progress (messy :O)
This commit is contained in:
@@ -7,11 +7,13 @@ from __future__ import annotations
|
|||||||
import os
|
import os
|
||||||
from collections import defaultdict
|
from collections import defaultdict
|
||||||
from functools import partial, wraps
|
from functools import partial, wraps
|
||||||
from typing import Callable, Literal, Optional
|
from typing import Any, Callable, Literal, Optional
|
||||||
|
|
||||||
|
from axolotl.utils.ctx_managers.utils import get_context_parallel_manager
|
||||||
import datasets
|
import datasets
|
||||||
import torch
|
import torch
|
||||||
from datasets import Dataset
|
from datasets import Dataset
|
||||||
|
from torch import nn
|
||||||
from torch.utils.data import (
|
from torch.utils.data import (
|
||||||
BatchSampler,
|
BatchSampler,
|
||||||
DataLoader,
|
DataLoader,
|
||||||
|
|||||||
@@ -17,7 +17,6 @@ from accelerate.utils import save_fsdp_model
|
|||||||
from datasets import Dataset
|
from datasets import Dataset
|
||||||
from huggingface_hub.errors import OfflineModeIsEnabled
|
from huggingface_hub.errors import OfflineModeIsEnabled
|
||||||
from peft import PeftConfig, PeftModel
|
from peft import PeftConfig, PeftModel
|
||||||
from torch.distributed.tensor.experimental import context_parallel
|
|
||||||
from transformers import PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
|
from transformers import PreTrainedModel, PreTrainedTokenizer, ProcessorMixin
|
||||||
from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
|
from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
|
||||||
from transformers.trainer import Trainer
|
from transformers.trainer import Trainer
|
||||||
@@ -34,7 +33,7 @@ from axolotl.loaders import (
|
|||||||
load_processor,
|
load_processor,
|
||||||
load_tokenizer,
|
load_tokenizer,
|
||||||
)
|
)
|
||||||
from axolotl.utils.ctx_managers.sequence_parallel import SequenceParallelContextManager
|
from axolotl.utils.ctx_managers.sequence_parallel import ContextParallelContextManager
|
||||||
from axolotl.utils.dict import DictDefault
|
from axolotl.utils.dict import DictDefault
|
||||||
from axolotl.utils.distributed import cleanup_distributed
|
from axolotl.utils.distributed import cleanup_distributed
|
||||||
from axolotl.utils.freeze import freeze_layers_except
|
from axolotl.utils.freeze import freeze_layers_except
|
||||||
@@ -205,32 +204,24 @@ def execute_training(
|
|||||||
)
|
)
|
||||||
|
|
||||||
if cfg.sequence_parallel_degree > 1:
|
if cfg.sequence_parallel_degree > 1:
|
||||||
if cfg.sdp_attention:
|
# Models to enter context parallel manager for
|
||||||
world_size = dist.get_world_size()
|
models = [trainer.model]
|
||||||
mesh_shape = (
|
if hasattr(trainer, "ref_model") and trainer.ref_model:
|
||||||
world_size // cfg.sequence_parallel_degree,
|
models.append(trainer.ref_model)
|
||||||
cfg.sequence_parallel_degree,
|
|
||||||
)
|
|
||||||
mesh = dist.DeviceMesh(
|
|
||||||
"cuda",
|
|
||||||
torch.tensor(list(range(world_size))).reshape(mesh_shape),
|
|
||||||
mesh_dim_names=("dp", "cp"),
|
|
||||||
)
|
|
||||||
stack.enter_context(context_parallel(mesh=mesh))
|
|
||||||
else: # flash_attention
|
|
||||||
models = [trainer.model]
|
|
||||||
if hasattr(trainer, "ref_model") and trainer.ref_model:
|
|
||||||
models.append(trainer.ref_model)
|
|
||||||
|
|
||||||
stack.enter_context(
|
# Attention backend
|
||||||
SequenceParallelContextManager(
|
backend = "sdp_attention" if cfg.sdp_attention else "flash_attention"
|
||||||
models=models,
|
|
||||||
sequence_parallel_degree=cfg.sequence_parallel_degree,
|
stack.enter_context(
|
||||||
gradient_accumulation_steps=cfg.gradient_accumulation_steps,
|
ContextParallelContextManager(
|
||||||
ring_attn_func=cfg.ring_attn_func,
|
models=models,
|
||||||
heads_k_stride=cfg.heads_k_stride,
|
backend=backend,
|
||||||
)
|
context_parallel_degree=cfg.sequence_parallel_degree,
|
||||||
|
gradient_accumulation_steps=cfg.gradient_accumulation_steps,
|
||||||
|
ring_attn_func=cfg.ring_attn_func,
|
||||||
|
heads_k_stride=cfg.heads_k_stride,
|
||||||
)
|
)
|
||||||
|
)
|
||||||
|
|
||||||
LOG.info("Starting trainer...")
|
LOG.info("Starting trainer...")
