support resize embed for zero3

src/llmtuner/model/patcher.py

@@ -5,6 +5,7 @@ import random
 from types import MethodType
 from typing import TYPE_CHECKING, Any, Dict, List, Optional, Tuple
 from datasets import load_dataset
+from contextlib import nullcontext
 
 from transformers import BitsAndBytesConfig, GPTQConfig, PreTrainedModel, PreTrainedTokenizerBase
 from transformers.integrations import is_deepspeed_zero3_enabled
@@ -28,7 +29,7 @@ SUPPORTED_CLASS_FOR_S2ATTN = [] # TODO: add llama
 def _noisy_mean_initialization(embed_weight: torch.Tensor, num_new_tokens: int):
     embedding_dim = embed_weight.size(1)
     avg_weight = embed_weight[:-num_new_tokens].mean(dim=0, keepdim=True)
-    noise_weight = torch.empty_like(avg_weight[-num_new_tokens:])
+    noise_weight = torch.empty_like(embed_weight[-num_new_tokens:])
     noise_weight.normal_(mean=0, std=(1.0 / math.sqrt(embedding_dim)))
     embed_weight[-num_new_tokens:] = avg_weight + noise_weight
 
@@ -37,17 +38,31 @@ def _resize_embedding_layer(model: "PreTrainedModel", tokenizer: "PreTrainedToke
     r"""
     Resize token embeddings.
     """
-    current_embedding_size = model.get_input_embeddings().weight.size(0)
+    if is_deepspeed_zero3_enabled():
+        import deepspeed
+        with deepspeed.zero.GatheredParameters(model.get_input_embeddings().weight, modifier_rank=None):
+            current_embedding_size = model.get_input_embeddings().weight.size(0)
+    else:
+        current_embedding_size = model.get_input_embeddings().weight.size(0)
     if len(tokenizer) > current_embedding_size:
         if not isinstance(model.get_output_embeddings(), torch.nn.Linear):
             logger.warning("Current model does not support resizing token embeddings.")
             return
 
         model.resize_token_embeddings(len(tokenizer), pad_to_multiple_of=64)
-        new_embedding_size = model.get_input_embeddings().weight.size(0)
-        num_new_tokens = new_embedding_size - current_embedding_size
-        _noisy_mean_initialization(model.get_input_embeddings().weight.data, num_new_tokens)
-        _noisy_mean_initialization(model.get_output_embeddings().weight.data, num_new_tokens)
+        if is_deepspeed_zero3_enabled():
+            import deepspeed
+            params = [model.get_input_embeddings().weight]
+            if model.get_output_embeddings() is not None and not model.config.tie_word_embeddings:
+                params.append(model.get_output_embeddings().weight)
+            context = deepspeed.zero.GatheredParameters(params, modifier_rank=0)
+        else:
+            context = nullcontext()
+        with context:
+            new_embedding_size = model.get_input_embeddings().weight.size(0)
+            num_new_tokens = new_embedding_size - current_embedding_size
+            _noisy_mean_initialization(model.get_input_embeddings().weight.data, num_new_tokens)
+            _noisy_mean_initialization(model.get_output_embeddings().weight.data, num_new_tokens)
 
         logger.info("Resized token embeddings from {} to {}.".format(current_embedding_size, new_embedding_size))
 
@@ -264,9 +279,6 @@ def patch_model(
         setattr(model, "_keys_to_ignore_on_save", ["lm_head.weight"])
 
     if model_args.resize_vocab:
-        if is_deepspeed_zero3_enabled():
-            raise ValueError("DeepSpeed ZeRO-3 is incompatible with vocab resizing.")
-
         _resize_embedding_layer(model, tokenizer)
 
     if is_trainable: