fusion-bench 0.2.20__py3-none-any.whl → 0.2.22__py3-none-any.whl

fusion_bench/models/modeling_smile_mistral/__init__.py

@@ -1,48 +1,6 @@
-# flake8: noqa F401
-from typing import TYPE_CHECKING
-
-from transformers.utils.import_utils import (
-    OptionalDependencyNotAvailable,
-    _LazyModule,
-    is_flax_available,
-    is_tf_available,
-    is_torch_available,
+from . import register
+from .configuration_smile_mistral import SmileMistralConfig
+from .modeling_smile_mistral import (
+    SmileMistralForCausalLM,
+    SmileMistralModel,
 )
-
-_import_structure = {
-    "configuration_smile_mistral": ["SmileMistralConfig"],
-}
-
-try:
-    if not is_torch_available():
-        raise OptionalDependencyNotAvailable()
-except OptionalDependencyNotAvailable:
-    pass
-else:
-    _import_structure["modeling_smile_mistral"] = [
-        "SmileMistralForCausalLM",
-        "SmileMistralModel",
-        "SmileMistralPreTrainedModel",
-    ]
-
-if TYPE_CHECKING:
-    from .configuration_smile_mistral import SmileMistralConfig
-
-    try:
-        if not is_torch_available():
-            raise OptionalDependencyNotAvailable()
-    except OptionalDependencyNotAvailable:
-        pass
-    else:
-        from .modeling_smile_mistral import (
-            SmileMistralForCausalLM,
-            SmileMistralModel,
-            SmileMistralPreTrainedModel,
-        )
-
-else:
-    import sys
-
-    sys.modules[__name__] = _LazyModule(
-        __name__, globals()["__file__"], _import_structure, module_spec=__spec__
-    )

fusion_bench/models/modeling_smile_qwen2/__init__.py

@@ -5,4 +5,4 @@ from .modeling_smile_qwen2 import (
     SmileQwen2ForQuestionAnswering,
     SmileQwen2ForSequenceClassification,
     SmileQwen2Model,
-)
+)

fusion_bench/models/modeling_smile_qwen2/modeling_smile_qwen2.py

@@ -24,7 +24,6 @@ from transformers.modeling_outputs import (
 from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
 from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
 from transformers.models.qwen2.modeling_qwen2 import (
-    QWEN2_INPUTS_DOCSTRING,
     Qwen2RMSNorm,
     Qwen2RotaryEmbedding,
     apply_rotary_pos_emb,
@@ -32,7 +31,6 @@ from transformers.models.qwen2.modeling_qwen2 import (
 )
 from transformers.processing_utils import Unpack
 from transformers.utils import (
-    LossKwargs,
     add_code_sample_docstrings,
     add_start_docstrings,
     add_start_docstrings_to_model_forward,
@@ -314,7 +312,6 @@ class SmileQwen2Model(SmileQwen2PreTrainedModel):
         self.embed_tokens = value
 
     @can_return_tuple
-    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -609,9 +606,6 @@ class SmileQwen2Model(SmileQwen2PreTrainedModel):
         return causal_mask
 
 
-class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs): ...
-
-
 class SmileQwen2ForCausalLM(SmileQwen2PreTrainedModel, GenerationMixin):
     _tied_weights_keys = ["lm_head.weight"]
     _tp_plan = {"lm_head": "colwise_rep"}
@@ -646,7 +640,6 @@ class SmileQwen2ForCausalLM(SmileQwen2PreTrainedModel, GenerationMixin):
 
     @can_return_tuple
     @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
-    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     @replace_return_docstrings(
         output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC
     )
@@ -663,7 +656,7 @@ class SmileQwen2ForCausalLM(SmileQwen2PreTrainedModel, GenerationMixin):
         output_hidden_states: Optional[bool] = None,
         cache_position: Optional[torch.LongTensor] = None,
         logits_to_keep: Union[int, torch.Tensor] = 0,
-        **kwargs: Unpack[KwargsForCausalLM],
+        **kwargs,
     ) -> CausalLMOutputWithPast:
         r"""
             labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
@@ -752,7 +745,9 @@ class SmileQwen2ForSequenceClassification(SmileQwen2PreTrainedModel):
     def __init__(self, config):
         super().__init__(config)
         self.num_labels = config.num_labels
-        self.model = SmileQwen2Model(config)  #* replace Qwen2Model with SmileQwen2Model
+        self.model = SmileQwen2Model(
+            config
+        )  # * replace Qwen2Model with SmileQwen2Model
         self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
 
