[Refactor] Rename CELossContext to LMHeadLossContext and refactor loss context base class

- Rename CELossContext to LMHeadLossContext for better semantic clarity
- Refactor BaseLossContext to be more abstract by removing LM-specific logic
- Move eager_mode and chunk_mode implementations from base class to LMHeadLossContext
- Make loss_ctx_cls and _loss_kwargs_cls abstract properties in BaseLossConfig
- Remove sp_split() and to() implementations from BaseLossKwargs base class
- Move sp_split() and to() to CELossKwargs subclass
- Update BaseRLLossKwargs to properly inherit and extend sp_split() and to() methods
- Add deprecation alias: CELossContext = LMHeadLossContext for backward compatibility
- Export LMHeadLossContext in __init__.py

ghstack-source-id: 1b3d648
Pull-Request: #1571

Author: HAOCHENYE

xtuner/v1/loss/__init__.py

@@ -1,5 +1,5 @@
 from .base_loss_ctx import BaseLossConfig, BaseLossContext, BaseLossKwargs
-from .ce_loss import CELossConfig, CELossContext
+from .ce_loss import CELossConfig, CELossContext, LMHeadLossContext
 from .chunk_loss import ChunkLoss
 from .moe_loss import (
     BalancingLoss,

xtuner/v1/loss/base_loss_ctx.py

@@ -3,17 +3,10 @@
 from typing import Annotated, Any, Literal, TypeVar
 
 import torch
-import torch.distributed as dist
 import torch.nn as nn
 from cyclopts import Parameter
 from pydantic import BaseModel, ConfigDict
 from torch.distributed.device_mesh import DeviceMesh
-from torch.distributed.nn.functional import all_reduce
-from typing_extensions import Self
-
-from xtuner.v1.loss.utils import sp_split
-
-from .chunk_loss import ChunkLoss
 
 
 # Do loss calibration among dp, sp and grad accumulation:
@@ -46,18 +39,13 @@
 
 
 class BaseLossKwargs(BaseModel):
-    """Everything needed to compute the loss."""
-
-    model_config = ConfigDict(title="loss keyword arguments", extra="forbid", arbitrary_types_allowed=True)
-    shifted_labels: torch.Tensor
+    """Everything needed to compute the loss.
 
-    def sp_split(self, sp_mesh: DeviceMesh) -> Self:
-        self.shifted_labels = sp_split(self.shifted_labels, sp_mesh=sp_mesh, split_dim=1, padding_value=-100)
-        return self
+    Subclasses should implement sp_split() and to() methods if they contain tensors that need to be split across
+    sequence parallel mesh or moved to device.
+    """
 
-    def to(self, device: torch.device | str) -> Self:
-        self.shifted_labels = self.shifted_labels.to(device)
-        return self
+    model_config = ConfigDict(title="loss keyword arguments", extra="forbid", arbitrary_types_allowed=True)
 
     def chunk(self, chunk_size) -> list["BaseLossKwargs"]:
         tensor_fields: dict[str, tuple[torch.Tensor, ...]] = {}
@@ -114,10 +102,13 @@ class BaseLossConfig(BaseModel):
     chunk_size: Annotated[int | None, Parameter(help="chunk size when mode is chunk")] = 1024
 
     @property
+    @abstractmethod
     def loss_ctx_cls(self) -> type["BaseLossContext"]:
         raise NotImplementedError
 
+    # TODO: private property maybe not a good idea
     @property
+    @abstractmethod
     def _loss_kwargs_cls(self) -> type["BaseLossKwargs"]:
         raise NotImplementedError
 
@@ -160,72 +151,10 @@ def __init__(self, loss_cfg: BaseLossConfig, loss_kwargs: BaseLossKwargs):
         self._batch_size = 1
 
