Merge pull request #223 from SylphAI-Inc/li

[Multi-hop RAG opt + improvement] [V0.3.0 Cycle Graph]

Author: Sylph-AI

adalflow/CHANGELOG.md

@@ -1,3 +1,9 @@
+## [0.2.7] - 2024-09-23
+
+### Improved
+- Better diagnose report for `Trainer.diagnose`.
+- Multi-hop RAG with handling of Cycle.
+
 ## [0.2.7] - TO Be Released
 ### Added
 - `Memory` is completed with `call` and `add_dialog_turn` methods.

adalflow/adalflow/__init__.py

@@ -1,7 +1,7 @@
 __version__ = "0.2.6"
 
 from adalflow.core.component import Component, fun_to_component
-from adalflow.core.container import Sequential
+from adalflow.core.container import Sequential, ComponentList
 from adalflow.core.base_data_class import DataClass, DataClassFormatType, required_field
 
 from adalflow.optim.grad_component import GradComponent
@@ -63,6 +63,10 @@
     BedrockAPIClient,
 )
 
+# data pipeline
+from adalflow.components.data_process.text_splitter import TextSplitter
+from adalflow.components.data_process.data_components import ToEmbeddings
+
 __all__ = [
     "Component",
     "fun_to_component",
@@ -72,7 +76,10 @@
     "required_field",
     # Container
     "Sequential",
+    "ComponentList",
+    # Grad Component
     "GradComponent",
+    # Functional Component
     "ModelClient",
     "Generator",
     "Embedder",
@@ -99,6 +106,9 @@
     "JsonOutputParser",
     "ListOutputParser",
     "DataClassParser",
+    # Data Pipeline
+    "TextSplitter",
+    "ToEmbeddings",
     # Types
     "GeneratorOutput",
     "EmbedderOutput",

adalflow/adalflow/core/__init__.py

@@ -1,7 +1,7 @@
 from .base_data_class import DataClass, required_field, DataClassFormatType
 
 from .component import Component, FunComponent, fun_to_component
-from .container import Sequential
+from .container import Sequential, ComponentList
 from .db import LocalDB
 from .default_prompt_template import DEFAULT_ADALFLOW_SYSTEM_PROMPT
 from .embedder import Embedder, BatchEmbedder
@@ -50,6 +50,7 @@
     "LocalDB",
     "Component",
     "Sequential",
+    "ComponentList",
     "FunComponent",
     "fun_to_component",
     "DataClass",

adalflow/adalflow/core/base_data_class.py

@@ -356,8 +356,6 @@ class TrecDataList(DataClass):
 
         return dict(ordered_dict)
 
-        return ordered_dict
-
     @classmethod
     def from_dict(cls, data: Dict[str, Any]) -> "DataClass":
         """Create a dataclass instance from a dictionary.

adalflow/adalflow/core/component.py

@@ -167,6 +167,7 @@ def use_teacher(self, mode: bool = True):
             component.use_teacher(mode)
         return self
 
+    # TODO: reassese trace, it should be turned on maybe all the time
     def trace(self, mode: bool = True):
         r"""Sets the component in tracing mode.This signal will be used in forward and backward to accumulate input and output."""
         if not isinstance(mode, bool):

adalflow/adalflow/core/container.py

@@ -1,14 +1,63 @@
-"""Container component for composing multiple components, such as Sequential."""
-
-from collections import OrderedDict
+"""
+Container component for composing multiple components, such as Sequential
+and ComponentList.
+
+This design draws inspiration from PyTorch’s modular
+container patterns, including `nn.Sequential` and `nn.ModuleList`. The
+`Container` component allows for grouping several components into one, enabling
+flexible and reusable model architectures.
+
+Design Motivation:
+-------------------
+This implementation follows the same principles as PyTorch’s component-based
+design, encouraging modularity, reusability, and extensibility. The `Container`
+component provides an easy way to manage multiple layers or other components,
+while ensuring that their parameters are properly registered and updated during
+training.
+
+Credits:
+---------
+The design of this component takes inspiration from the PyTorch project
+(https://pytorch.org). PyTorch is an open-source deep learning framework,
+licensed under a BSD-style license. Although this code is not part of the
+official PyTorch library, it mirrors the same design principles.
+
+For more details on PyTorch’s licensing, refer to:
+https://github.com/pytorch/pytorch/blob/main/LICENSE
+
+Usage Example:
+--------------
+    class MyModule(nn.Module):
+        def __init__(self):
+            super().__init__()
+
+            self.model = nn.Sequential(
+                  nn.Conv2d(1,20,5),
+                  nn.ReLU(),
+                  nn.Conv2d(20,64,5),
+                  nn.ReLU()
+                )
+            self.linears = nn.ModuleList([nn.Linear(10, 10) for i in range(10)])
+
+        def forward(self, x):
+            # ModuleList can act as an iterable, or be indexed using ints
+            for i, l in enumerate(self.linears):
+                x = self.linears[i // 2](x) + l(x)
+            return x
+
+"""
+
+from collections import OrderedDict, abc as container_abcs
 import operator
-from itertools import islice
-from typing import TypeVar, Dict, Union, Iterable, Iterator, Any, overload
+from itertools import islice, chain
+from typing import TypeVar, Dict, Union, Iterable, Iterator, Any, overload, Optional
 
 from adalflow.core.component import Component
 
 T = TypeVar("T", bound=Component)
 