|
||||||
trainer.train(resume_from_checkpoint=resume_from_checkpoint)
|
trainer.train(resume_from_checkpoint=resume_from_checkpoint)
|
||||||
|
|||||||
@@ -3,4 +3,4 @@
|
|||||||
# pylint: disable=unused-import
|
# pylint: disable=unused-import
|
||||||
# flake8: noqa
|
# flake8: noqa
|
||||||
|
|
||||||
from .sequence_parallel import SequenceParallelContextManager
|
from .sequence_parallel import ContextParallelContextManager
|
||||||
|
|||||||
@@ -2,11 +2,13 @@
|
|||||||
|
|
||||||
import functools
|
import functools
|
||||||
import inspect
|
import inspect
|
||||||
|
from typing import Literal
|
||||||
|
|
||||||
import torch
|
import torch
|
||||||
import torch.distributed as dist
|
import torch.distributed as dist
|
||||||
from torch import nn
|
from torch.distributed.tensor.experimental import context_parallel
|
||||||
from torch.utils.hooks import RemovableHandle
|
from torch.utils.hooks import RemovableHandle
|
||||||
|
from transformers import PreTrainedModel
|
||||||
from transformers.modeling_outputs import CausalLMOutputWithPast
|
from transformers.modeling_outputs import CausalLMOutputWithPast
|
||||||
from transformers.utils import ModelOutput
|
from transformers.utils import ModelOutput
|
||||||
|
|
||||||
@@ -15,177 +17,41 @@ from axolotl.monkeypatch.ring_attn import (
|
|||||||
patch_prepare_data_loader,
|
patch_prepare_data_loader,
|
||||||
patch_prepare_device_mesh,
|
patch_prepare_device_mesh,
|
||||||
register_ring_attn,
|
register_ring_attn,
|
||||||
update_ring_attn_params,
|
|
||||||
)
|
)
|
||||||
from axolotl.utils.schemas.enums import RingAttnFunc
|
from axolotl.utils.schemas.enums import RingAttnFunc
|
||||||
|
from axolotl.utils.ctx_managers.utils import get_context_parallel_manager
|
||||||
|
|
||||||
|
|
||||||
# TODO(djsaunde): implement zigzag, stripe patterns here (and elsewhere) in this
|
class ContextParallelContextManager:
|
||||||
# module. Currently, we just focus on batch ring and varlen llama3 for simplicity.
|
"""Context manager for context parallelism operations.
|
||||||
def apply_sequence_parallelism(
|
|
||||||
batch: dict[str, torch.Tensor],
|
|
||||||
local_rank: int,
|
|
||||||
local_world_size: int,
|
|
||||||
gradient_accumulation_steps: int,
|
|
||||||
ring_attn_func: RingAttnFunc, # pylint: disable=unused-argument
|
|
||||||
) -> tuple[dict[str, torch.Tensor], int, int]:
|
|
||||||
"""
|
|
||||||
Apply sequence parallelism slicing to a batch.
|
|
||||||
|
|
||||||
Special handling is implemented for integer logits_to_keep, which indicates
|
This class provides a context that will automatically apply context parallelism
|
||||||
to only keep the last N tokens in the sequence during generation.
|
|
||||||
|
|
||||||
Args:
|
|
||||||
batch: Batch dictionary (e.g., input_ids, attention_mask, etc.).
|
|
||||||
local_rank: Local rank in the sequence parallel group.
|
|
||||||
local_world_size: World size of the sequence parallel group.
|
|
||||||
gradient_accumulation_steps: Number of steps to accumulate gradients over.
|
|
||||||
ring_attn_func: Which ring attention function to use. Currently unused, but
|
|
||||||
related to above TODO.
|
|
||||||
|
|
||||||
Returns:
|
|
||||||
tuple of:
|
|
||||||
- Batch dictionary with sliced tensors.
|
|
||||||
- The original sequence length before padding.
|
|
||||||
- The number of padding tokens added.
