move more things to kd plugin

src/axolotl/core/trainer_builder.py

@@ -47,7 +47,6 @@
     AxolotlTrainer,
     ReLoRATrainer,
 )
-from axolotl.core.trainers.kd import AxolotlKDTrainer
 from axolotl.core.training_args import (
     AxolotlCPOConfig,
     AxolotlDPOConfig,
@@ -77,7 +76,6 @@
 from axolotl.utils.chat_templates import get_chat_template_from_config
 from axolotl.utils.collators import (
     BatchSamplerDataCollatorForSeq2Seq,
-    DataCollatorForKD,
     DataCollatorForSeq2Seq,
     MambaDataCollator,
     V2BatchSamplerDataCollatorForSeq2Seq,
@@ -306,8 +304,6 @@ def _get_trainer_cls(self):
             return AxolotlMambaTrainer
         if self.cfg.reward_model:
             return AxolotlRewardTrainer
-        if self.cfg.trainer == "kd":
-            return AxolotlKDTrainer
         return AxolotlTrainer
 
     def build(self, total_num_steps):
@@ -797,7 +793,6 @@ def build_collator(
             Union[
                 V2BatchSamplerDataCollatorForSeq2Seq,
                 BatchSamplerDataCollatorForSeq2Seq,
-                DataCollatorForKD,
                 DataCollatorForSeq2Seq,
                 DataCollatorWithFlattening,
                 RewardDataCollatorWithPadding,
@@ -828,7 +823,9 @@ def build_collator(
                 collator_args.pop(0)
                 kwargs.pop("pad_to_multiple_of", None)
                 kwargs.pop("padding", None)
-            elif self.cfg.trainer == "kd":
+            elif self.cfg.kd_trainer:
+                from axolotl.integrations.kd.collator import DataCollatorForKD
+
                 collator = DataCollatorForKD
             else:
                 collator = DataCollatorForSeq2Seq

src/axolotl/integrations/kd/__init__.py

@@ -0,0 +1,22 @@
+"""
+Plugin init to add KD support to Axolotl.
+"""
+from axolotl.integrations.base import BasePlugin
+
+from .args import KDArgs  # pylint: disable=unused-import. # noqa: F401
+
+
+class KDPlugin(BasePlugin):
+    """
+    Plugin for KD support in Axolotl.
+    """
+
+    def get_input_args(self):
+        return "axolotl.integrations.kd.KDArgs"
+
+    def get_trainer_cls(self, cfg):
+        if cfg.kd_trainer:
+            from .trainer import AxolotlKDTrainer
+
+            return AxolotlKDTrainer
+        return None

src/axolotl/integrations/kd/args.py

@@ -0,0 +1,19 @@
+"""
+Plugin args for KD support.
+"""
+from typing import Optional
+
+from pydantic import BaseModel
+
+
+class KDArgs(BaseModel):
+    """
+    Input args for knowledge distillation.
+    """
+
+    kd_trainer: Optional[bool] = None  # whether to use KD trainer
+    kd_ce_alpha: Optional[
+        float
+    ] = None  # loss coefficient for cross-entropy loss during KD
+    kd_alpha: Optional[float] = None  # loss coefficient for KD loss
+    kd_temperature: Optional[float] = None  # temperature for sampling during KD

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -614,13 +614,6 @@ class Config:
         bool
     ] = None  # whether to use weighting in DPO trainer. If none, default is false in the trainer.
 
-    trainer: Optional[Literal["kd"]] = None
-    kd_ce_alpha: Optional[
-        float
-    ] = None  # loss coefficient for cross-entropy loss during KD
-    kd_alpha: Optional[float] = None  # loss coefficient for KD loss
-    kd_temperature: Optional[float] = None  # temperature for sampling during KD
-
     datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset], min_length=1)] = None  # type: ignore
     test_datasets: Optional[conlist(Union[SFTDataset, DPODataset, KTODataset], min_length=1)] = None  # type: ignore
     shuffle_merged_datasets: Optional[bool] = True