Spaces:
Sleeping
Sleeping
File size: 4,103 Bytes
c1e438c | 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 | import torch
import traceback
from core.device import DEVICE
from emotion.tracker import EmotionTracker
from reasoning.persona import PersonaController
class MVIState:
def __init__(self):
self.modules = {}
self.health = {}
def register(self, name, module, status=True):
self.modules[name] = module
self.health[name] = status
def get(self, name):
return self.modules.get(name)
def report(self):
return self.health
def safe_load(name, loader_fn, state):
"""
Attempts to load a module and updates state health.
"""
try:
module = loader_fn()
state.register(name, module, True)
print(f"[LOAD SUCCESS] {name}")
except Exception:
traceback.print_exc()
state.register(name, None, False)
print(f"[LOAD FAILED] {name}")
def initialize_mvi():
"""
Central MVI loader — loads all modules (language, emotion, vision, voice, memory)
and registers them into MVIState with health tracking.
"""
from language.tokenizer import SimpleTokenizer
from language.embeddings import EmbeddingLayer
from language.encoder import SentenceEncoder
from language.intent import IntentClassifier
from emotion.sentiment_model import SentimentRegressor
from vision.image_encoder import ImageEncoder
from vision.video_encoder import VideoEncoder
from core.voice_encoder import VoiceEncoder
from memory.short_term import ShortTermMemory
from memory.long_term import LongTermMemory
state = MVIState()
# ---------- EMOTION TRACKER & PERSONA ----------
state.register("emotion_tracker", EmotionTracker(window_size=6))
state.register("persona", PersonaController())
# ---------- LANGUAGE ----------
def load_language():
tokenizer = SimpleTokenizer()
tokenizer.load_vocab("artifacts/vocab.json")
embedder = EmbeddingLayer(
tokenizer.vocab_size,
pad_index=tokenizer.vocab[tokenizer.PAD_TOKEN]
).to(DEVICE)
encoder = SentenceEncoder().to(DEVICE)
encoder.load_state_dict(torch.load("artifacts/sentence_encoder.pt", map_location=DEVICE))
intent = IntentClassifier(
input_dim=encoder.projection.out_features,
intent_labels=["question", "advice", "statement"]
).to(DEVICE)
intent.load_state_dict(torch.load("artifacts/intent_classifier.pt", map_location=DEVICE))
return {
"tokenizer": tokenizer,
"embedder": embedder.eval(),
"encoder": encoder.eval(),
"intent": intent.eval()
}
safe_load("language", load_language, state)
# ---------- EMOTION ----------
def load_emotion():
model = SentimentRegressor(input_dim=128).to(DEVICE)
model.load_state_dict(torch.load("artifacts/sentiment_regressor.pt", map_location=DEVICE))
return model.eval()
safe_load("emotion", load_emotion, state)
# ---------- VISION ----------
def load_image():
model = ImageEncoder(embed_dim=128).to(DEVICE)
model.load_state_dict(torch.load("artifacts/image_encoder.pt", map_location=DEVICE))
return model.eval()
safe_load("vision_image", load_image, state)
def load_video():
model = VideoEncoder(embed_dim=128).to(DEVICE)
model.load_state_dict(torch.load("artifacts/video_encoder.pt", map_location=DEVICE))
return model.eval()
safe_load("vision_video", load_video, state)
# ---------- VOICE ----------
def load_voice():
model = VoiceEncoder(embed_dim=128).to(DEVICE)
model.load_state_dict(torch.load("artifacts/voice_encoder_best.pt", map_location=DEVICE))
return model.eval()
safe_load("voice", load_voice, state)
# ---------- MEMORY ----------
def load_memory():
return {
"stm": ShortTermMemory(max_len=10),
"ltm": LongTermMemory(dim=128)
}
safe_load("memory", load_memory, state)
print("\n[MVI SYSTEM HEALTH]")
print(state.report())
return state
|