|
|
||||||
"""
|
|
||||||
original_seq_len = batch["input_ids"].size(1)
|
|
||||||
|
|
||||||
# Update ring attention params if needed
|
|
||||||
if batch.get("position_ids") is not None:
|
|
||||||
update_ring_attn_params(position_ids=batch["position_ids"])
|
|
||||||
else:
|
|
||||||
# If position_ids aren't already in the batch, create them
|
|
||||||
batch["position_ids"] = torch.arange(
|
|
||||||
0,
|
|
||||||
original_seq_len,
|
|
||||||
dtype=torch.long,
|
|
||||||
device=batch["input_ids"].device,
|
|
||||||
).expand(batch["input_ids"].size(0), -1)
|
|
||||||
|
|
||||||
if "logits_to_keep" in batch and isinstance(batch["logits_to_keep"], int):
|
|
||||||
logits_to_keep = batch["logits_to_keep"]
|
|
||||||
|
|
||||||
# Calculate which positions in the full sequence contain the last N tokens
|
|
||||||
start_position = max(0, original_seq_len - logits_to_keep)
|
|
||||||
chunk_size = original_seq_len // local_world_size
|
|
||||||
rank_start = local_rank * chunk_size
|
|
||||||
rank_end = rank_start + chunk_size
|
|
||||||
|
|
||||||
# Create a boolean mask tensor for this rank's chunk
|
|
||||||
mask = torch.zeros(
|
|
||||||
chunk_size,
|
|
||||||
dtype=torch.bool,
|
|
||||||
device=batch["input_ids"].device,
|
|
||||||
)
|
|
||||||
|
|
||||||
if rank_end > start_position:
|
|
||||||
# Calculate how many of the last N tokens fall within this rank's range
|
|
||||||
tokens_in_rank = min(rank_end, original_seq_len) - max(
|
|
||||||
rank_start, start_position
|
|
||||||
)
|
|
||||||
|
|
||||||
# Calculate where these tokens start in the local chunk
|
|
||||||
local_start_idx = max(0, start_position - rank_start)
|
|
||||||
|
|
||||||
# Set the appropriate positions in the mask to True
|
|
||||||
mask[local_start_idx : local_start_idx + tokens_in_rank] = True
|
|
||||||
|
|
||||||
# Replace the integer with the boolean mask
|
|
||||||
batch["logits_to_keep"] = mask
|
|
||||||
|
|
||||||
# Add padding to make sequence length divisible by local_world_size
|
|
||||||
total_seq_len = original_seq_len
|
|
||||||
pad_len = 0
|
|
||||||
divisor = min(local_world_size, 64)
|
|
||||||
if total_seq_len % divisor != 0:
|
|
||||||
pad_len = divisor - (total_seq_len % divisor)
|
|
||||||
|
|
||||||
# Apply padding to all relevant tensors
|
|
||||||
for key in batch:
|
|
||||||
if (
|
|
||||||
isinstance(batch[key], torch.Tensor)
|
|
||||||
and batch[key].dim() > 1
|
|
||||||
and batch[key].size(1) == total_seq_len
|
|
||||||
):
|
|
||||||
# Create padding tensor
|
|
||||||
pad_value = -100 if key == "labels" else 0
|
|
||||||
padding = torch.full(
|
|
||||||
(batch[key].size(0), pad_len, *batch[key].shape[2:]),
|
|
||||||
pad_value,
|
|
||||||
dtype=batch[key].dtype,
|
|
||||||
device=batch[key].device,
|
|
||||||
)
|
|
||||||
|
|
||||||
# Concatenate padding to the right side of the tensor
|
|
||||||
batch[key] = torch.cat([batch[key], padding], dim=1)
|
|
||||||
if key == "logits_to_keep":
|
|
||||||
# Create padding tensor
|
|
||||||
padding = torch.ones(
|
|
||||||
1,
|
|
||||||
dtype=batch[key].dtype,
|
|
||||||
device=batch[key].device,
|
|
||||||
)
|
|
||||||
|
|
||||||
# Concatenate padding to the right side of the tensor
|
|
||||||
batch[key] = torch.cat([batch[key], padding], dim=0)
|
|
||||||
|
|
||||||
# Update the total sequence length after padding
|
|
||||||
total_seq_len = batch["input_ids"].size(1)
|
|
||||||
|
|
||||||
# Slice batch for sequence parallel
|
|
||||||
for key in batch:
|
|
||||||
if not isinstance(batch[key], torch.Tensor) or batch[key].dim() <= 1:
|
|
||||||
continue
|
|
||||||
|
|
||||||
# Split in sequential fashion and grab this rank's chunk
|
|
||||||
if batch[key].size(1) == total_seq_len:
|
|
||||||
batch[key] = (
|
|
||||||
batch[key].chunk(local_world_size, dim=1)[local_rank].contiguous()
|
|
||||||
)
|
|
||||||
elif key == "logits_to_keep":
|
|
||||||
batch[key] = (
|
|
||||||
batch[key].chunk(local_world_size, dim=0)[local_rank].contiguous()
|
|
||||||
)
|
|
||||||
|
|
||||||
# Handle num_items_in_batch
|
|
||||||
if "num_items_in_batch" in batch:
|
|
||||||
# Approximation; this needed since num_items_in_batch may be counted across
|
|
||||||
# all samples in a gradient accumulated batch, not on a per-step basis.
|
|
||||||
batch["num_items_in_batch"] = (
|
|
||||||
batch["labels"] != -100
|
|
||||||
).sum() * gradient_accumulation_steps
|
|
||||||
|
|
||||||
return batch, original_seq_len, pad_len
|
|
||||||
|
|
||||||
|
|
||||||
class SequenceParallelContextManager:
|
|
||||||
"""Context manager for sequence parallelism operations.
|
|
||||||
|
|
||||||
This class provides a context that will automatically apply sequence parallelism
|
|
||||||
during model forward passes using a pre-forward hook, and gather outputs from
|
during model forward passes using a pre-forward hook, and gather outputs from
|
||||||
across the sequence parallelism group using a post-forward hook.
|
across the context parallelism group using a post-forward hook.
|
||||||
|
|
||||||
Args:
|
Args:
|
||||||
models: List of models to apply sequence parallelism to pre- and post- forward
|
models: List of models to apply context parallelism to pre- and post- forward
|
||||||
hooks.
|
hooks.
|
||||||
sequence_parallel_degree: Number of processes to split sequences over.
|
backend: Which attention backend to use.
|
||||||
|
context_parallel_degree: Number of processes to split sequences over.
|
||||||
gradient_accumulation_steps: Number of steps to accumulate gradients over.