     @staticmethod
-    @abstractmethod
-    def build_batches(loss_ctx_list: list[_BaseLossContextT], *args, **kwargs) -> list[_BaseLossContextT]: ...
-
-    @abstractmethod
-    def loss_fn(
-        self,
-        hidden_states: torch.Tensor,
-        head_weight: torch.Tensor,
-        head_bias: torch.Tensor | None,
-        loss_kwargs: BaseLossKwargs,
-    ) -> tuple[torch.Tensor, tuple[torch.Tensor | None, dict[str, Any]]]:
-        """Step 2.a and 2.b in the loss calculation."""
-        ...
-
-    def eager_mode(
-        self,
-        hidden_states: torch.Tensor,
-        head_weight: torch.Tensor,
-        head_bias: torch.Tensor | None,
-        loss_kwargs: BaseLossKwargs,
-    ):
-        return self.loss_fn(hidden_states, head_weight, head_bias, loss_kwargs)
-
-    def chunk_mode(
-        self,
-        hidden_states: torch.Tensor,
-        head_weight: torch.Tensor,
-        head_bias: torch.Tensor | None,
-        loss_kwargs: BaseLossKwargs,
-    ):
-        assert self.loss_cfg.chunk_size is not None, "chunk_size must be set in chunk mode"
-
-        chunks = loss_kwargs.chunk(self.loss_cfg.chunk_size)
-        loss, extra_info = ChunkLoss.apply(
-            hidden_states, head_weight, head_bias, self.loss_fn, chunks, self.loss_cfg.chunk_size
-        )
-        return loss, (None, extra_info)
-
-    def forward(
-        self,
-        hidden_states: torch.Tensor,
-        head_weight: torch.Tensor,
-        head_bias: torch.Tensor | None = None,
-    ) -> tuple[torch.Tensor, tuple[torch.Tensor | None, dict[str, Any]]]:
-        from xtuner.v1.model.utils.misc import ModelForwardExtraLogInfo
-
-        assert self.loss_kwargs is not None, "loss_kwargs must be set before calling forward"
-        if head_bias is not None:
-            raise NotImplementedError("Loss does not support head_bias yet.")
-
-        if self.loss_cfg.mode == "eager":
-            loss, (logits, extra_info) = self.eager_mode(hidden_states, head_weight, head_bias, self.loss_kwargs)
-        else:
-            loss, (logits, extra_info) = self.chunk_mode(hidden_states, head_weight, head_bias, self.loss_kwargs)
-
-        # TODO: yanhuida, should be removed
-        if not isinstance(extra_info, ModelForwardExtraLogInfo):
-            extra_info = ModelForwardExtraLogInfo(extra_info)
-
-        extra_info["local_base_loss"] = loss.detach().clone()
-
-        # Step 2.c in the loss calculation: reduce the loss over all ranks using all_reduce with autograd support
-        if dist.is_initialized():
-            loss = all_reduce(loss, op=dist.ReduceOp.SUM, group=dist.group.WORLD)
-
-        return loss, (logits, extra_info)
+    def build_batches(loss_ctx_list: list[_BaseLossContextT], *args, **kwargs) -> list[_BaseLossContextT]:
+        for ctx in loss_ctx_list:
+            ctx._batch_size = len(loss_ctx_list)
+        return loss_ctx_list
 
     @classmethod
     def cat(cls: type[_BaseLossContextT], chunks: list[_BaseLossContextT]) -> _BaseLossContextT:

xtuner/v1/loss/ce_loss.py

@@ -6,8 +6,10 @@
 import torch.nn.functional as F
 from cyclopts import Parameter
 from torch.distributed.device_mesh import DeviceMesh
+from torch.distributed.nn.functional import all_reduce
 
 from xtuner.v1.loss import BaseLossConfig, BaseLossContext, BaseLossKwargs
+from xtuner.v1.loss.chunk_loss import ChunkLoss
 from xtuner.v1.utils.device import get_device
 
 # from xtuner.v1.profiler.prober import ProberList
@@ -37,7 +39,11 @@ class CELossConfig(BaseLossConfig):
     def loss_ctx_cls(self) -> type["CELossContext"]:
         return CELossContext
 
-    def model_post_init(self, __context: Any) -> None:
+    @property
+    def _loss_kwargs_cls(self) -> type["CELossKwargs"]:
+        return CELossKwargs
+
+    def model_post_init(self, _context: Any) -> None:
         if self.mode == "liger":
             assert self.loss_reduction == "token", "Currently, cannot use liger kernel with sample or square reduction"
 
@@ -80,8 +86,16 @@ class CELossKwargs(BaseLossKwargs):
     shifted_labels: torch.Tensor
     loss_weight: torch.Tensor | None = None
 
+    def sp_split(self, sp_mesh: DeviceMesh) -> "CELossKwargs":
+        self.shifted_labels = sp_split(self.shifted_labels, sp_mesh=sp_mesh, split_dim=1, padding_value=-100)
+        return self
+
+    def to(self, device: torch.device | str) -> "CELossKwargs":
+        self.shifted_labels = self.shifted_labels.to(device)
+        return self
+
 
-class CELossContext(BaseLossContext):
+class LMHeadLossContext(BaseLossContext):
     """Cross-entropy loss context for language models.
 
