reference/code/SpecForge/scripts/train_dflash.py

#!/usr/bin/env python3
# coding=utf-8
"""DFlash Training Script."""

import argparse
import functools
import logging
import math
import os
import shutil
import time
import warnings
from typing import Optional, Tuple

import torch
import torch.distributed as dist
from accelerate.utils import set_seed
from torch.distributed.fsdp import BackwardPrefetch
from torch.distributed.fsdp import FullyShardedDataParallel as FSDP
from torch.distributed.fsdp import MixedPrecision, ShardingStrategy, StateDictType
from torch.distributed.fsdp.wrap import transformer_auto_wrap_policy
from torch.utils.data import DataLoader
from tqdm import tqdm
from transformers import AutoConfig

from datasets import load_dataset
from specforge.args import SGLangBackendArgs, TrackerArgs
from specforge.core.dflash import OnlineDFlashModel
from specforge.data import build_eagle3_dataset, prepare_dp_dataloaders
from specforge.distributed import destroy_distributed, get_dp_group, init_distributed
from specforge.modeling.draft.dflash import DFlashDraftModel
from specforge.modeling.target.dflash_target_model import (
    DFlashTargetModel,
    get_dflash_target_model,
)
from specforge.modeling.target.target_utils import TargetEmbeddingsAndHead
from specforge.optimizer import BF16Optimizer
from specforge.tracker import create_tracker
from specforge.utils import (
    get_last_checkpoint,
    get_local_device,
    load_tokenizer,
    print_on_rank0,
    print_with_rank,
)


def parse_args():
    parser = argparse.ArgumentParser(description="Train DFlash Draft Model")

    model_group = parser.add_argument_group("model")
    model_group.add_argument("--target-model-path", type=str, required=True)
    model_group.add_argument(
        "--target-model-backend",
        type=str,
        default="hf",
        choices=["sglang", "hf"],
        help="Backend for target model: 'sglang' (service) or 'hf' (local)",
    )
    model_group.add_argument("--draft-config-path", type=str, default=None)
    model_group.add_argument("--block-size", type=int, default=16)
    model_group.add_argument("--num-draft-layers", type=int, default=1)
    model_group.add_argument(
        "--mask-token-id",
        type=int,
        default=None,
        help="MASK token ID. If not provided, auto-detect from tokenizer.",
    )
    model_group.add_argument(
        "--attention-backend",
        type=str,
        default="flex_attention",
        choices=["eager", "sdpa", "flex_attention"],
        help="Attention backend for draft model.",
    )
    model_group.add_argument(
        "--trust-remote-code", action="store_true", help="Trust remote code"
    )
    model_group.add_argument(
        "--num-anchors",
        type=int,
        default=512,
        help="Number of anchor positions per sequence",
    )
    model_group.add_argument(
        "--loss-decay-gamma",
        type=float,
        default=None,
        help="Gamma for exponential loss decay weighting (paper Eq.4). "
        "Suggested: 7 for block_size=16, 5 for 10, 4 for 8. None disables. "
        "Only applies when --loss-type dflash.",
    )
    model_group.add_argument(
        "--loss-type",
        type=str,
        default=None,
        choices=[
            "dflash",
            "vp_drafter",
            "dpace",
            "dpace-cumulative-confidence-only",
            "dpace-continuation-value-only",
        ],
        help=(
            "Training objective. If omitted, reads dflash_config.training_mode or "
            "dflash_config.loss_type from the draft config, defaulting to dflash."
        ),
    )
    model_group.add_argument(
        "--dpace-alpha",
        type=float,
        default=0.5,
        help="Smoothing alpha for D-PACE position weights.",
    )
    model_group.add_argument(
        "--prefix-weight-base",
        type=float,
        default=None,
        help=(
            "VP-Drafter prefix length sampling base. Values below 1 prefer shorter "
            "visible prefixes; defaults to dflash_config.prefix_weight_base or 0.9."
        ),
    )
    model_group.add_argument(
        "--embedding-key",
        type=str,
        default=None,
        help="Embedding weight key in the target model. "
        "Default: 'model.embed_tokens.weight' for standard models, "
        "'model.language_model.embed_tokens.weight' for multimodal models like Qwen3.5-A3B.",
    )
    model_group.add_argument(
        "--lm-head-key",
        type=str,
        default=None,
        help="LM head weight key in the target model. Default: 'lm_head.weight'.",
    )

    dataset_group = parser.add_argument_group("dataset")
    dataset_group.add_argument("--train-data-path", type=str, required=True)
    dataset_group.add_argument("--eval-data-path", type=str, default=None)
    dataset_group.add_argument("--chat-template", type=str, default="qwen")
    dataset_group.add_argument("--is-preformatted", action="store_true")
    dataset_group.add_argument("--dataloader-num-workers", type=int, default=8)
    dataset_group.add_argument(
        "--build-dataset-num-proc",
        type=int,
        default=int(os.environ.get("SPECFORGE_DATA_NUM_PROC", 8)),
    )

    training_group = parser.add_argument_group("training")
    training_group.add_argument("--num-epochs", type=int, default=6)
    training_group.add_argument("--batch-size", type=int, default=1)
    training_group.add_argument("--learning-rate", type=float, default=6e-4)
    training_group.add_argument("--max-length", type=int, default=3072)
    training_group.add_argument("--warmup-ratio", type=float, default=0.04)
    training_group.add_argument("--max-grad-norm", type=float, default=1.0)
    training_group.add_argument("--accumulation-steps", type=int, default=1)
    training_group.add_argument("--seed", type=int, default=42)
    training_group.add_argument("--resume", action="store_true")

    output_group = parser.add_argument_group("output")
    output_group.add_argument("--output-dir", type=str, required=True)
    output_group.add_argument("--cache-dir", type=str, default="./cache")
    output_group.add_argument("--log-interval", type=int, default=50)
    output_group.add_argument("--eval-interval", type=int, default=1000)
    output_group.add_argument("--save-interval", type=int, default=1000)

    optimization_group = parser.add_argument_group("optimization")
    optimization_group.add_argument(
        "--tp-size",
        type=int,
        default=1,
        help="The size of the tensor parallel for the target model",
    )

    tracker_group = parser.add_argument_group("tracker")
    TrackerArgs.add_args(tracker_group)

    dist_group = parser.add_argument_group("distributed")
    dist_group.add_argument("--dist-timeout", type=int, default=30)

    # SGLang specific args
    sglang_group = parser.add_argument_group("sglang backend")
    SGLangBackendArgs.add_args(sglang_group)

    return parser.parse_args()


def build_models(args) -> Tuple[DFlashTargetModel, DFlashDraftModel]:
    """Build target model (backend wrapper) and draft model."""
    print_on_rank0(
        f"Loading target model from {args.target_model_path} using {args.target_model_backend} backend"
    )

    target_model_kwargs = {}
    if args.target_model_backend == "sglang":
        target_model_kwargs = SGLangBackendArgs.from_args(args).to_kwargs()

    device = get_local_device()
    device_type = device.type

    target_model = get_dflash_target_model(
        pretrained_model_name_or_path=args.target_model_path,
        backend=args.target_model_backend,
        torch_dtype=torch.bfloat16,
        device=device_type if args.target_model_backend == "hf" else None,
        trust_remote_code=args.trust_remote_code,
        **target_model_kwargs,
    )

    if args.draft_config_path:
        draft_config = AutoConfig.from_pretrained(args.draft_config_path)
        print_on_rank0(f"Loaded draft config from {args.draft_config_path}")
        # Warn if command-line args differ from config
        if (
            hasattr(draft_config, "block_size")
            and draft_config.block_size != args.block_size
        ):
            print_on_rank0(
                f"Warning: checkpoint block_size ({draft_config.block_size}) differs from "
                f"command-line arg ({args.block_size}). Using checkpoint value."
            )
    else:
        target_config = AutoConfig.from_pretrained(args.target_model_path)
        draft_config = AutoConfig.from_pretrained(args.target_model_path)
        draft_config.num_hidden_layers = args.num_draft_layers
        draft_config.block_size = args.block_size
        draft_config.num_target_layers = target_config.num_hidden_layers
        print_on_rank0("Auto-generated draft config from target model")

    if not hasattr(draft_config, "dflash_config") or draft_config.dflash_config is None:
        draft_config.dflash_config = {}

    args.loss_type = (
        args.loss_type
        or draft_config.dflash_config.get("training_mode")
        or draft_config.dflash_config.get("loss_type")
        or "dflash"
    )
    if args.prefix_weight_base is None:
        args.prefix_weight_base = draft_config.dflash_config.get(
            "prefix_weight_base", 0.9
        )

    draft_config._attn_implementation = args.attention_backend
    print_on_rank0(f"Using attention backend: {args.attention_backend}")
    print_on_rank0(f"Using DFlash training loss_type: {args.loss_type}")

    draft_model = DFlashDraftModel(draft_config).to(device=device, dtype=torch.bfloat16)

