File size: 18,210 Bytes
c8b77b5 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 |
"""
Empathy Training Module for MangoMAS Local
This module implements specialized training for empathy and emotional intelligence,
adapted from the AWS backup system for local training.
"""
import json
import logging
import os
import random
from typing import Any, Dict, List
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import Dataset
from ..core_framework import SpecializedTrainingModule, TrainingModuleConfig
logger = logging.getLogger(__name__)
class EmpathyDataset(Dataset):
"""Dataset for training empathy and emotional intelligence capabilities."""
def __init__(self, data_path: str, tokenizer, max_length: int = 768):
"""
Initialize the empathy dataset.
Args:
data_path: Path to the empathy data file
tokenizer: Tokenizer for text processing
max_length: Maximum sequence length
"""
self.tokenizer = tokenizer
self.max_length = max_length
self.data = self._load_data(data_path)
logger.info(f"Loaded empathy dataset with {len(self.data)} examples")
def _load_data(self, data_path: str) -> List[Dict]:
"""Load empathy training data."""
data = []
with open(data_path, "r", encoding="utf-8") as f:
for line in f:
try:
item = json.loads(line.strip())
# Validate required fields for empathy data
if (
"user_message" in item
and "emotional_state" in item
and "empathetic_response" in item
):
data.append(item)
except (json.JSONDecodeError, KeyError) as e:
logger.warning(f"Skipping invalid empathy data: {e}")
return data
def __len__(self) -> int:
"""Return the number of examples in the dataset."""
return len(self.data)
def __getitem__(self, idx: int) -> Dict[str, torch.Tensor]:
"""Get a training example."""
item = self.data[idx]
# Format the empathy example
user_message = item["user_message"]
emotional_state = item["emotional_state"]
empathetic_response = item["empathetic_response"]
# Additional fields if available
emotional_cues = item.get("emotional_cues", [])
context = item.get("context", "")
# Construct the text with empathy markers
text = f"User: {user_message}\n\n"
# Include emotional analysis section for training
text += f"Emotional State: {emotional_state}\n"
if emotional_cues:
text += "Emotional Cues:\n"
for i, cue in enumerate(emotional_cues):
text += f"{i+1}. {cue}\n"
text += "\n"
if context:
text += f"Context: {context}\n\n"
text += f"Empathetic Response: {empathetic_response}"
# Tokenize
encoding = self.tokenizer(
text,
truncation=True,
padding="max_length",
max_length=self.max_length,
return_tensors="pt",
)
return {
"input_ids": encoding["input_ids"].squeeze(),
"attention_mask": encoding["attention_mask"].squeeze(),
"labels": encoding["input_ids"].squeeze().clone(),
"user_message": user_message,
"emotional_state": emotional_state,
"empathetic_response": empathetic_response,
}
class EmpathyEvaluator:
"""Evaluator for empathy and emotional intelligence capabilities."""
def __init__(self, tokenizer):
"""
Initialize the empathy evaluator.
Args:
tokenizer: Tokenizer for text processing
"""
self.tokenizer = tokenizer
self.metrics = {
"emotional_recognition": 0.0,
"empathetic_language": 0.0,
"supportive_tone": 0.0,
"personalization": 0.0,
}
# Empathetic language markers
self.empathetic_phrases = [
"understand",
"feel",
"appreciate",
"recognize",
"acknowledge",
"must be",
"sounds like",
"seems like",
"I hear you",
"that's difficult",
"that's challenging",
"I'm sorry",
"thank you for sharing",
"I can imagine",
]
# Emotional state categories
self.emotional_states = {
"positive": [
"happy",
"excited",
"grateful",
"proud",
"hopeful",
"inspired",
],
"negative": [
"sad",
"angry",
"frustrated",
"anxious",
"disappointed",
"overwhelmed",
],
"neutral": [
"confused",
"uncertain",
"curious",
"surprised",
"contemplative",
],
}
def evaluate(self, model, eval_dataset: EmpathyDataset) -> Dict[str, float]:
"""
Evaluate empathy capabilities on the provided dataset.
