Update instruction refinement

pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "hatchling.build"
 [project]
 name = "draive"
 description = "Framework designed to simplify and accelerate the development of LLM-based applications."
-version = "0.75.3"
+version = "0.75.4"
 readme = "README.md"
 maintainers = [
   { name = "Kacper Kaliński", email = "kacper.kalinski@miquido.com" },

src/draive/helpers/instruction_refinement.py

@@ -490,7 +490,7 @@ async def _generate_strategy_metadata[
 </format>
 """  # noqa: E501
 
-    response = await Stage.completion(
+    response: MultimodalContent = await Stage.completion(
         f"<failure_analysis>\n{failure_report}\n</failure_analysis>",
         instruction=strategy_prompt,
     ).execute()
@@ -501,18 +501,18 @@ async def _generate_strategy_metadata[
         "strategy",
         content=response,
     ):
-        name_elem: MultimodalTagElement | None = MultimodalTagElement.parse_first(
+        name_element: MultimodalTagElement | None = MultimodalTagElement.parse_first(
             "name",
             content=strategy_element.content,
         )
-        approach_elem: MultimodalTagElement | None = MultimodalTagElement.parse_first(
+        approach_element: MultimodalTagElement | None = MultimodalTagElement.parse_first(
             "approach",
             content=strategy_element.content,
         )
 
-        if name_elem and approach_elem:
-            strategy_name: str = name_elem.content.to_str().strip()
-            strategy_approach: str = approach_elem.content.to_str().strip()
+        if name_element and approach_element:
+            strategy_name: str = name_element.content.to_str().strip()
+            strategy_approach: str = approach_element.content.to_str().strip()
             strategies.append((strategy_name, strategy_approach))
 
     return strategies
@@ -530,7 +530,7 @@ async def _generate_instruction_content[
     guidelines: str | None,
 ) -> str:
     # Analyze failures
-    failure_report = evaluation_result.report(
+    failure_report: str = evaluation_result.report(
         include_passed=False,
         include_details=True,
     )
@@ -559,12 +559,17 @@ async def _generate_instruction_content[
 </format>
 """  # noqa: E501
 
-    response = await Stage.completion(
+    ctx.log_info(f"Generating updated instruction using {strategy_name}:\n\n{strategy_approach}")
+    response: MultimodalContent = await Stage.completion(
         f"<failure_analysis>\n{failure_report}\n</failure_analysis>",
         instruction=refinement_prompt,
     ).execute()
 
-    return response.to_str().strip()
+    updated_instruction: str = response.to_str().strip()
+
+    ctx.log_info(f"Prepared updated instruction using {strategy_name}:\n\n{updated_instruction}")
+
+    return updated_instruction
 
 
 async def _generate_refined_instructions[
@@ -577,7 +582,7 @@ async def _generate_refined_instructions[
     guidelines: str | None,
 ) -> Sequence[tuple[str, Instruction]]:
     # Step 1: Generate strategy metadata
-    strategy_metadata = await _generate_strategy_metadata(
+    strategy_metadata: Sequence[tuple[str, str]] = await _generate_strategy_metadata(
         evaluation_result=evaluation_result,
         parent_strategy=parent_strategy,
         guidelines=guidelines,
@@ -593,7 +598,7 @@ async def _generate_refined_instructions[
             evaluation_result=evaluation_result,
             guidelines=guidelines,
         )
-        refined_instruction = instruction.updated(content=refined_content)
+        refined_instruction: Instruction = instruction.updated(content=refined_content)
         refined_strategies.append((strategy_name, refined_instruction))
 
     # Validate and return results
@@ -691,7 +696,10 @@ async def tree_finalization(
         # Update result with best instruction and updated state
         return state.updated(
             refinement_state,
-            result=MultimodalContent.of(best_instruction.content),
+            result=MultimodalContent.of(
+                best_instruction.content,
+                meta={"score": best_score},
+            ),
         )
 
     return tree_finalization
@@ -738,8 +746,6 @@ def _log_tree_statistics(
         depth_distribution[node.depth] = depth_distribution.get(node.depth, 0) + 1
 
     max_depth: int = max(depth_distribution.keys()) if depth_distribution else 0
-    theoretical_max: int = 2 ** (max_depth + 1) - 1
-    efficiency: float = active_nodes / theoretical_max if theoretical_max > 0.0 else 0.0
 
     # Build path to best node
     best_path: list[str] = []
@@ -757,6 +763,5 @@ def _log_tree_statistics(
         f"- Active nodes: {active_nodes}\n"
         f"- Pruned nodes: {pruned_count} ({pruned_count / total_nodes * 100:.1f}%)\n"
         f"- Max depth reached: {max_depth}\n"
-        f"- Exploration efficiency: {efficiency:.1%}\n"
         f"- Best path: {' -> '.join(best_path) if best_path else 'None'}"
     )