Merge pull request #6362 from hiyouga/hiyouga/mllm_packing

[model] generalized packing

src/llamafactory/__init__.py

@@ -30,7 +30,7 @@
   longlora:
     transformers>=4.41.2,<=4.46.1
   packing:
-    transformers>=4.41.2,<=4.46.1
+    transformers>=4.43.0,<=4.46.1
 
 Disable version checking: DISABLE_VERSION_CHECK=1
 Enable VRAM recording: RECORD_VRAM=1

src/llamafactory/extras/constants.py

@@ -81,19 +81,6 @@
 
 STAGES_USE_PAIR_DATA = {"rm", "dpo"}
 
-SUPPORTED_CLASS_FOR_BLOCK_DIAG_ATTN = {
-    "cohere",
-    "falcon",
-    "gemma",
-    "gemma2",
-    "llama",
-    "mistral",
-    "phi",
-    "phi3",
-    "qwen2",
-    "starcoder2",
-}
-
 SUPPORTED_CLASS_FOR_S2ATTN = {"llama"}
 
 VIDEO_PLACEHOLDER = os.environ.get("VIDEO_PLACEHOLDER", "<video>")

src/llamafactory/model/model_utils/packing.py

@@ -44,13 +44,14 @@
 from transformers.utils.versions import require_version
 
 from ...extras import logging
-from ...extras.constants import SUPPORTED_CLASS_FOR_BLOCK_DIAG_ATTN
 from ...extras.packages import is_transformers_version_greater_than
 
 
-if TYPE_CHECKING:
-    from transformers import PretrainedConfig
+if is_transformers_version_greater_than("4.43.0"):
+    import transformers.modeling_flash_attention_utils
+
 
+if TYPE_CHECKING:
     from ...hparams import ModelArguments
 
 
@@ -113,45 +114,10 @@ def get_unpad_data(attention_mask: "torch.Tensor") -> Tuple["torch.Tensor", "tor
     return indices, cu_seqlens, max_seqlen_in_batch
 
 
-def _patch_for_block_diag_attn(model_type: str) -> None:
-    require_version("transformers>=4.41.2,<=4.46.1", "To fix: pip install transformers>=4.41.2,<=4.46.1")
-    if is_transformers_version_greater_than("4.43.0"):
-        import transformers.modeling_flash_attention_utils
-
-        transformers.modeling_flash_attention_utils._get_unpad_data = get_unpad_data
-        return
-
-    import transformers.models
-
-    if model_type == "cohere":
-        transformers.models.cohere.modeling_cohere._get_unpad_data = get_unpad_data
-    elif model_type == "falcon":
-        transformers.models.falcon.modeling_falcon._get_unpad_data = get_unpad_data
-    elif model_type == "gemma":
-        transformers.models.gemma.modeling_gemma._get_unpad_data = get_unpad_data
-    elif model_type == "gemma2":
-        transformers.models.gemma2.modeling_gemma2._get_unpad_data = get_unpad_data
-    elif model_type == "llama":
-        transformers.models.llama.modeling_llama._get_unpad_data = get_unpad_data
-    elif model_type == "mistral":
-        transformers.models.mistral.modeling_mistral._get_unpad_data = get_unpad_data
-    elif model_type == "phi":
-        transformers.models.phi.modeling_phi._get_unpad_data = get_unpad_data
-    elif model_type == "phi3":
-        transformers.models.phi3.modeling_phi3._get_unpad_data = get_unpad_data
-    elif model_type == "qwen2":
-        transformers.models.qwen2.modeling_qwen2._get_unpad_data = get_unpad_data
-    elif model_type == "starcoder2":
-        transformers.models.starcoder2.modeling_starcoder2._get_unpad_data = get_unpad_data
-
-
-def configure_packing(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None:
+def configure_packing(model_args: "ModelArguments", is_trainable: bool) -> None:
     if not is_trainable or not model_args.block_diag_attn:
         return
 
-    model_type = getattr(config, "model_type", None)
-    if model_type in SUPPORTED_CLASS_FOR_BLOCK_DIAG_ATTN:
-        _patch_for_block_diag_attn(model_type)
-        logger.info_rank0("Using block diagonal attention for sequence packing without cross-attention.")
-    else:
-        raise ValueError("Current model does not support block diagonal attention.")
+    require_version("transformers>=4.43.0,<=4.46.1", "To fix: pip install transformers>=4.43.0,<=4.46.1")
+    transformers.modeling_flash_attention_utils._get_unpad_data = get_unpad_data
+    logger.info_rank0("Using block diagonal attention for sequence packing without cross-attention.")

src/llamafactory/model/patcher.py

@@ -96,7 +96,7 @@ def patch_config(
     configure_quantization(config, tokenizer, model_args, init_kwargs)
     configure_moe(config, model_args, is_trainable)
     configure_visual_model(config)
-    configure_packing(config, model_args, is_trainable)
+    configure_packing(model_args, is_trainable)
 
     if model_args.use_cache and not is_trainable:
         setattr(config, "use_cache", True)