|
gradient_accumulation_steps: Number of steps to accumulate gradients over.
|
||||||
ring_attn_func: Which ring attention function to use. Currently unused.
|
ring_attn_func: Which ring attention function to use. Currently unused.
|
||||||
heads_k_stride: Sequence parallelism K head stride size. Passed through to
|
heads_k_stride: Context parallelism K head stride size. Passed through to
|
||||||
`varlen_llama3` `ring_flash_attn` implementation.
|
`varlen_llama3` `ring_flash_attn` implementation.
|
||||||
"""
|
"""
|
||||||
|
|
||||||
def __init__(
|
def __init__(
|
||||||
self,
|
self,
|
||||||
models: list[nn.Module],
|
models: list[PreTrainedModel],
|
||||||
sequence_parallel_degree: int,
|
backend: Literal["sdp_attention", "flash_attention"],
|
||||||
|
context_parallel_degree: int,
|
||||||
gradient_accumulation_steps: int,
|
gradient_accumulation_steps: int,
|
||||||
ring_attn_func: RingAttnFunc,
|
ring_attn_func: RingAttnFunc,
|
||||||
heads_k_stride: int | None,
|
heads_k_stride: int | None,
|
||||||
):
|
):
|
||||||
self.models = models
|
self.models = models
|
||||||
self.sequence_parallel_degree = sequence_parallel_degree
|
self.backend = backend
|
||||||
|
self.context_parallel_degree = context_parallel_degree
|
||||||
self.gradient_accumulation_steps = gradient_accumulation_steps
|
self.gradient_accumulation_steps = gradient_accumulation_steps
|
||||||
self.ring_attn_func = ring_attn_func
|
self.ring_attn_func = ring_attn_func
|
||||||
self.heads_k_stride = heads_k_stride
|
self.heads_k_stride = heads_k_stride
|
||||||
@@ -204,14 +70,34 @@ class SequenceParallelContextManager:
|
|||||||
self.pad_len = 0
|
self.pad_len = 0
|
||||||
|
|
||||||
# Create a partially applied version of the apply_sequence_parallelism function
|
# Create a partially applied version of the apply_sequence_parallelism function
|
||||||
self.apply_sequence_parallelism = functools.partial(
|
self.apply_context_parallelism = functools.partial(
|
||||||
apply_sequence_parallelism,
|
apply_context_parallelism,
|
||||||
local_rank=self.local_rank,
|
local_rank=self.local_rank,
|
||||||
local_world_size=self.local_world_size,
|
local_world_size=self.local_world_size,
|
||||||
gradient_accumulation_steps=self.gradient_accumulation_steps,
|
gradient_accumulation_steps=self.gradient_accumulation_steps,
|
||||||
ring_attn_func=self.ring_attn_func,
|
ring_attn_func=self.ring_attn_func,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
# SPDA CP initialization
|
||||||
|
world_size = dist.get_world_size()
|
||||||
|
mesh_shape = (
|
||||||
|
world_size // self.context_parallel_degree,
|
||||||
|
self.context_parallel_degree,
|
||||||
|
)
|
||||||
|
world_mesh = dist.DeviceMesh(
|
||||||
|
"cuda",
|
||||||
|
torch.tensor(list(range(world_size))).reshape(mesh_shape),
|
||||||
|
mesh_dim_names=("dp", "cp"),
|
||||||
|
)
|
||||||
|
self.context_parallel_managers = []
|
||||||
|
for model in models:
|
||||||
|
ctx_manager = get_context_parallel_manager(
|
||||||
|
enabled=self.context_parallel_degree > 1,
|
||||||
|
world_mesh=world_mesh,
|
||||||
|
model=model,
|
||||||
|
)
|
||||||
|
self.context_parallel_managers.append(ctx_manager)
|
||||||
|
|
||||||
def __enter__(self):
|
def __enter__(self):
|
||||||
self._register_model_hooks()
|
self._register_model_hooks()
|
||||||
|
|
||||||
@@ -226,22 +112,25 @@ class SequenceParallelContextManager:
|
|||||||
# TODO(djsaunde): Un-patch attention and accelerate functions (low priority)
|
# TODO(djsaunde): Un-patch attention and accelerate functions (low priority)
|
||||||
|
|
||||||
def _register_ring_attn(self):
|
def _register_ring_attn(self):
|
||||||
# Initialize ring attn for sequence parallelism
|
if self.backend == "flash_attention":
|
||||||
register_ring_attn(
|
# Initialize ring attn for context parallelism
|
||||||
sequence_parallel_degree=self.sequence_parallel_degree,
|
register_ring_attn(
|
||||||
heads_k_stride=self.heads_k_stride,
|
sequence_parallel_degree=self.context_parallel_degree,
|
||||||
ring_attn_func=self.ring_attn_func,
|
heads_k_stride=self.heads_k_stride,
|
||||||
)
|
ring_attn_func=self.ring_attn_func,
|
||||||
|
)
|
||||||
|
else:
|
||||||
|
stack.enter_context(context_parallel(mesh=mesh))
|
||||||
|
|
||||||
# Patches for accelerate functionality
|
# Patches for accelerate functionality
|
||||||
patch_prepare_data_loader()
|
patch_prepare_data_loader()
|
||||||
patch_prepare_device_mesh(
|
patch_prepare_device_mesh(
|
||||||
sequence_parallel_degree=self.sequence_parallel_degree
|
sequence_parallel_degree=self.context_parallel_degree
|
||||||
)
|
)
|
||||||
|
|
||||||
def _register_model_hooks(self):
|
def _register_model_hooks(self):
|
||||||
# Forward pre-hook to apply sequence parallelism
|
# Forward pre-hook to apply sequence parallelism
|
||||||
def sequence_parallel_pre_hook(_, args, kwargs):
|
def cp_flash_pre_hook(_, args, kwargs):
|
||||||
# Get parameter names from the model's forward function
|
# Get parameter names from the model's forward function
|
||||||
forward_params = list(
|
forward_params = list(
|
||||||
inspect.signature(self.models[0].forward).parameters.keys()
|
inspect.signature(self.models[0].forward).parameters.keys()
|
||||||
@@ -257,13 +146,13 @@ class SequenceParallelContextManager:
|
|||||||
|
|
||||||
# Apply sequence parallelism to updated kwargs
|
# Apply sequence parallelism to updated kwargs
|
||||||
updated_kwargs, self.original_seq_len, self.pad_len = (
|
updated_kwargs, self.original_seq_len, self.pad_len = (
|
||||||
self.apply_sequence_parallelism(updated_kwargs)
|
self.apply_context_parallelism(updated_kwargs)
|
||||||
)
|
)
|
||||||
|
|
||||||
return remaining_args, updated_kwargs
|
return remaining_args, updated_kwargs
|
||||||
|
|
||||||
# Forward post-hook to gather outputs
|
# Forward post-hook to gather outputs
|
||||||
def sequence_parallel_post_hook(_, __, output: ModelOutput) -> ModelOutput:
|
def cp_flash_post_hook(_, __, output: ModelOutput) -> ModelOutput:
|
||||||
# Gather the sharded outputs
|
# Gather the sharded outputs
|
||||||
output = self._gather_outputs(output)
|
output = self._gather_outputs(output)
|
||||||
|
|
||||||
@@ -277,15 +166,19 @@ class SequenceParallelContextManager:
|
|||||||
|
|
||||||
return output
|
return output
|
||||||
|
|
||||||
|
def cp_sdpa_pre_hook(_, args, kwargs):
|
||||||
|
with self.context_parallel_managers[?](list(inputs.values())):
|
||||||
|
|
||||||
|
|
||||||
# Register both hooks
|
# Register both hooks
|
||||||
for model in self.models:
|
for model in self.models:
|
||||||
self.hook_handles.append(
|
self.hook_handles.append(
|
||||||
model.register_forward_pre_hook(
|
model.register_forward_pre_hook(
|
||||||
sequence_parallel_pre_hook, with_kwargs=True
|
cp_flash_pre_hook, with_kwargs=True
|
||||||
)
|
)
|
||||||
)
|
)
|
||||||
self.hook_handles.append(
|
self.hook_handles.append(
|
||||||
model.register_forward_hook(sequence_parallel_post_hook)
|
model.register_forward_hook(cp_flash_post_hook)
|
||||||
)
|
)
|
||||||
|
|
||||||
def _gather_outputs(self, output: CausalLMOutputWithPast) -> CausalLMOutputWithPast:
|
def _gather_outputs(self, output: CausalLMOutputWithPast) -> CausalLMOutputWithPast:
|
||||||
145
src/axolotl/utils/ctx_managers/context_parallel/utils.py
Normal file
145
src/axolotl/utils/ctx_managers/context_parallel/utils.py
Normal file
@@ -0,0 +1,145 @@
|
|||||||
|
"""Utils for context parallel context manager."""
|
||||||
|
|
||||||
|
import torch
|
||||||
|
|
||||||
|
from axolotl.monkeypatch.ring_attn.patch import update_ring_attn_params
|
||||||
|
from axolotl.utils.schemas.enums import RingAttnFunc
|
||||||
|
|
||||||
|
|
||||||
|
# TODO(djsaunde): implement zigzag, stripe patterns here (and elsewhere) in this
|
||||||
|
# module. Currently, we just focus on batch ring and varlen llama3 for simplicity.
|
||||||
|
def apply_context_parallelism(
|
||||||
|
batch: dict[str, torch.Tensor],
|
||||||
|
local_rank: int,
|
||||||
|
local_world_size: int,
|
||||||
|
gradient_accumulation_steps: int,
|
||||||
|
ring_attn_func: RingAttnFunc, # pylint: disable=unused-argument
|
||||||
|
) -> tuple[dict[str, torch.Tensor], int, iwnt]:
|
||||||
|
"""
|
||||||
|
Apply context parallelism slicing to a batch.
|
||||||
|
|
||||||
|
Special handling is implemented for integer logits_to_keep, which indicates
|
||||||
|
to only keep the last N tokens in the input sequence during generation.
|
||||||
|
|
||||||
|
Args:
|
||||||
|
batch: Batch dictionary (e.g., input_ids, attention_mask, etc.).
|
||||||
|
local_rank: Local rank in the context parallel group.
|
||||||
|
local_world_size: World size of the context parallel group.
|
||||||
|
gradient_accumulation_steps: Number of steps to accumulate gradients over.
|
||||||
|
ring_attn_func: Which ring attention function to use. Currently unused, but
|
||||||
|
related to above TODO.