Args:
model: The model to evaluate
eval_dataset: Dataset of empathy examples
Returns:
Dictionary of evaluation metrics
"""
model.eval()
device = next(model.parameters()).device
# Reset metrics
for key in self.metrics:
self.metrics[key] = 0.0
total_examples = min(
len(eval_dataset), 50
) # Limit to 50 examples for efficiency
with torch.no_grad():
for idx in range(total_examples):
example = eval_dataset[idx]
user_message = example["user_message"]
expected_emotional_state = example["emotional_state"]
# Generate response without providing emotional state
prompt = f"User: {user_message}\n\nProvide an empathetic response:"
input_ids = self.tokenizer(prompt, return_tensors="pt").input_ids.to(
device
)
generated_ids = model.generate(
input_ids, max_length=256, temperature=0.7, num_return_sequences=1
)
generated_text = self.tokenizer.decode(
generated_ids[0], skip_special_tokens=True
)
# Evaluate empathy quality
self._evaluate_empathy(
user_message=user_message,
expected_emotional_state=expected_emotional_state,
expected_response=example["empathetic_response"],
generated_response=generated_text,
)
# Calculate averages
for key in self.metrics:
self.metrics[key] /= total_examples
return self.metrics
def _evaluate_empathy(
self,
user_message: str,
expected_emotional_state: str,
expected_response: str,
generated_response: str,
) -> None:
"""
Evaluate empathy quality for a specific example.
Args:
user_message: The user's message
expected_emotional_state: Expected identified emotional state
expected_response: Expected empathetic response
generated_response: The response generated by the model
"""
# 1. Emotional recognition - check if response acknowledges correct emotion
emotional_category = None
for category, emotions in self.emotional_states.items():
if any(emotion in expected_emotional_state.lower() for emotion in emotions):
emotional_category = category
break
if emotional_category:
# Check if response contains words matching the emotional category
emotion_words = self.emotional_states[emotional_category]
emotion_recognition = any(
word in generated_response.lower() for word in emotion_words
)
self.metrics["emotional_recognition"] += 1.0 if emotion_recognition else 0.0
else:
# Default partial score if we couldn't categorize
self.metrics["emotional_recognition"] += 0.5
# 2. Empathetic language - check for empathetic phrases
empathy_phrase_count = sum(
1
for phrase in self.empathetic_phrases
if phrase in generated_response.lower()
)
self.metrics["empathetic_language"] += min(1.0, empathy_phrase_count / 2)
# 3. Supportive tone - simplified check for supportive language
supportive_score = 0.0
if (
"here for you" in generated_response.lower()
or "support" in generated_response.lower()
):
supportive_score += 0.5
if (
"help" in generated_response.lower()
or "advice" in generated_response.lower()
):
supportive_score += 0.3
if any(
phrase in generated_response.lower()
for phrase in ["let me know", "is there anything", "can i"]
):
supportive_score += 0.2
self.metrics["supportive_tone"] += min(1.0, supportive_score)
# 4. Personalization - check if response refers to specific details from user message
user_specific_terms = set(user_message.lower().split()) - {
"i",
"me",
"my",
"mine",
"am",
"was",
"the",
"a",
"an",
}
generated_terms = set(generated_response.lower().split())
specific_term_overlap = len(user_specific_terms.intersection(generated_terms))
self.metrics["personalization"] += min(1.0, specific_term_overlap / 3)
class EmpathyTrainingModule(SpecializedTrainingModule):
"""Specialized training module for empathy and emotional intelligence capabilities."""
def __init__(self, config: TrainingModuleConfig, tokenizer):
"""
Initialize the empathy training module.
Args:
config: Module configuration
tokenizer: Tokenizer for text processing
"""
super().__init__(config, tokenizer)
# Initialize empathy-specific components
self.data_path = config.data_path or "data/processed/empathy_train.jsonl"
self.evaluator = EmpathyEvaluator(tokenizer)
# Empathy-specific loss
self.empathy_loss = nn.CrossEntropyLoss(ignore_index=-100)
# Training metrics
self.metrics = {
"empathy_loss": 0.0,
"emotion_recognition_rate": 0.0,
"empathetic_language_score": 0.0,
}
logger.info("Initialized empathy training module")
def prepare_batch(self, batch: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]:
"""
Prepare a batch of data for empathy training.
Args:
batch: The input batch from the dataloader
Returns:
Processed batch ready for empathy training
"""
# Extract empathy-specific elements if they exist
if all(
key in batch
for key in ["user_message", "emotional_state", "empathetic_response"]
):
# This is already an empathy-specific batch
return batch
# For general conversation batches, we need to identify emotional content
# This is a simplified placeholder implementation
return batch
def compute_loss(
self, student_outputs: Any, teacher_outputs: Any, batch: Dict[str, torch.Tensor]
) -> torch.Tensor:
"""
Compute the empathy-specific loss.