         # Initialize weights and apply final processing
@@ -765,7 +760,6 @@ class SmileQwen2ForSequenceClassification(SmileQwen2PreTrainedModel):
         self.model.embed_tokens = value
 
     @can_return_tuple
-    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,
@@ -852,7 +846,9 @@ class SmileQwen2ForQuestionAnswering(SmileQwen2PreTrainedModel):
 
     def __init__(self, config):
         super().__init__(config)
-        self.transformer = SmileQwen2Model(config)  #* replace Qwen2Model with SmileQwen2Model
+        self.transformer = SmileQwen2Model(
+            config
+        )  # * replace Qwen2Model with SmileQwen2Model
         self.qa_outputs = nn.Linear(config.hidden_size, 2)
 
         # Initialize weights and apply final processing
@@ -865,7 +861,6 @@ class SmileQwen2ForQuestionAnswering(SmileQwen2PreTrainedModel):
         self.transformer.embed_tokens = value
 
     @can_return_tuple
-    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
     def forward(
         self,
         input_ids: Optional[torch.LongTensor] = None,

fusion_bench/models/modeling_smile_qwen2/register.py

@@ -1,10 +1,7 @@
 from transformers import AutoConfig, AutoModel, AutoModelForCausalLM
 
 from .configuration_smile_qwen2 import SmileQwen2Config
-from .modeling_smile_qwen2 import (
-    SmileQwen2ForCausalLM,
-    SmileQwen2Model,
-)
+from .modeling_smile_qwen2 import SmileQwen2ForCausalLM, SmileQwen2Model
 
 AutoConfig.register("smile_qwen2", SmileQwen2Config)
 AutoModel.register(SmileQwen2Config, SmileQwen2Model)

fusion_bench/models/parameter_dict.py

@@ -74,7 +74,7 @@ class ParameterDictModel(nn.Module):
                     name.split("."),
                     param,
                     check_parent=True,
-                    parent_builder=self.__class__,
+                    parent_builder=__class__,
                 )
 
     def __repr__(self):

fusion_bench/models/sparse_we_moe.py

@@ -11,6 +11,7 @@ from torch.func import functional_call
 from torch.nn import functional as F
 from tqdm.auto import tqdm
 
+from fusion_bench.models.utils import del_attr, get_attr, set_attr
 from fusion_bench.utils.state_dict_arithmetic import (
     state_dict_sub,
     state_dict_weighted_sum,
@@ -20,59 +21,6 @@ from fusion_bench.utils.type import StateDictType
 log = logging.getLogger(__name__)
 
 
-def join_list(list_of_list: List[List]):
-    ans = []
-    for l in list_of_list:
-        ans.extend(l)
-    return ans
-
-
-def del_attr(obj, names: List[str]):
-    """
-    Deletes an attribute from an object recursively.
-
-    Args:
-        obj (object): Object to delete attribute from.
-        names (list): List of attribute names to delete recursively.
-    """
-    if len(names) == 1:
-        delattr(obj, names[0])
-    else:
-        del_attr(getattr(obj, names[0]), names[1:])
-
-
-def set_attr(obj, names: List[str], val):
-    """
-    Sets an attribute of an object recursively.
-
-    Args:
-        obj (object): Object to set attribute of.
-        names (list): List of attribute names to set recursively.
-        val (object): Value to set the attribute to.
-    """
-    if len(names) == 1:
-        setattr(obj, names[0], val)
-    else:
-        set_attr(getattr(obj, names[0]), names[1:], val)
-
-
-def get_attr(obj, names: List[str]):
-    """
-    Gets an attribute of an object recursively.
-
-    Args:
-        obj (object): Object to get attribute of.
-        names (list): List of attribute names to get recursively.
-
-    Returns:
-        object: The attribute of the object.
-    """
-    if len(names) == 1:
-        return getattr(obj, names[0])
-    else:
-        return get_attr(getattr(obj, names[0]), names[1:])
-
-
 class Depth_0_Gate(nn.Module):
     def __init__(self, num_experts: int):
         super().__init__()

fusion_bench/models/utils.py

@@ -3,6 +3,8 @@ from typing import List
 import torch
 from torch import nn
 
+from fusion_bench.utils.type import StateDictType
+
 
 def del_attr(obj, names: List[str]):
     """
@@ -50,6 +52,30 @@ def get_attr(obj, names: List[str]):
         return get_attr(getattr(obj, names[0]), names[1:])
 
