- Add ConsensusSnapshot and ConsensusSlot classes for structured consensus state representation. - Introduce OnlineLanguageLearner for stateful online learning with experience replay and EWC regularization. - Create WorkspaceEvent and ActionEvent classes for serializable event types. - Develop InMemoryVectorMemoryStore for storing and querying workspace events. - Implement ActionPolicy to manage action selection and motor parameterization. - Establish a unified WorkspaceRuntime for managing the agent's lifecycle and interactions. - Enhance tests in test_smoke.py to cover new functionalities and ensure correctness.
105 lines
3.4 KiB
Python
105 lines
3.4 KiB
Python
"""
|
|
Continual learning utilities shared by interactive runtimes.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
import torch
|
|
import torch.nn.functional as F
|
|
|
|
from .language_model import (
|
|
EWCOptimizer,
|
|
ExperienceReplay,
|
|
ExpertPlasticity,
|
|
)
|
|
|
|
|
|
class OnlineLanguageLearner:
|
|
"""
|
|
Stateful online learner for chat/runtime usage.
|
|
|
|
Unlike the old ad hoc SGD update, this keeps replay, EWC regularization,
|
|
and periodic consolidation alive across the whole session.
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
model,
|
|
config,
|
|
tokenizer,
|
|
device: str = "cpu",
|
|
lr: float = 5e-3,
|
|
ewc_lambda: float = 0.05,
|
|
seq_len: int = 48,
|
|
replay_batch_size: int = 4,
|
|
replay_weight: float = 0.5,
|
|
consolidate_every: int = 20,
|
|
):
|
|
self.model = model
|
|
self.config = config
|
|
self.tokenizer = tokenizer
|
|
self.device = device
|
|
self.seq_len = seq_len
|
|
self.replay_batch_size = replay_batch_size
|
|
self.replay_weight = replay_weight
|
|
self.consolidate_every = consolidate_every
|
|
|
|
self.optimizer = EWCOptimizer(model, lr=lr, ewc_lambda=ewc_lambda)
|
|
self.replay_buffer = ExperienceReplay(capacity=500, seq_len=seq_len)
|
|
self.plasticity = ExpertPlasticity(num_experts=config.num_experts)
|
|
self.step_count = 0
|
|
|
|
def _prepare_sequence(self, text: str) -> torch.Tensor | None:
|
|
token_ids = self.tokenizer.encode(text)
|
|
if len(token_ids) < 4:
|
|
return None
|
|
if len(token_ids) < self.seq_len:
|
|
token_ids = token_ids * (self.seq_len // len(token_ids) + 1)
|
|
token_ids = token_ids[:self.seq_len]
|
|
return torch.tensor([token_ids], dtype=torch.long, device=self.device)
|
|
|
|
def learn_text(self, text: str) -> dict | None:
|
|
token_seq = self._prepare_sequence(text)
|
|
if token_seq is None:
|
|
return None
|
|
|
|
self.model.train()
|
|
logits, info = self.model(token_seq)
|
|
pred = logits[:, :-1]
|
|
target = token_seq[:, 1:]
|
|
loss = F.cross_entropy(
|
|
pred.reshape(-1, self.config.vocab_size),
|
|
target.reshape(-1),
|
|
)
|
|
|
|
replay_loss = torch.tensor(0.0, device=self.device)
|
|
replay_seq = self.replay_buffer.sample(self.replay_batch_size)
|
|
if replay_seq is not None:
|
|
replay_seq = replay_seq.to(self.device)
|
|
replay_logits, _ = self.model(replay_seq)
|
|
replay_pred = replay_logits[:, :-1]
|
|
replay_target = replay_seq[:, 1:]
|
|
replay_loss = F.cross_entropy(
|
|
replay_pred.reshape(-1, self.config.vocab_size),
|
|
replay_target.reshape(-1),
|
|
)
|
|
|
|
total_task_loss = loss + self.replay_weight * replay_loss
|
|
total_loss = self.optimizer.step(total_task_loss)
|
|
|
|
self.plasticity.update(info["alpha"].detach(), token_seq.detach())
|
|
self.replay_buffer.push(token_seq.detach().cpu())
|
|
self.step_count += 1
|
|
|
|
if self.step_count % self.consolidate_every == 0:
|
|
self.optimizer.consolidate(token_seq.detach(), n_samples=10)
|
|
|
|
self.model.eval()
|
|
return {
|
|
"loss": float(loss.item()),
|
|
"replay_loss": float(replay_loss.item()),
|
|
"total_loss": float(total_loss),
|
|
"w_norm_mean": float(info["w_norm"].mean().item()),
|
|
"usage": self.plasticity.usage.tolist(),
|
|
"step": self.step_count,
|
|
}
|