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
tuple of:
|
||||||
|
- Batch dictionary with sliced tensors.
|
||||||
|
- The original sequence length before padding.
|
||||||
|
- The number of padding tokens added.
|
||||||
|
"""
|
||||||
|
original_seq_len = batch["input_ids"].size(1)
|
||||||
|
|
||||||
|
# Update ring attention params if needed
|
||||||
|
if batch.get("position_ids") is not None:
|
||||||
|
update_ring_attn_params(position_ids=batch["position_ids"])
|
||||||
|
else:
|
||||||
|
# If position_ids aren't already in the batch, create them
|
||||||
|
batch["position_ids"] = torch.arange(
|
||||||
|
0,
|
||||||
|
original_seq_len,
|
||||||
|
dtype=torch.long,
|
||||||
|
device=batch["input_ids"].device,
|
||||||
|
).expand(batch["input_ids"].size(0), -1)
|
||||||
|
|
||||||
|
if "logits_to_keep" in batch and isinstance(batch["logits_to_keep"], int):
|
||||||
|
logits_to_keep = batch["logits_to_keep"]
|
||||||
|
|
||||||
|
# Calculate which positions in the full sequence contain the last N tokens
|
||||||
|
start_position = max(0, original_seq_len - logits_to_keep)
|
||||||
|
chunk_size = original_seq_len // local_world_size
|
||||||
|
rank_start = local_rank * chunk_size
|
||||||
|
rank_end = rank_start + chunk_size
|
||||||
|
|
||||||
|
# Create a boolean mask tensor for this rank's chunk
|
||||||
|
mask = torch.zeros(
|
||||||
|
chunk_size,
|
||||||
|
dtype=torch.bool,
|
||||||
|
device=batch["input_ids"].device,
|
||||||
|
)
|
||||||
|
|
||||||
|
if rank_end > start_position:
|
||||||
|
# Calculate how many of the last N tokens fall within this rank's range
|
||||||
|
tokens_in_rank = min(rank_end, original_seq_len) - max(
|
||||||
|
rank_start, start_position
|
||||||
|
)
|
||||||
|
|
||||||
|
# Calculate where these tokens start in the local chunk
|
||||||
|
local_start_idx = max(0, start_position - rank_start)
|
||||||
|
|
||||||
|
# Set the appropriate positions in the mask to True
|
||||||
|
mask[local_start_idx : local_start_idx + tokens_in_rank] = True
|
||||||
|
|
||||||
|
# Replace the integer with the boolean mask
|
||||||
|
batch["logits_to_keep"] = mask
|
||||||
|
|
||||||
|
# Add padding to make sequence length divisible by local_world_size
|
||||||
|
total_seq_len = original_seq_len
|
||||||
|
pad_len = 0
|
||||||
|
divisor = min(local_world_size, 64)
|
||||||
|
if total_seq_len % divisor != 0:
|
||||||
|
pad_len = divisor - (total_seq_len % divisor)
|
||||||
|
|
||||||
|
# Apply padding to all relevant tensors
|
||||||
|
for key in batch:
|
||||||
|
if (
|
||||||
|
isinstance(batch[key], torch.Tensor)
|
||||||
|
and batch[key].dim() > 1
|
||||||
|
and batch[key].size(1) == total_seq_len
|
||||||
|
):
|
||||||
|
# Create padding tensor
|
||||||
|
pad_value = -100 if key == "labels" else 0
|
||||||
|
padding = torch.full(
|
||||||
|
(batch[key].size(0), pad_len, *batch[key].shape[2:]),
|
||||||
|
pad_value,
|
||||||
|
dtype=batch[key].dtype,
|
||||||
|
device=batch[key].device,
|
||||||
|
)
|
||||||
|
|
||||||
|
# Concatenate padding to the right side of the tensor
|
||||||
|
batch[key] = torch.cat([batch[key], padding], dim=1)
|
||||||
|
if key == "logits_to_keep":
|
||||||
|
# Create padding tensor
|
||||||
|
padding = torch.ones(
|
||||||
|
1,
|
||||||
|
dtype=batch[key].dtype,
|
||||||
|
device=batch[key].device,
|
||||||
|
)
|
||||||
|
|
||||||
|
# Concatenate padding to the right side of the tensor
|
||||||
|
batch[key] = torch.cat([batch[key], padding], dim=0)
|
||||||
|
|
||||||
|
# Update the total sequence length after padding
|
||||||
|
total_seq_len = batch["input_ids"].size(1)
|
||||||
|
|
||||||
|
# Slice batch for context parallel
|
||||||
|
for key in batch:
|
||||||
|
if not isinstance(batch[key], torch.Tensor) or batch[key].dim() <= 1:
|
||||||
|
continue
|
||||||
|
|
||||||
|
# Split in sequential fashion and grab this rank's chunk
|
||||||
|
if batch[key].size(1) == total_seq_len:
|
||||||
|
batch[key] = (
|
||||||
|
batch[key].chunk(local_world_size, dim=1)[local_rank].contiguous()
|
||||||
|
)
|
||||||
|
elif key == "logits_to_keep":
|
||||||
|
batch[key] = (
|
||||||
|
batch[key].chunk(local_world_size, dim=0)[local_rank].contiguous()
|
||||||
|
)
|
||||||
|
|
||||||
|
# Handle num_items_in_batch
|
||||||
|
if "num_items_in_batch" in batch:
