],
        output_dir: str = "./tuned_model",
        num_train_epochs: int = 3,
        learning_rate: float = 5e-5,
        batch_size: int = 4,
        gradient_accumulation_steps: int = 4,
        max_seq_length: int = 512,
        format_type: str = "interleaved",
        positive_weight: float = 0.8,
        device: str = "cuda" if torch.cuda.is_available() else "cpu",
        **kwargs
    ) -> Tuple[Any, Dict[str, Any]]:
        """
        Tune a model using supervised fine-tuning.
        
        Args:
            model_name: Name of the base model
            trajectories: List of training trajectories
            output_dir: Directory to save the model
            num_train_epochs: Number of training epochs
            learning_rate: Learning rate
            batch_size: Batch size
            gradient_accumulation_steps: Gradient accumulation steps
            max_seq_length: Maximum sequence length
            format_type: Format type for trajectories
            positive_weight: Weight for positive examples
            device: Device to use for training
            **kwargs: Additional tuning parameters
            
        Returns:
            Tuple of (tuned_model, training_metrics)
        """
        print(f"Starting supervised fine-tuning of {model_name}")
        
        # Create output directory
        os.makedirs(output_dir, exist_ok=True)
        
        # Load model and tokenizer
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(model_name)
        
        # Ensure the tokenizer has a pad token
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token
        
        # Prepare training data
        print("Preparing training data...")
        
        # Separate positive and negative trajectories
        positive_trajectories = [t for t in trajectories if t.is_positive]
        negative_trajectories = [t for t in trajectories if not t.is_positive]
        
        print(f"Found {len(positive_trajectories)} positive and {len(negative_trajectories)} negative trajectories")
        
        # Calculate sample counts based on positive weight
        total_samples = len(trajectories)
        target_positive = int(total_samples * positive_weight)
        target_negative = total_samples - target_positive
        
        # Sample trajectories to achieve desired ratio
        if len(positive_trajectories) > target_positive:
            positive_trajectories = np.random.choice(positive_trajectories, target_positive, replace=False).tolist()
        
        if len(negative_trajectories) > target_negative:
            negative_trajectories = np.random.choice(negative_trajectories, target_negative, replace=False).tolist()
        
        # Combine trajectories
        sampled_trajectories = positive_trajectories + negative_trajectories
        np.random.shuffle(sampled_trajectories)
        
        print(f"Using {len(positive_trajectories)} positive and {len(negative_trajectories)} negative trajectories for training")
        
        # Format trajectories for training
        training_texts = []
        
        for trajectory in tqdm(sampled_trajectories, desc="Formatting trajectories"):
            formatted = trajectory.to_training_format(format_type)
            training_texts.append(formatted)
        
        # Tokenize training data
        def tokenize_function(examples):
            return tokenizer(
                examples["text"],
                padding="max_length",
                truncation=True,
                max_length=max_seq_length
            )
        
        # Create dataset
        dataset = Dataset.from_dict({"text": training_texts})
        tokenized_dataset = dataset.map(
            tokenize_function,
            batched=True,
            remove_columns=["text"]
        )
        
        # Set up training arguments
        training_args = TrainingArguments(
            output_dir=output_dir,
            num_train_epochs=num_train_epochs,
            per_device_train_batch_size=batch_size,
            gradient_accumulation_steps=gradient_accumulation_steps,
            learning_rate=learning_rate,
            weight_decay=0.01,
            save_strategy="epoch",
            save_total_limit=2,
            logging_dir=f"{output_dir}/logs",
            logging_steps=10,
            report_to="none"
        )
        
        # Create data collator
        data_collator = DataCollatorForLanguageModeling(
            tokenizer=tokenizer,
            mlm=False
        )
        
        # Create trainer
        trainer = Trainer(
            model=model,
            args=training_args,
            train_dataset=tokenized_dataset,
            data_collator=data_collator
        )
        
        # Train the model
        print("Starting training...")
        train_result = trainer.train()
        
        # Save the model
        print(f"Saving model to {output_dir}")
        trainer.save_model(output_dir)
        tokenizer.save_pretrained(output_dir)
        
        # Return the model and metrics
        metrics = {
            "train_loss": train_result.training_loss,
            "train_runtime": train_result.metrics["train_runtime"],
            "samples_per_second": train_result.metrics["train_samples_per_second"],
            "num_train_samples": len(tokenized_dataset)
        }
        
        return model, metrics
    
    def save_model(self, model: Any, path: str) -> None:
        """
        Save the tuned model.
        
        Args:
            model: Tuned model
            path: Path to save the model
        """
        model.save_pretrained(path)
    
    def load_model(self, path: str) -> Any:
        """
        Load a tuned model.
        
        Args:
            path: Path to the model
            
        Returns:
            Loaded model
        """
        return AutoModelForCausalLM.from_pretrained(path)


class ParameterEfficientFineTuner(AgentTuner):
    """Tune agents using parameter-efficient fine-tuning methods."""
    
    def __init__(self):
        """Initialize the parameter-efficient fine-tuner."""
        super().__init__("parameter_efficient_fine_tuning")
    
    def tune(
        self, 
        model_name: str,
        trajectories: List[Trajectory],
        output_dir: str = "./tuned_model",
        method: str = "lora",  # 'lora', 'prefix', 'prompt_tuning'
        num_train_epochs: int = 3,
        learning_rate: float = 1e-4,
        batch_size: int = 4,
        gradient_accumulation_steps: int = 4,
        max_seq_length: int = 512,
        format_type: str = "interleaved",
        positive_weight: float = 0.8,
        device: str = "cuda" if torch.cuda.is_available() else "cpu",
        **kwargs
    ) -> Tuple[Any, Dict[str, Any]]:
        """
        Tune a model using parameter-efficient methods.
        