    target_model.set_capture_layers(draft_model.target_layer_ids)

    print_on_rank0(
        f"Draft config: block_size={draft_config.block_size}, "
        f"num_hidden_layers={draft_config.num_hidden_layers}, "
        f"num_target_layers={draft_config.num_target_layers}"
    )
    print_on_rank0(
        f"Draft model parameters: {sum(p.numel() for p in draft_model.parameters()):,}"
    )

    return target_model, draft_model


def build_dataloader(args, tokenizer) -> Tuple[DataLoader, Optional[DataLoader]]:
    """Build train and eval dataloaders."""
    import hashlib

    cache_params_string = (
        f"{args.train_data_path}-"
        f"{args.max_length}-"
        f"{args.chat_template}-"
        f"{args.target_model_path}"
    )
    cache_key = hashlib.md5(cache_params_string.encode()).hexdigest()

    train_dataset = load_dataset("json", data_files=args.train_data_path)["train"]
    train_eagle3_dataset = build_eagle3_dataset(
        dataset=train_dataset,
        tokenizer=tokenizer,
        chat_template=args.chat_template,
        max_length=args.max_length,
        is_preformatted=args.is_preformatted,
        cache_dir=os.path.join(args.cache_dir, "processed_dataset"),
        cache_key=cache_key,
        num_proc=args.build_dataset_num_proc,
    )

    min_loss_tokens = 2 * args.block_size
    original_size = len(train_eagle3_dataset)
    train_eagle3_dataset = train_eagle3_dataset.filter(
        lambda x: x["loss_mask"].sum() >= min_loss_tokens
    )
    print_on_rank0(
        f"Filtered train dataset: {original_size} -> {len(train_eagle3_dataset)} samples"
    )

    train_dataloader = prepare_dp_dataloaders(
        train_eagle3_dataset,
        args.batch_size,
        num_workers=args.dataloader_num_workers,
        shuffle=True,
        process_group=get_dp_group(),
    )

    eval_dataloader = None
    if args.eval_data_path:
        eval_dataset = load_dataset("json", data_files=args.eval_data_path)["train"]
        eval_eagle3_dataset = build_eagle3_dataset(
            dataset=eval_dataset,
            tokenizer=tokenizer,
            chat_template=args.chat_template,
            max_length=args.max_length,
            is_preformatted=args.is_preformatted,
        )
        eval_dataloader = prepare_dp_dataloaders(
            eval_eagle3_dataset,
            args.batch_size,
            num_workers=args.dataloader_num_workers,
            shuffle=False,
            process_group=get_dp_group(),
        )

    return train_dataloader, eval_dataloader


def save_checkpoint(args, epoch, step, dflash_model, draft_model, optimizer):
    """Save checkpoint."""
    save_dir = os.path.join(args.output_dir, f"epoch_{epoch}_step_{step}")
    if dist.get_rank() == 0:
        os.makedirs(save_dir, exist_ok=True)
    dist.barrier()

    with FSDP.state_dict_type(dflash_model, StateDictType.FULL_STATE_DICT):
        state_dict = dflash_model.state_dict()
        draft_state_dict = {
            k.replace("draft_model.", ""): v
            for k, v in state_dict.items()
            if "draft_model." in k
        }

        if dist.get_rank() == 0:
            torch.save(
                {
                    "epoch": epoch,
                    "global_step": step,
                    "args": args,
                    **optimizer.state_dict(),
                },
                os.path.join(save_dir, "training_state.pt"),
            )

            draft_model.save_pretrained(save_dir, state_dict=draft_state_dict)

            modeling_src = os.path.join(
                os.path.dirname(__file__),
                "..",
                "specforge",
                "modeling",
                "draft",
                "dflash.py",
            )
            modeling_dst = os.path.join(save_dir, "dflash.py")
            if os.path.exists(modeling_src):
                shutil.copy(modeling_src, modeling_dst)

            print_on_rank0(f"Saved checkpoint to {save_dir}")

    dist.barrier()


def record_metrics(
    args,
    loss: float,
    accuracy: float,
    global_step: int,
    tracker,
    optimizer,
    train_dataloader=None,
    mode: str = "train",
) -> None:
    logdict = {}

    if mode == "train" and optimizer is not None:
        logdict["train/lr"] = optimizer.get_learning_rate()

    logdict[f"{mode}/loss"] = loss
    logdict[f"{mode}/accuracy"] = accuracy

    print_on_rank0(
        f"{mode.capitalize()} - Step {global_step} [{global_step}/{args.num_epochs * len(train_dataloader) // args.accumulation_steps}?], Loss: {loss:.4f}, Acc: {accuracy:.4f}"
    )

    tracker.log(logdict, step=global_step)


def main():