|
||||||
|
# Approximation; this needed since num_items_in_batch may be counted across
|
||||||
|
# all samples in a gradient accumulated batch, not on a per-step basis.
|
||||||
|
batch["num_items_in_batch"] = (
|
||||||
|
batch["labels"] != -100
|
||||||
|
).sum() * gradient_accumulation_steps
|
||||||
|
|
||||||
|
return batch, original_seq_len, pad_len
|
||||||
103
src/axolotl/utils/ctx_managers/utils.py
Normal file
103
src/axolotl/utils/ctx_managers/utils.py
Normal file
@@ -0,0 +1,103 @@
|
|||||||
|
import contextlib
|
||||||
|
from typing import Callable, Generator, Optional, Union
|
||||||
|
|
||||||
|
import torch
|
||||||
|
|
||||||
|
from torch.distributed.tensor.experimental import context_parallel
|
||||||
|
from torch.distributed.tensor.experimental._attention import set_rotate_method
|
||||||
|
from torch.nn.attention import sdpa_kernel, SDPBackend
|
||||||
|
from torch.nn.attention.flex_attention import BlockMask
|
||||||
|
from transformers import PreTrainedModel
|
||||||
|
|
||||||
|
|
||||||
|
def _get_sdpa_context() -> (
|
||||||
|
Callable[[Optional[Generator[None, None, None]]], Generator[None, None, None]]
|
||||||
|
):
|
||||||
|
"""
|
||||||
|
Creates a context manager to confine to flash/efficient/cuDNN attention backends.
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
A context manager function that takes an optional context parallel context.
|
||||||
|
"""
|
||||||
|
|
||||||
|
@contextlib.contextmanager
|
||||||
|
def context(cp_context: Union[Generator[None, None, None], None] = None):
|
||||||
|
with contextlib.ExitStack() as stack:
|
||||||
|
if cp_context is not None:
|
||||||
|
stack.enter_context(
|
||||||
|
sdpa_kernel(
|
||||||
|
[
|
||||||
|
SDPBackend.FLASH_ATTENTION,
|
||||||
|
SDPBackend.EFFICIENT_ATTENTION,
|
||||||
|
SDPBackend.CUDNN_ATTENTION,
|
||||||
|
]
|
||||||
|
)
|
||||||
|
)
|
||||||
|
stack.enter_context(cp_context)
|
||||||
|
|
||||||
|
yield
|
||||||
|
|
||||||
|
return context
|
||||||
|
|
||||||
|
|
||||||
|
def get_context_parallel_manager(
|
||||||
|
*,
|
||||||
|
enabled: bool = False,
|
||||||
|
world_mesh: torch.distributed.DeviceMesh,
|
||||||
|
model: PreTrainedModel,
|
||||||
|
) -> Callable[[list[torch.Tensor]], Generator[None, None, None]]:
|
||||||
|
"""
|
||||||
|
Context manager for applying context parallelism to a model. In addition to applying the
|
||||||
|
standard context manager to patch SDPA and shard model inputs and buffers along the sequence
|
||||||
|
dimension, this context manager also calls into _get_sdpa_context to filter to acceptable SDPA backends.
|
||||||
|
|
||||||
|
Args:
|
||||||
|
enabled: Whether context parallel is enabled. Default: False
|
||||||
|
world_mesh: Global device mesh.
|
||||||
|
model: Model to apply context parallelism to.
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
A context manager applying context parallelism if enabled is True. Otherwise a context manager
|
||||||
|
disabling the math SDPA backend.
|
||||||
|
|
||||||
|
Raises:
|
||||||
|
ValueError: if enabled is True but world_mesh does not contain a "cp" dimension
|
||||||
|
"""
|
||||||
|
|
||||||
|
if enabled and "cp" not in world_mesh.mesh_dim_names:
|
||||||
|
raise ValueError(
|
||||||
|
"Context parallel is enabled but no context parallel device mesh is provided."
