👐 DeepSpeed integration for GRPO (#2652)

trl/models/__init__.py

@@ -20,7 +20,7 @@
 _import_structure = {
     "modeling_base": ["GeometricMixtureWrapper", "PreTrainedModelWrapper", "create_reference_model"],
     "modeling_value_head": ["AutoModelForCausalLMWithValueHead", "AutoModelForSeq2SeqLMWithValueHead"],
-    "utils": ["SUPPORTED_ARCHITECTURES", "setup_chat_format", "unwrap_model_for_generation"],
+    "utils": ["SUPPORTED_ARCHITECTURES", "prepare_deepspeed", "setup_chat_format", "unwrap_model_for_generation"],
 }
 
 try:
@@ -39,7 +39,7 @@
 if TYPE_CHECKING:
     from .modeling_base import GeometricMixtureWrapper, PreTrainedModelWrapper, create_reference_model
     from .modeling_value_head import AutoModelForCausalLMWithValueHead, AutoModelForSeq2SeqLMWithValueHead
-    from .utils import SUPPORTED_ARCHITECTURES, setup_chat_format, unwrap_model_for_generation
+    from .utils import SUPPORTED_ARCHITECTURES, prepare_deepspeed, setup_chat_format, unwrap_model_for_generation
 
     try:
         if not is_diffusers_available():

trl/models/utils.py

@@ -14,6 +14,7 @@
 
 import itertools
 from contextlib import contextmanager
+from copy import deepcopy
 from dataclasses import dataclass
 from typing import TYPE_CHECKING, Literal, Optional, Union
 
@@ -193,3 +194,37 @@ def unwrap_model_for_generation(
                 add_hooks(model)
     else:
         yield unwrapped_model
+
+
+def prepare_deepspeed(model, accelerator):
+    # Adapted from accelerate: https://github.com/huggingface/accelerate/blob/739b135f8367becb67ffaada12fe76e3aa60fefd/src/accelerate/accelerator.py#L1473
+    deepspeed_plugin = accelerator.state.deepspeed_plugin
+    config_kwargs = deepcopy(deepspeed_plugin.deepspeed_config)
+    stage = config_kwargs["zero_optimization"]["stage"]
+
+    if model is not None:
+        hidden_size = (
+            max(model.config.hidden_sizes)
+            if getattr(model.config, "hidden_sizes", None)
+            else getattr(model.config, "hidden_size", None)
+        )
+        if hidden_size is not None and stage == 3:
+            # Note that `stage3_prefetch_bucket_size` can produce DeepSpeed messages like: `Invalidate trace cache
+            # @ step 0: expected module 1, but got module 0`
+            # This is expected and is not an error, see: https://github.com/microsoft/DeepSpeed/discussions/4081
+            config_kwargs.update(
+                {
+                    "zero_optimization.reduce_bucket_size": hidden_size * hidden_size,
+                    "zero_optimization.stage3_param_persistence_threshold": 10 * hidden_size,
+                    "zero_optimization.stage3_prefetch_bucket_size": 0.9 * hidden_size * hidden_size,
+                }
+            )
+
+    # If ZeRO-3 is used, we shard both the active and reference model.
+    # Otherwise, we assume the reference model fits in memory and is initialized on each device with ZeRO
+    # disabled (stage 0)
+    if stage != 3:
+        config_kwargs["zero_optimization"]["stage"] = 0
+    model, *_ = deepspeed.initialize(model=model, config=config_kwargs)
+    model.eval()
+    return model

trl/trainer/grpo_trainer.py

@@ -32,10 +32,11 @@
     TrainerCallback,
     is_wandb_available,
 )
+from transformers.integrations.deepspeed import is_deepspeed_zero3_enabled
 from transformers.utils import is_peft_available
 
 from ..data_utils import apply_chat_template, is_conversational, maybe_apply_chat_template
-from ..models import create_reference_model, unwrap_model_for_generation
+from ..models import create_reference_model, prepare_deepspeed, unwrap_model_for_generation
 from .grpo_config import GRPOConfig
 from .utils import generate_model_card, get_comet_experiment_url
 
@@ -158,6 +159,7 @@ def __init__(
         # Trained model
         model_init_kwargs = args.model_init_kwargs or {}
         if isinstance(model, str):
+            model_id = model
             torch_dtype = model_init_kwargs.get("torch_dtype")
             if isinstance(torch_dtype, torch.dtype) or torch_dtype == "auto" or torch_dtype is None:
                 pass  # torch_dtype is already a torch.dtype or "auto" or None
@@ -171,6 +173,7 @@ def __init__(
                 )
             model = AutoModelForCausalLM.from_pretrained(model, **model_init_kwargs)
         else:
+            model_id = model.config._name_or_path
             if args.model_init_kwargs is not None:
                 raise ValueError(
                     "You passed `model_init_kwargs` to the `GRPOConfig`, but your model is already instantiated. "
@@ -181,7 +184,9 @@ def __init__(
             model = get_peft_model(model, peft_config)
 
         # Reference model
-        if peft_config is None:
+        if is_deepspeed_zero3_enabled():
+            self.ref_model = AutoModelForCausalLM.from_pretrained(model_id, **model_init_kwargs)
+        elif peft_config is None:
             # If PEFT configuration is not provided, create a reference model based on the initial model.
             self.ref_model = create_reference_model(model)
         else:
@@ -269,7 +274,10 @@ def data_collator(features):  # No data collation is needed in GRPO
         self.model_accepts_loss_kwargs = False
 
         if self.ref_model is not None:
-            self.ref_model = self.accelerator.prepare_model(self.ref_model, evaluation_mode=True)
+            if self.is_deepspeed_enabled:
+                self.ref_model = prepare_deepspeed(self.ref_model, self.accelerator)
+            else:
+                self.ref_model = self.accelerator.prepare_model(self.ref_model, evaluation_mode=True)
 
         for i, reward_func in enumerate(self.reward_funcs):
             if isinstance(reward_func, PreTrainedModel):