     Args:
@@ -163,6 +177,7 @@ def build_batches(  # type: ignore[override]
 
         for loss_ctx in loss_ctx_list:
             loss_ctx._batch_size = len(loss_ctx_list)
+            assert loss_ctx.loss_kwargs.loss_weight is not None
             loss_ctx.loss_kwargs.loss_weight /= global_denominator + 1e-12
         return loss_ctx_list
 
@@ -195,15 +210,30 @@ def loss_fn(
 
         return loss, (logits, {})
 
+    def eager_mode(
+        self,
+        hidden_states: torch.Tensor,
+        head_weight: torch.Tensor,
+        head_bias: torch.Tensor | None,
+        loss_kwargs: CELossKwargs,
+    ) -> tuple[torch.Tensor, tuple[torch.Tensor | None, dict[str, Any]]]:
+        return self.loss_fn(hidden_states, head_weight, head_bias, loss_kwargs)
+
     def chunk_mode(
         self,
         hidden_states: torch.Tensor,
         head_weight: torch.Tensor,
         head_bias: torch.Tensor | None,
         loss_kwargs: CELossKwargs,
-    ):
+    ) -> tuple[torch.Tensor, tuple[torch.Tensor | None, dict[str, Any]]]:
         if self.loss_cfg.mode == "chunk":
-            return super().chunk_mode(hidden_states, head_weight, head_bias, loss_kwargs)
+            assert self.loss_cfg.chunk_size is not None, "chunk_size must be set in chunk mode"
+
+            chunks = loss_kwargs.chunk(self.loss_cfg.chunk_size)
+            loss, extra_info = ChunkLoss.apply(
+                hidden_states, head_weight, head_bias, self.loss_fn, chunks, self.loss_cfg.chunk_size
+            )
+            return loss, (None, extra_info)
         else:
             assert self.liger_loss_fct is not None, "liger_loss_fct must be initialized in liger mode"
             shifted_labels = loss_kwargs.shifted_labels  # (bs, seq_len)
@@ -225,3 +255,36 @@ def chunk_mode(
     @property
     def batch_size(self) -> int:
         return self._batch_size
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        head_weight: torch.Tensor,
+        head_bias: torch.Tensor | None = None,
+    ) -> tuple[torch.Tensor, tuple[torch.Tensor | None, dict[str, Any]]]:
+        from xtuner.v1.model.utils.misc import ModelForwardExtraLogInfo
+
+        assert self.loss_kwargs is not None, "loss_kwargs must be set before calling forward"
+        if head_bias is not None:
+            raise NotImplementedError("Loss does not support head_bias yet.")
+
+        if self.loss_cfg.mode == "eager":
+            loss, (logits, extra_info) = self.eager_mode(hidden_states, head_weight, head_bias, self.loss_kwargs)
+        else:
+            loss, (logits, extra_info) = self.chunk_mode(hidden_states, head_weight, head_bias, self.loss_kwargs)
+
+        # TODO: yanhuida, should be removed
+        if not isinstance(extra_info, ModelForwardExtraLogInfo):
+            extra_info = ModelForwardExtraLogInfo(extra_info)
+
+        extra_info["local_base_loss"] = loss.detach().clone()
+
+        # Step 2.c in the loss calculation: reduce the loss over all ranks using all_reduce with autograd support
+        if dist.is_initialized():
+            loss = all_reduce(loss, op=dist.ReduceOp.SUM, group=dist.group.WORLD)
+
+        return loss, (logits, extra_info)
+
+
+# Deprecated: Use LMHeadLossContext instead. Will be removed in version 1.1.0
+CELossContext = LMHeadLossContext

xtuner/v1/loss/moe_loss.py

@@ -1,12 +1,11 @@
-from typing import Annotated, Any, Literal
+from typing import Annotated, Literal
 
 import torch
 import torch.nn as nn
 from cyclopts import Parameter
 from pydantic import BaseModel, ConfigDict
 from torch import distributed as dist
 from torch.distributed._functional_collectives import all_reduce
-from torch.distributed.device_mesh import DeviceMesh
 