|
||||||
|
)
|
||||||
|
# TODO: context parallel for multimodal models requires extra work
|
||||||
|
if enabled and not isinstance(model, TransformerDecoder):
|
||||||
|
raise ValueError("Context parallel is only supported for text models")
|
||||||
|
# TODO: this is a hacky proxy for whether we use flex for chunked attention
|
||||||
|
# remove this once flex is supported
|
||||||
|
if enabled and any([layer.mask_mod is not None for layer in model.layers]):
|
||||||
|
raise ValueError("Context parallel with flex attention is not yet supported")
|
||||||
|
model_buffers = list(model.buffers())
|
||||||
|
|
||||||
|
@contextlib.contextmanager
|
||||||
|
def context(model_inputs: list[torch.Tensor]):
|
||||||
|
# Create context parallel context if enabled
|
||||||
|
cp_context = None
|
||||||
|
if enabled and any([isinstance(input, BlockMask) for input in model_inputs]):
|
||||||
|
raise ValueError(
|
||||||
|
"Context parallel with flex attention is not yet supported"
|
||||||
|
)
|
||||||
|
if enabled:
|
||||||
|
set_rotate_method("allgather")
|
||||||
|
cp_context = context_parallel(
|
||||||
|
world_mesh["cp"],
|
||||||
|
buffers=model_inputs + model_buffers,
|
||||||
|
buffer_seq_dims=[1] * len(model_inputs) + [0] * len(model_buffers),
|
||||||
|
no_restore_buffers=set(model_inputs),
|
||||||
|
)
|
||||||
|
|
||||||
|
# Create and enter the train context with the optional cp_context
|
||||||
|
sdpa_context = _get_sdpa_context()
|
||||||
|
|
||||||
|
with sdpa_context(cp_context):
|
||||||
|
yield
|
||||||
|
|
||||||
|
return context
|
||||||
@@ -15,7 +15,7 @@ from axolotl.monkeypatch.ring_attn import (
|
|||||||
register_ring_attn,
|
register_ring_attn,
|
||||||
set_ring_attn_group,
|
set_ring_attn_group,
|
||||||
)
|
)
|
||||||
from axolotl.utils.ctx_managers.sequence_parallel import apply_sequence_parallelism
|
from axolotl.utils.ctx_managers.sequence_parallel import apply_context_parallelism
|
||||||
from axolotl.utils.dict import DictDefault
|
from axolotl.utils.dict import DictDefault
|
||||||
from axolotl.utils.schemas.enums import RingAttnFunc
|
from axolotl.utils.schemas.enums import RingAttnFunc
|
||||||
from axolotl.utils.schemas.trl import TRLConfig
|
from axolotl.utils.schemas.trl import TRLConfig
|
||||||
@@ -328,7 +328,7 @@ class TestApplySequenceParallelism:
|
|||||||
"""Test that function returns original batch when world size is 1."""
|
"""Test that function returns original batch when world size is 1."""
|
||||||
mock_get_ring_attn_group.return_value = 0
|
mock_get_ring_attn_group.return_value = 0
|
||||||
|
|
||||||
result, _, _ = apply_sequence_parallelism(
|
result, _, _ = apply_context_parallelism(
|
||||||
batch=sequence_parallel_batch,
|
batch=sequence_parallel_batch,
|
||||||
local_rank=0,
|
local_rank=0,
|
||||||
local_world_size=1,
|
local_world_size=1,
|
||||||
@@ -347,7 +347,7 @@ class TestApplySequenceParallelism:
|
|||||||
batch = sequence_parallel_batch
|
batch = sequence_parallel_batch
|
||||||
seq_len = batch["input_ids"].size(1)
|
seq_len = batch["input_ids"].size(1)
|
||||||
|
|
||||||
result, _, _ = apply_sequence_parallelism(
|
result, _, _ = apply_context_parallelism(
|
||||||
batch=batch,
|
batch=batch,
|
||||||
local_rank=0,
|
local_rank=0,
|
||||||
local_world_size=2,
|
local_world_size=2,
|
||||||
@@ -374,7 +374,7 @@ class TestApplySequenceParallelism:
|
|||||||
seq_len = batch["input_ids"].size(1)
|
seq_len = batch["input_ids"].size(1)
|
||||||
original_input_ids = batch["input_ids"].clone()
|
original_input_ids = batch["input_ids"].clone()
|
||||||
|
|
||||||
result, _, _ = apply_sequence_parallelism(
|
result, _, _ = apply_context_parallelism(
|
||||||
batch=batch,
|
batch=batch,
|
||||||
local_rank=1,
|
local_rank=1,
|
||||||
local_world_size=2,
|
local_world_size=2,
|
||||||
@@ -440,7 +440,7 @@ class TestApplySequenceParallelism:
|
|||||||
|
|
||||||
# Create a partially applied function
|
# Create a partially applied function
|
||||||
rank0_ring_parallel = functools.partial(
|
rank0_ring_parallel = functools.partial(
|
||||||
apply_sequence_parallelism,
|
apply_context_parallelism,
|
||||||
local_rank=0,
|
local_rank=0,
|
||||||
local_world_size=2,
|
local_world_size=2,
|
||||||
gradient_accumulation_steps=1,
|
gradient_accumulation_steps=1,
|
||||||
@@ -466,7 +466,7 @@ class TestApplySequenceParallelism:
|
|||||||
original_input_ids = batch["input_ids"].clone()
|
original_input_ids = batch["input_ids"].clone()
|
||||||
|
|
||||||
# This should run without error even though position_ids is missing
|
# This should run without error even though position_ids is missing
|
||||||
result, _, _ = apply_sequence_parallelism(
|
result, _, _ = apply_context_parallelism(
|
||||||
batch=batch,
|
batch=batch,
|
||||||
local_rank=0,
|
local_rank=0,
|
||||||
local_world_size=2,
|
local_world_size=2,
|
||||||
|
|||||||
Reference in New Issue
Block a user