 
+def check_parameterNamesMatch(checkpoints: List[StateDictType]) -> None:
+    """
+    Checks that the parameter names of the given checkpoints match.
+
+    Args:
+        checkpoints (List[Dict[str, float]]): A list of checkpoints, where each checkpoint is a dictionary of parameter names and their corresponding values.
+
+    Raises:
+        ValueError: If the number of checkpoints is less than 2 or if the parameter names of any two checkpoints differ.
+
+    """
+    parameter_names = set(checkpoints[0].keys())
+
+    if len(checkpoints) >= 2:
+        # raise ValueError("Number of models is less than 2.")
+        for checkpoint in checkpoints[1:]:
+            current_parameterNames = set(checkpoint.keys())
+            if current_parameterNames != parameter_names:
+                raise ValueError(
+                    "Differing parameter names in models. "
+                    f"The different parameters are {parameter_names.symmetric_difference(current_parameterNames)}"
+                )
+
+
 def find_layers_with_type(
     module: nn.Module,
     layer_types=[nn.Linear],

fusion_bench/models/we_moe.py

@@ -8,64 +8,12 @@ from torch import Tensor, nn
 from torch.func import functional_call
 from torch.nn import functional as F
 
+from fusion_bench.models.utils import del_attr, get_attr, set_attr
 from fusion_bench.utils.type import StateDictType
 
 log = logging.getLogger(__name__)
 
 
-def join_list(list_of_list: List[List]):
-    ans = []
-    for l in list_of_list:
-        ans.extend(l)
-    return ans
-
-
-def del_attr(obj, names: List[str]):
-    """
-    Deletes an attribute from an object recursively.
-
-    Args:
-        obj (object): Object to delete attribute from.
-        names (list): List of attribute names to delete recursively.
-    """
-    if len(names) == 1:
-        delattr(obj, names[0])
-    else:
-        del_attr(getattr(obj, names[0]), names[1:])
-
-
-def set_attr(obj, names: List[str], val):
-    """
-    Sets an attribute of an object recursively.
-
-    Args:
-        obj (object): Object to set attribute of.
-        names (list): List of attribute names to set recursively.
-        val (object): Value to set the attribute to.
-    """
-    if len(names) == 1:
-        setattr(obj, names[0], val)
-    else:
-        set_attr(getattr(obj, names[0]), names[1:], val)
-
-
-def get_attr(obj, names: List[str]):
-    """
-    Gets an attribute of an object recursively.
-
-    Args:
-        obj (object): Object to get attribute of.
-        names (list): List of attribute names to get recursively.
-
-    Returns:
-        object: The attribute of the object.
-    """
-    if len(names) == 1:
-        return getattr(obj, names[0])
-    else:
-        return get_attr(getattr(obj, names[0]), names[1:])
-
-
 class Depth_0_Gate(nn.Module):
     def __init__(self, num_experts: int):
         super().__init__()

fusion_bench/models/wrappers/ensemble.py

@@ -1,4 +1,4 @@
-from typing import Any, Callable, Dict, List, cast
+from typing import Any, Callable, Dict, List, Union, cast
 
 import numpy as np
 import torch
@@ -6,7 +6,9 @@ from omegaconf import ListConfig
 from torch import Tensor, nn
 
 
-def aggregate_tensors(outputs: List[Any], aggregate_fn: Callable) -> Tensor:
+def aggregate_tensors(
+    outputs: List[Any], aggregate_fn: Callable
+) -> Union[Tensor, Dict, List, None]:
     """
     Aggregates a list of outputs using the provided aggregation function.
 
@@ -84,7 +86,7 @@ class EnsembleModule(nn.Module):
         """
         return torch.stack(outputs).mean(dim=0)
 
-    def forward(self, *args, **kwargs):
+    def forward(self, *args: Any, **kwargs: Any) -> Any:
         """
         Performs a forward pass by averaging the outputs of the models.
 
@@ -150,7 +152,7 @@ class WeightedEnsembleModule(nn.Module):
         weights = cast(Tensor, self.weights).view(-1, *([1] * outputs[0].dim()))
         return (torch.stack(outputs) * weights).sum(dim=0)
 
-    def forward(self, *args, **kwargs):
+    def forward(self, *args: Any, **kwargs: Any) -> Any:
         """
         Performs a forward pass by computing the weighted average of the models' outputs.
 

fusion_bench/models/wrappers/layer_wise_fusion.py

@@ -49,7 +49,7 @@ def get_layer_wise_weights(
     return torch.full((num_models, num_layers), init_values, dtype=dtype)
 
 
-def _fuse_weights(layer_wise_weight: Tensor, tensors: List[Tensor]):
+def _fuse_weights(layer_wise_weight: Tensor, tensors: List[Tensor]) -> Tensor:
     """
     Fuse the layer-wise weights with the given state dictionaries.