Witllm/wit/lit_module.py

from functools import cache
from typing import Dict, Optional

import pytorch_lightning as pl
import torch
import torchmetrics

from modeling_wit import QWenLMHeadModel
from wit.configuration import ModelConfig

from transformers import AutoConfig


class LitModule(pl.LightningModule):
    def __init__(
        self,
        pretrained_model_dir: str = None,
        learning_rate: float = 0.0001,
        config: ModelConfig = None,
        use_tril_attention_mask: str = False,
    ):
        super().__init__()
        self.save_hyperparameters()
        if config == None:
            config = ModelConfig()
        model = QWenLMHeadModel(config)
        if pretrained_model_dir != None:
            from modelscope import snapshot_download

            model = model.from_pretrained(snapshot_download(pretrained_model_dir))
        self.llm = self.register_core_module(model)
        self.learning_rate = learning_rate
        self.use_tril_attention_mask = use_tril_attention_mask
        self.metric_loss = torchmetrics.MeanMetric()
        self.vocab_size = self.llm.config.vocab_size
        self.metric_accuracy = torchmetrics.Accuracy(
            task="multiclass",
            num_classes=self.vocab_size,
        )

    @cache
    def get_batch_tril_matrix(self, block_size: int, batch_size: Optional[int] = None) -> torch.Tensor:
        matrix = torch.ones(block_size, block_size).tril()
        if batch_size is not None:
            matrix = matrix.repeat(batch_size, 1, 1)
        return matrix

    def register_core_module(self, module: torch.nn.Module) -> torch.nn.Module:
        object.__setattr__(self, "__core_module__", module)
        return module

    def training_step(self, batch: Dict[str, torch.Tensor], batch_idx):
        batch_size, block_size = batch["input_ids"].shape
        if self.use_tril_attention_mask:
            batch["attention_mask"] = self.get_batch_tril_matrix(block_size, batch_size=batch_size).to(self.device)
        outputs, loss = self.llm(**batch)
        self.log("train_loss", loss, rank_zero_only=True)
        return loss

    def validation_step(self, batch: Dict[str, torch.Tensor], batch_idx):
        outputs, loss = self.llm(**batch, return_dict=True)
        logits = outputs[..., :-1, :]
        logits = logits.contiguous().view(-1, logits.size(-1))
        labels = batch["labels"][..., 1:]
        labels = labels.contiguous().view(-1)
        label_mask = labels < self.vocab_size
        logits_m = logits[label_mask]
        labels_m = labels[label_mask]
        # m = torch.max(logits, 1).indices.cpu().numpy()
        # ll = labels.cpu().numpy()
        self.metric_accuracy.update(logits_m, labels_m)
        self.metric_loss.update(loss)

    def on_validation_epoch_end(self) -> None:
        self.log("val_loss", self.metric_loss, rank_zero_only=True)
        self.log("accuracy", self.metric_accuracy, rank_zero_only=True)

    def configure_optimizers(self):
        optimizer = torch.optim.AdamW(self.trainer.model.parameters(), lr=self.learning_rate)
        return optimizer

    def configure_callbacks(self):
        checkpoint_callback = pl.callbacks.ModelCheckpoint(
            monitor="accuracy",
            mode="max",
            filename="{epoch:02d}-{accuracy:.4f}",
        )
        early_stop_callback = pl.callbacks.EarlyStopping(
            monitor="accuracy",
            min_delta=0.001,
            patience=3,
            mode="max",
            stopping_threshold=1,
        )
        lr_monitor = pl.callbacks.LearningRateMonitor(logging_interval="step")
        return [lr_monitor]
        # return [checkpoint_callback, lr_monitor]
        # return [checkpoint_callback, early_stop_callback]