 from xtuner.v1.utils.device import get_device
 
@@ -223,7 +222,7 @@ def forward(
 
         tokens_per_expert_global = tokens_per_expert.to(router_weights.dtype)  # (nlayers, ne)
         if self.loss_cfg.balancing_loss_global_average and dist.is_initialized():
-            tokens_per_expert_global = all_reduce(tokens_per_expert_global, "sum", dist.group.WORLD)
+            tokens_per_expert_global = all_reduce(tokens_per_expert_global, "sum", dist.group.WORLD)  # type: ignore
             tokens_global = tokens_per_expert_global.sum(-1)  # (nlayers, )
             seqlen_global = tokens_global // num_experts_per_tok
             routing_weights_sum_global = all_reduce_autograd(router_weights.sum(dim=1), "sum", dist.group.WORLD)
@@ -327,7 +326,9 @@ def forward(self, router_logits: torch.Tensor) -> torch.Tensor:
         if self.loss_cfg.z_loss_global_average and dist.is_initialized():
             unmasked_num = router_logits.shape[1]
             unmasked_num_rank = torch.tensor(unmasked_num, device=router_logits.device, dtype=torch.int64)
-            unmasked_num_global = all_reduce(unmasked_num_rank, "sum", dist.group.WORLD)
+            group = dist.group.WORLD
+            assert group is not None
+            unmasked_num_global = all_reduce(unmasked_num_rank, "sum", group)
             world_size = dist.get_world_size()
             loss = loss * unmasked_num * world_size / unmasked_num_global
 

xtuner/v1/model/base.py

@@ -8,7 +8,7 @@
 from itertools import chain
 from pathlib import Path
 from shutil import copy, copytree
-from typing import Annotated, Any, Generator, Iterable, Literal, Mapping, Sequence, TypedDict, cast
+from typing import Annotated, Any, Generator, Iterable, Literal, Mapping, Sequence, cast
 
 import torch
 import torch.distributed as dist
@@ -699,7 +699,8 @@ def build_loss_ctx_batch(
         if lm_loss_ctx_list is not None:
             loss_ctx_cls = lm_loss_ctx_list[0].__class__
             lm_loss_ctx_list = loss_ctx_cls.build_batches(
-                lm_loss_ctx_list, cu_seq_lens_list=cu_seq_lens_list, sp_mesh=sp_mesh)
+                lm_loss_ctx_list, cu_seq_lens_list=cu_seq_lens_list, sp_mesh=sp_mesh
+            )
 
             if lm_loss_ctx_list is not None:
                 for i, lm_loss_ctx in enumerate(lm_loss_ctx_list):
@@ -1864,10 +1865,7 @@ def _collect_full_state_dict(self, module: nn.Module):
         return ret
 
     def _build_loss_ctx(
-            self,
-            loss_ctx_cfg: BaseLossConfig | None,
-            data_batch: list[dict],
-            sp_mesh: DeviceMesh | None
+        self, loss_ctx_cfg: BaseLossConfig | None, data_batch: list[dict], sp_mesh: DeviceMesh | None
     ) -> list[BaseLossContext] | None:
         if loss_ctx_cfg is None:
             return None
@@ -1878,9 +1876,9 @@ def _build_loss_ctx(
         if first_loss_ctx is None:
             return None
         else:
-            ret = [first_loss_ctx] + [
-                loss_ctx_cfg.build(data=data, sp_mesh=sp_mesh) for data in data_batch[1:]]
-            return ret
+            ret = [first_loss_ctx] + [loss_ctx_cfg.build(data=data, sp_mesh=sp_mesh) for data in data_batch[1:]]
+            return ret  # type: ignore[return-value]
+
     # NOTE: Add this overload for inferring the return type for easier type checking and using
     @overload  # type: ignore
     def __call__(  # type: ignore

xtuner/v1/model/dense/dense.py

@@ -1,6 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from pathlib import Path
-from typing import Self, cast, Literal
+from typing import Self, cast
 
 import torch
 import torch.distributed as dist
@@ -19,7 +19,7 @@
 from xtuner.v1.config import FSDPConfig
 from xtuner.v1.data_proto import SequenceContext
 from xtuner.v1.float8.float8_handler import Float8Handler
-from xtuner.v1.loss import CELossContext, BaseLossContext
+from xtuner.v1.loss import BaseLossContext, CELossContext
 from xtuner.v1.model.base import (
     DEFAULT_FLOAT8_CFG,
     BaseModel,
@@ -78,7 +78,7 @@ def __init__(self, config: TransformerConfig):
     def forward(
         self,
         seq_ctx: SequenceContext,  # todo(@yehaochen): support intra layer micro-batch
-        loss_ctx: dict[Literal["lm"], BaseLossContext] | None = None,
+        loss_ctx: dict[str, BaseLossContext | list[BaseLossContext]] | None = None,
     ) -> ModelOutputs:
         input_ids = seq_ctx.input_ids
         position_ids = seq_ctx.position_ids
@@ -116,7 +116,7 @@ def forward(
             output["logits"] = logits
         else:
             # Training mode
-            loss, (logits, extra_info) = self.lm_head(hidden_states, loss_ctx["lm"])
+            loss, (logits, extra_info) = self.lm_head(hidden_states, loss_ctx["lm"])  # type: ignore[call-overload]
             output["loss"] = loss
             output["logits"] = logits
             output["extra_info"] = extra_info