    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO,
    )
    logging.getLogger().setLevel(logging.INFO)
    warnings.filterwarnings(
        "ignore",
        "The .grad attribute of a Tensor that is not a leaf Tensor is being accessed",
    )

    args = parse_args()
    set_seed(args.seed)

    init_distributed(timeout=args.dist_timeout, tp_size=args.tp_size)
    print_with_rank("Initialized distributed")

    draft_model_last_checkpoint = None
    ckpt_info = (0, 0)
    if args.resume and os.path.isdir(args.output_dir):
        draft_model_last_checkpoint, ckpt_info = get_last_checkpoint(args.output_dir)
        print(f"Last checkpoint detected: {draft_model_last_checkpoint}")

    # If resuming, load config from checkpoint to ensure consistency
    if draft_model_last_checkpoint:
        checkpoint_config_path = os.path.join(
            draft_model_last_checkpoint, "config.json"
        )
        if os.path.exists(checkpoint_config_path):
            print(f"Loading draft config from checkpoint: {checkpoint_config_path}")
            args.draft_config_path = checkpoint_config_path

    target_model, draft_model = build_models(args)

    resume_state = None
    if draft_model_last_checkpoint:
        loaded_model = DFlashDraftModel.from_pretrained(
            draft_model_last_checkpoint, torch_dtype=torch.bfloat16
        )
        draft_model.load_state_dict(loaded_model.state_dict())
        del loaded_model
        print("Loaded draft model weights from checkpoint")

        training_state_path = os.path.join(
            draft_model_last_checkpoint, "training_state.pt"
        )
        if os.path.exists(training_state_path):
            resume_state = torch.load(
                training_state_path, map_location="cpu", weights_only=False
            )
            print(
                f"Will resume from epoch {resume_state['epoch']}, "
                f"step {resume_state['global_step']}"
            )

    tokenizer = load_tokenizer(args.target_model_path)

    if args.mask_token_id is not None:
        mask_token_id = args.mask_token_id
    elif (
        dflash_config := getattr(draft_model.config, "dflash_config", {})
    ) and dflash_config.get("mask_token_id") is not None:
        mask_token_id = dflash_config["mask_token_id"]
    elif tokenizer.mask_token_id is not None:
        mask_token_id = tokenizer.mask_token_id
    else:
        tokenizer.add_special_tokens({"mask_token": "<|MASK|>"})
        mask_token_id = tokenizer.mask_token_id
    print_on_rank0(f"Using mask_token_id: {mask_token_id}")

    draft_model.mask_token_id = mask_token_id
    draft_model.config.dflash_config["mask_token_id"] = mask_token_id
    draft_model.config.dflash_config["target_layer_ids"] = draft_model.target_layer_ids
    print_on_rank0(f"dflash_config: {draft_model.config.dflash_config}")

    train_dataloader, eval_dataloader = build_dataloader(args, tokenizer)

    steps_per_epoch = math.ceil(len(train_dataloader) / args.accumulation_steps)
    total_steps = args.num_epochs * steps_per_epoch
    print_on_rank0(f"Total training steps: {total_steps}")

    print_on_rank0("Loading target embeddings and head...")
    device = get_local_device()
    device_type = device.type
    target_components = TargetEmbeddingsAndHead.from_pretrained(
        args.target_model_path,
        embed_key=args.embedding_key,
        lm_head_key=args.lm_head_key,
        device=device_type,
        trust_remote_code=args.trust_remote_code,
    )

    dflash_model = OnlineDFlashModel(
        draft_model=draft_model,
        target_lm_head=target_components.lm_head,
        target_embed_tokens=target_components.embed_tokens,
        block_size=draft_model.block_size,
        mask_token_id=mask_token_id,
        attention_backend=args.attention_backend,
        num_anchors=args.num_anchors,
        loss_decay_gamma=args.loss_decay_gamma,
        loss_type=args.loss_type,
        dpace_alpha=args.dpace_alpha,
        prefix_weight_base=args.prefix_weight_base,
    )

