build cycle multi-hop rag, trainable, added component graph with cycl…

…e, output graph as simplified version of the dynamic computation graph

adalflow/adalflow/core/generator.py

@@ -506,6 +506,7 @@ def forward(
                         self.model_kwargs, model_kwargs
                     ),
                 }
+
                 output = self.call(**input_args, id=id)
         # 2. Generate a Parameter object from the output
         combined_prompt_kwargs = compose_model_kwargs(self.prompt_kwargs, prompt_kwargs)
@@ -527,9 +528,12 @@ def forward(
             name=self.name + "_output",
             role_desc=f"Output from (llm) {self.name}",
             param_type=ParameterType.GENERATOR_OUTPUT,
+            data_id=id,
         )
         response.set_predecessors(predecessors)
-        response.trace_forward_pass(input_args=input_args, full_response=output)
+        response.trace_forward_pass(
+            input_args=input_args, full_response=output, id=self.id, name=self.name
+        )
         # *** special to the generator ***
         response.trace_api_kwargs(api_kwargs=self._trace_api_kwargs)
         # attach the demo to the demo parameter
@@ -755,6 +759,7 @@ def _backward_through_one_predecessor(
             score=response._score,  # add score to gradient
             param_type=ParameterType.GRADIENT,
             from_response_id=response.id,
+            data_id=response.data_id,
         )
         pred.add_gradient(var_gradient)
         pred.set_score(response._score)

adalflow/adalflow/core/retriever.py

@@ -128,7 +128,13 @@ def forward(
         )
         if input is None:
             raise ValueError("Input cannot be empty")
-        response = super().forward(input, top_k=top_k, **kwargs)
+        response = super().forward(input, top_k=top_k, id=id, **kwargs)
+        response.trace_forward_pass(
+            input_args={"input": input, "top_k": top_k},
+            full_response=response,
+            id=self.id,
+            name=self.name,
+        )
         response.param_type = (
             ParameterType.RETRIEVER_OUTPUT
         )  # be more specific about the type

adalflow/adalflow/optim/grad_component.py

@@ -2,6 +2,7 @@
 
 from typing import TYPE_CHECKING
 from collections import OrderedDict
+import uuid
 import logging
 
 if TYPE_CHECKING:
@@ -13,6 +14,7 @@
 from adalflow.core.component import Component
 from adalflow.optim.function import BackwardContext
 
+
 __all__ = ["GradComponent"]
 log = logging.getLogger(__name__)
 
@@ -31,10 +33,12 @@ class GradComponent(Component):
     """
     backward_engine: "BackwardEngine"
     _component_type = "grad"
+    id = None
 
     def __init__(self, *args, **kwargs):
         super().__init__()
         super().__setattr__("backward_engine", None)
+        super().__setattr__("id", str(uuid.uuid4()))
 
     def __call__(self, *args, **kwargs):
         if self.training:
@@ -122,9 +126,15 @@ def forward(self, *args, **kwargs) -> "Parameter":
             name=self.name + "_output",
             role_desc=self.name + " response",
             param_type=ParameterType.OUTPUT,
+            data_id=kwargs.get("id", None),
         )
         response.set_predecessors(predecessors)
-        response.trace_forward_pass(input_args=input_args, full_response=call_response)
+        response.trace_forward_pass(
+            input_args=input_args,
+            full_response=call_response,
+            id=self.id,
+            name=self.name,
+        )
         response.set_grad_fn(
             BackwardContext(
                 backward_fn=self.backward,