Args:
student_outputs: Outputs from the student model
teacher_outputs: Outputs from the teacher model
batch: The processed input batch
Returns:
Empathy-specific loss tensor
"""
# Get logits from outputs
student_logits = (
student_outputs.logits
if hasattr(student_outputs, "logits")
else student_outputs
)
teacher_logits = (
teacher_outputs.logits
if hasattr(teacher_outputs, "logits")
else teacher_outputs
)
# Standard distillation loss calculation
student_logits = student_logits[:, :-1, :].contiguous()
teacher_logits = teacher_logits[:, :-1, :].contiguous()
target_ids = batch["labels"][:, 1:].contiguous()
# KL divergence loss for distillation
temperature = 2.0
kl_loss = F.kl_div(
F.log_softmax(student_logits / temperature, dim=-1),
F.softmax(teacher_logits / temperature, dim=-1),
reduction="batchmean",
) * (temperature**2)
# Cross-entropy loss against labels
ce_loss = self.empathy_loss(
student_logits.view(-1, student_logits.size(-1)), target_ids.view(-1)
)
# Combined loss with balanced weighting
loss = 0.5 * ce_loss + 0.5 * kl_loss
# Update metrics
self.metrics["empathy_loss"] = loss.item()
return loss
def get_metrics(self) -> Dict[str, float]:
"""
Get metrics specific to empathy training.
Returns:
Dictionary of metric names and values
"""
return self.metrics
def generate_synthetic_empathy_data(
self, output_path: str, num_samples: int = 1000
) -> None:
"""
Generate synthetic empathy training data.
Args:
output_path: Path to save the generated data
num_samples: Number of samples to generate
"""
# This is a simplified implementation based on the AWS backup
# In a full implementation, this would be much more sophisticated
empathy_templates = [
{
"user_message": "I just lost my job after 5 years at the company. I'm feeling really lost and worried about how I'll pay my bills.",
"emotional_state": "anxious and worried",
"emotional_cues": [
"Lost job unexpectedly",
"Financial insecurity",
"Identity/purpose uncertainty",
"Use of 'lost' to describe feelings",
],
"empathetic_response": "I'm really sorry to hear about losing your job, especially after dedicating 5 years to the company. That's incredibly difficult. It's completely understandable to feel lost and worried about your financial situation right now. Sudden changes like this can be overwhelming. Would it help to talk through some immediate steps you might take, or would you prefer to just share how you're processing this right now?",
"context": "Professional setback, financial stress",
},
{
"user_message": "My dog passed away yesterday. He was with me for 12 years and I don't know how to handle this emptiness in my home.",
"emotional_state": "grief and sadness",
"emotional_cues": [
"Loss of longtime companion",
"Reference to 'emptiness'",
"Uncertainty about coping",
"Recent timing (yesterday)",
],
"empathetic_response": "I'm so sorry about the loss of your dog. After 12 years together, they're truly family, and that leaves such a profound emptiness. The love you shared for so many years speaks to what a caring person you are and what a special connection you had. There's no timeline for grief, especially for a companion who was part of your daily life for so long. Would it help to talk about some of your favorite memories of them?",
"context": "Pet loss, grief",
},
{
"user_message": "I just got accepted to my dream university! I've been working toward this for years and can't believe it finally happened.",
"emotional_state": "excited and proud",
"emotional_cues": [
"Exclamation mark showing excitement",
"Achievement of long-term goal",
"Expression of disbelief at success",
"Years of effort mentioned",
],
"empathetic_response": "Congratulations! That's absolutely wonderful news about your acceptance to your dream university! Years of hard work and dedication have paid off, and you have every right to feel excited and proud of this achievement. It's such a special feeling when a goal you've worked toward for so long finally becomes reality. I'd love to hear more about the university and what you're most looking forward to as you start this new chapter!",
"context": "Academic achievement, celebration",
},
]
# Generate variations
output_data = []
for _ in range(num_samples):
template = random.choice(empathy_templates)
# Create a variation to avoid exact duplicates
variation = template.copy()
# Add metadata
variation["metadata"] = {
"generated": True,
"timestamp": (
torch.cuda.get_device_name(0)
if torch.cuda.is_available()
else "CPU"
),
}
output_data.append(variation)
# Save to file
os.makedirs(os.path.dirname(output_path), exist_ok=True)
with open(output_path, "w", encoding="utf-8") as f:
for item in output_data:
f.write(json.dumps(item) + "\n")
logger.info(
f"Generated {len(output_data)} synthetic empathy examples at {output_path}"
)
|