+__all__ = ["Sequential", "ComponentList"]
+
 
 class Sequential(Component):
     __doc__ = r"""A sequential container.
@@ -311,3 +360,177 @@ def extend(self, components: Iterable[Component]) -> "Sequential":
         for component in components:
             self.append(component)
         return self
+
+
+def _addindent(s_: str, numSpaces: int):
+    s = s_.split("\n")
+    # don't do anything for single-line stuff
+    if len(s) == 1:
+        return s_
+    first = s.pop(0)
+    s = [(numSpaces * " ") + line for line in s]
+    s = "\n".join(s)
+    s = first + "\n" + s
+    return s
+
+
+class ComponentList(Component):
+    __doc__ = r"""Holds subcomponents in a list.
+
+    :class:`adalflow.core.ComponentList` can be indexed like a regular Python list, but
+    the components it holds are properly registered, and will be visible by all
+    :class:`adalflow.core.Component` methods.
+
+    Args:
+        components (iterable, optional): an iterable of components to add
+
+    Examples:
+
+    .. code-block:: python
+
+        # Example of how to use ComponentList
+        class MyComponents(Component):
+            def __init__(self):
+                super().__init__()
+                self.llms = ComponentList([adal.Generator() for i in range(10)])
+
+            def forward(self, x):
+                for layer in self.layers:
+                    x = layer(x)
+                return x
+    """
+    _components: Dict[str, Component] = OrderedDict()
+
+    def __init__(self, components: Optional[Iterable[Component]] = None) -> None:
+        super().__init__()
+        if components is not None:
+            self += components
+
+    def _get_abs_string_index(self, idx):
+        """Get the absolute index as a string."""
+        idx = operator.index(idx)
+        if not (-len(self) <= idx < len(self)):
+            raise IndexError(f"index {idx} is out of range")
+        if idx < 0:
+            idx += len(self)
+        return str(idx)
+
+    def __getitem__(self, idx: Union[int, slice]) -> Union[Component, "ComponentList"]:
+        """Retrieve a component or a slice of components."""
+        if isinstance(idx, slice):
+            return self.__class__(list(self._components.values())[idx])
+        else:
+            return self._components[self._get_abs_string_index(idx)]
+
+    def __setitem__(self, idx: int, component: Component) -> None:
+        """Set a component at the given index."""
+        idx = self._get_abs_string_index(idx)
+        return setattr(self, str(idx), component)
+
+    def __delitem__(self, idx: Union[int, slice]) -> None:
+        """Delete a component or a slice of components."""
+        if isinstance(idx, slice):
+            for k in range(len(self._components))[idx]:
+                delattr(self, str(k))
+        else:
+            delattr(self, self._get_abs_string_index(idx))
+        # To preserve numbering, self._components is being reconstructed with modules after deletion
+        str_indices = [str(i) for i in range(len(self._components))]
+        self._components = OrderedDict(
+            list(zip(str_indices, self._components.values()))
+        )
+
+    def __len__(self) -> int:
+        """Return the number of components."""
+        return len(self._components)
+
+    def __iter__(self) -> Iterator[Component]:
+        """Iterate over the components."""
+        return iter(self._components.values())
+
+    def __iadd__(self, components: Iterable[Component]) -> "ComponentList":
+        """Add multiple components using the `+=` operator."""
+
+        return self.extend(components)
+
+    def __add__(self, other: Iterable[Component]) -> "ComponentList":
+        """Concatenate two ComponentLists."""
+
+        combined = ComponentList()
+        for i, component in enumerate(chain(self, other)):
+            combined.add_component(str(i), component)
+        return combined
+
+    def __repr__(self):
+        """Return a custom repr for ModuleList that compresses repeated module representations."""
+        list_of_reprs = [repr(item) for item in self]
+        if len(list_of_reprs) == 0:
+            return self._get_name() + "()"
+
+        start_end_indices = [[0, 0]]
+        repeated_blocks = [list_of_reprs[0]]
+        for i, r in enumerate(list_of_reprs[1:], 1):
+            if r == repeated_blocks[-1]:
+                start_end_indices[-1][1] += 1
+                continue
+
+            start_end_indices.append([i, i])
+            repeated_blocks.append(r)
+
+        lines = []
+        main_str = self._get_name() + "("
+        for (start_id, end_id), b in zip(start_end_indices, repeated_blocks):
+            local_repr = f"({start_id}): {b}"  # default repr
+
+            if start_id != end_id:
+                n = end_id - start_id + 1
+                local_repr = f"({start_id}-{end_id}): {n} x {b}"
+
+            local_repr = _addindent(local_repr, 2)
+            lines.append(local_repr)
+
+        main_str += "\n  " + "\n  ".join(lines) + "\n"
+        main_str += ")"
+        return main_str
+
+    def __dir__(self):
+        keys = super().__dir__()
+        keys = [key for key in keys if not key.isdigit()]
+        return keys
+
+    def insert(self, index: int, component: Component) -> None:
+        """Insert a component at the specified index."""
+        for i in range(len(self._components), index, -1):
+            self._components[str(i)] = self._components[str(i - 1)]
+        self._components[str(index)] = component
+
+    def pop(self, index: Union[int, slice]) -> Component:
+        """Remove and return a component at the given index."""
+        component = self[index]
+        del self[index]
+        return component
+
+    def append(self, component: Component) -> "ComponentList":
+        """Append a component to the list."""
+        # self._components[str(len(self))] = component
+        self.add_component(str(len(self)), component)
+        return self
+
+    def extend(self, components: Iterable[Component]) -> "ComponentList":
+        """Extend the list by appending multiple components."""
+        # for component in components:
+        #     self.append(component)
+        # return self
+
+        if not isinstance(components, container_abcs.Iterable):
+            raise TypeError(
+                "ModuleList.extend should be called with an "
+                "iterable, but got " + type(components).__name__
+            )
+        offset = len(self)
+        for i, component in enumerate(components):
+            self.add_component(str(offset + i), component)
+        return self
+
+
+# TODO: need to do the same to ParameterList and ParameterDict, ModuleDict