    # Wrap each transformer block as its own FSDP unit so that all-gather /
    # reduce-scatter overlap with compute. Without an auto_wrap_policy the
    # whole model is a single FSDP unit, forcing every collective onto the
    # critical path with no overlap. The block class is resolved from the
    # draft model's `_no_split_modules` so this stays architecture-agnostic
    # rather than hardcoding a specific decoder-layer class.
    fsdp_kwargs = dict(
        use_orig_params=True,
        forward_prefetch=True,
        backward_prefetch=BackwardPrefetch.BACKWARD_PRE,
        limit_all_gathers=True,
        mixed_precision=MixedPrecision(
            param_dtype=torch.bfloat16,
            buffer_dtype=torch.bfloat16,
        ),
        sharding_strategy=ShardingStrategy.SHARD_GRAD_OP,
    )
    block_names = set(getattr(draft_model, "_no_split_modules", None) or [])
    block_classes = {
        type(m) for m in dflash_model.modules() if type(m).__name__ in block_names
    }
    if block_classes:
        fsdp_kwargs["auto_wrap_policy"] = functools.partial(
            transformer_auto_wrap_policy,
            transformer_layer_cls=block_classes,
        )
    else:
        print_with_rank(
            "No _no_split_modules on draft model; falling back to single-unit "
            "FSDP wrap (no compute-comm overlap)."
        )
    dflash_model = FSDP(dflash_model, **fsdp_kwargs)
    print_with_rank("Initialized FSDP")

    start_epoch = ckpt_info[0]
    global_step = ckpt_info[1]

    optimizer = BF16Optimizer(
        draft_model,
        lr=args.learning_rate,
        max_grad_norm=args.max_grad_norm,
        warmup_ratio=args.warmup_ratio,
        total_steps=total_steps,
    )

    if resume_state is not None:
        optimizer.load_state_dict(resume_state)
        start_epoch = resume_state["epoch"]
        global_step = resume_state["global_step"]
        del resume_state
        print_on_rank0(
            f"Restored optimizer/scheduler state: "
            f"epoch={start_epoch}, step={global_step}, "
            f"lr={optimizer.get_learning_rate():.6f}"
        )

    skip_steps = global_step - start_epoch * len(train_dataloader)

    print_on_rank0(f"Initializing tracker (report_to={args.report_to})...")
    tracker = create_tracker(args, args.output_dir)
    print_on_rank0("Tracker initialized successfully.")

    last_time = time.time()
    print_on_rank0(f"Starting training from epoch {start_epoch}, step {global_step}")

    for epoch in range(start_epoch, args.num_epochs):
        train_dataloader.sampler.set_epoch(epoch)
        draft_model.train()

        if dist.get_rank() == 0:
            progress_bar = tqdm(
                train_dataloader, desc=f"Training Epoch {epoch}", leave=True
            )
        else:
            progress_bar = train_dataloader

        for step_in_epoch, data in enumerate(progress_bar):
            if epoch == start_epoch and step_in_epoch < skip_steps:
                continue
            global_step += 1

            input_ids = data["input_ids"].to(device, non_blocking=True)
            attention_mask = data["attention_mask"].to(device, non_blocking=True)
            loss_mask = data["loss_mask"].to(device, non_blocking=True)
            target_output = target_model.generate_dflash_data(
                input_ids, attention_mask, loss_mask
            )
            hidden_states = target_output.hidden_states.to(device, non_blocking=True)

            loss, accuracy = dflash_model(
                input_ids=input_ids,
                hidden_states=hidden_states,
                loss_mask=loss_mask,
            )

            (loss / args.accumulation_steps).backward()

            if global_step % args.accumulation_steps == 0:
                optimizer.step()

            if global_step % args.log_interval == 0:
                loss_log = loss.clone()
                acc_log = accuracy.clone()
                dist.all_reduce(loss_log)
                dist.all_reduce(acc_log)
                loss_log = loss_log / dist.get_world_size()
                acc_log = acc_log / dist.get_world_size()

                record_metrics(
                    args,
                    loss_log.item(),
                    acc_log.item(),
                    global_step,
                    tracker,
                    optimizer,
                    train_dataloader,
                    mode="train",
                )

            if dist.get_rank() == 0:
                elapsed = time.time() - last_time
                last_time = time.time()
                progress_bar.set_postfix(
                    {
                        "loss": f"{loss.item():.4f}",
                        "acc": f"{accuracy.item():.4f}",
                        "iter_time": f"{elapsed:.2f}s",
                    }
                )

            if global_step % args.save_interval == 0:
                save_checkpoint(
                    args, epoch, global_step, dflash_model, draft_model, optimizer
                )

    save_checkpoint(
        args, args.num_epochs, global_step, dflash_model, draft_model, optimizer
    )

    tracker.close()
    destroy_distributed()


if __name__ == "__main__":
    main()