        Args:
            model_name: Name of the base model
            trajectories: List of training trajectories
            output_dir: Directory to save the model
            method: PEFT method to use
            num_train_epochs: Number of training epochs
            learning_rate: Learning rate
            batch_size: Batch size
            gradient_accumulation_steps: Gradient accumulation steps
            max_seq_length: Maximum sequence length
            format_type: Format type for trajectories
            positive_weight: Weight for positive examples
            device: Device to use for training
            **kwargs: Additional tuning parameters
            
        Returns:
            Tuple of (tuned_model, training_metrics)
        """
        try:
            from peft import (
                get_peft_model, LoraConfig, PrefixTuningConfig, 
                PromptTuningConfig, TaskType, PeftModel
            )
        except ImportError:
            raise ImportError("PEFT library is required for parameter-efficient fine-tuning. Install it with 'pip install peft'.")
        
        print(f"Starting parameter-efficient fine-tuning of {model_name} using {method}")
        
        # Create output directory
        os.makedirs(output_dir, exist_ok=True)
        
        # Load model and tokenizer
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForCausalLM.from_pretrained(model_name)
        
        # Ensure the tokenizer has a pad token
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token
        
        # Configure PEFT method
        if method == "lora":
            peft_config = LoraConfig(
                task_type=TaskType.CAUSAL_LM,
                r=16,
                lora_alpha=32,
                lora_dropout=0.1,
                target_modules=["q_proj", "v_proj"]
            )
        elif method == "prefix":
            peft_config = PrefixTuningConfig(
                task_type=TaskType.CAUSAL_LM,
                num_virtual_tokens=20,
                prefix_projection=True
            )
        elif method == "prompt_tuning":
            peft_config = PromptTuningConfig(
                task_type=TaskType.CAUSAL_LM,
                num_virtual_tokens=20,
                tokenizer_name_or_path=model_name
            )
        else:
            raise ValueError(f"Unsupported PEFT method: {method}")
        
        # Create PEFT model
        model = get_peft_model(model, peft_config)
        model.print_trainable_parameters()
        
        # Prepare training data (same as SupervisedFineTuner)
        print("Preparing training data...")
        
        # Separate positive and negative trajectories
        positive_trajectories = [t for t in trajectories if t.is_positive]
        negative_trajectories = [t for t in trajectories if not t.is_positive]
        
        print(f"Found {len(positive_trajectories)} positive and {len(negative_trajectories)} negative trajectories")
        
        # Calculate sample counts based on positive weight
        total_samples = len(trajectories)
        target_positive = int(total_samples * positive_weight)
        target_negative = total_samples - target_positive
        
        # Sample trajectories to achieve desired ratio
        if len(positive_trajectories) > target_positive:
            positive_trajectories = np.random.choice(positive_trajectories, target_positive, replace=False).tolist()
        
        if len(negative_trajectories) > target_negative:
            negative_trajectories = np.random.choice(negative_trajectories, target_negative, replace=False).tolist()
        
        # Combine trajectories
        sampled_trajectories = positive_trajectories + negative_trajectories
        np.random.shuffle(sampled_trajectories)
        
        print(f"Using {len(positive_trajectories)} positive and {len(negative_trajectories)} negative trajectories for training")
        
        # Format trajectories for training
        training_texts = []
        
        for trajectory in tqdm(sampled_trajectories, desc="Formatting trajectories"):
            formatted = trajectory.to_training_format(format_type)
            training_texts.append(formatted)
        
        # Tokenize training data
        def tokenize_function(examples):
            return tokenizer(
                examples["text"],
                padding="max_length",
                truncation=True,
                max_length=max_seq_length
            )
        
        # Create dataset
        dataset = Dataset.from_dict({"text": training_texts})
        tokenized_dataset = dataset.map(
            tokenize_function,
            batched=True,
            remove_columns=["text"]
        )
        
        # Set up training arguments
        training_args = TrainingArguments(
            output_dir=output_dir,
            num_train_epochs=num_train_epochs,
            per_device_train_batch_size=batch_size,
            gradient_accumulation_steps=gradient_accumulation_steps,
            learning_rate=learning_rate,
            weight_decay=0.01,
            save_strategy="epoch",
            save_total_limit=2,
            logging_dir=f"{output_dir}/logs",
            logging_steps=10,
            report_to="none"
        )
        
        # Create data collator
        data_collator = DataCollatorForLanguageModeling(
            tokenizer=tokenizer,
            mlm=False
        )
        
        # Create trainer
        trainer = Trainer(
            model=model,
            args=training_args,
            train_dataset=tokenized_dataset,
            data_collator=data_collator
        )
        
        # Train the model
        print("Starting training...")
        train_result = trainer.train()
        
        # Save the model
        print(f"Saving model to {output_dir}")
        trainer.save_model(output_dir)
        tokenizer.save_pretrained(output_dir)
        
        # Return the model and metrics
        metrics = {
            "train_loss": train_result.training_loss,
            "train_runtime": train_result.metrics["train_runtime"],
            "samples_per_second": train_result.metrics["train_samples_per_second"],
            "num_train_samples": len(tokenized_dataset),
            "peft_method": method
        }
        
        return model, metrics
    
    def save_model(self, model: Any, path: str) -> None:
        """
        Save the tuned model.
        
        Args:
            model: Tuned model
            path: Path to save the model
        """
        model.save_pretrained(path)
    
  
(Content truncated due to size limit. Use line ranges to read in chunks)