finitearth
diff --git a/‎.coverage‎
0 Bytes b/‎.coverage‎
0 Bytes
diff --git a/‎.pre-commit-config.yaml‎
Lines changed: 1 addition & 1 deletion b/‎.pre-commit-config.yaml‎
Lines changed: 1 addition & 1 deletion
diff --git a/‎notebooks/getting_started.ipynb‎
Lines changed: 85 additions & 86 deletions b/‎notebooks/getting_started.ipynb‎
Lines changed: 85 additions & 86 deletions
diff --git a/‎promptolution/config.py‎
Lines changed: 11 additions & 2 deletions b/‎promptolution/config.py‎
Lines changed: 11 additions & 2 deletions
diff --git a/‎promptolution/helpers.py‎
Lines changed: 6 additions & 2 deletions b/‎promptolution/helpers.py‎
Lines changed: 6 additions & 2 deletions
diff --git a/‎promptolution/llms/vllm.py‎
Lines changed: 4 additions & 3 deletions b/‎promptolution/llms/vllm.py‎
Lines changed: 4 additions & 3 deletions
diff --git a/‎promptolution/optimizers/__init__.py‎
Lines changed: 26 additions & 2 deletions b/‎promptolution/optimizers/__init__.py‎
Lines changed: 26 additions & 2 deletions
diff --git a/‎promptolution/optimizers/base_optimizer.py‎
Lines changed: 1 addition & 2 deletions b/‎promptolution/optimizers/base_optimizer.py‎
Lines changed: 1 addition & 2 deletions
@@ -1,5 +1,5 @@
 fail_fast: true
-exclude: '^(?!promptolution/).*$|^promptolution/templates.py'
+exclude: '^(?!(promptolution/|tests/)).*$|^promptolution/templates.py'
 repos:
   - repo: https://github.com/gitleaks/gitleaks
     rev: v8.18.2
 
@@ -18,11 +18,20 @@ def __init__(self, **kwargs):
         for key, value in kwargs.items():
             setattr(self, key, value)
 
+    def __setattr__(self, name, value):
+        """Override attribute setting to track used attributes."""
+        # Set the attribute using the standard mechanism
+        object.__setattr__(self, name, value)
+        if not name.startswith("_") and not callable(value):
+            self._used_attributes.add(name)
+
     def __getattribute__(self, name):
         """Override attribute access to track used attributes."""
         # Get the attribute using the standard mechanism
-        value = object.__getattribute__(self, name)
-
+        try:
+            value = object.__getattribute__(self, name)
+        except AttributeError:
+            return None
         if not name.startswith("_") and not callable(value):
             self._used_attributes.add(name)
 
 
@@ -46,7 +46,11 @@ def run_optimization(df, config: ExperimentConfig):
     """
     llm = get_llm(config=config)
     predictor = get_predictor(llm, config=config)
+
     config.task_description = config.task_description + " " + predictor.extraction_description
+    if config.optimizer == "capo" and config.subsample_strategy is None:
+        logger.info("CAPO requires block evaluation strategy. Setting it to 'sequential_block'.")
+        config.subsample_strategy = "sequential_block"
 
     task = get_task(df, config)
     optimizer = get_optimizer(
@@ -55,7 +59,7 @@ def run_optimization(df, config: ExperimentConfig):
         task=task,
         config=config,
     )
-
+    logger.info("Starting optimization...")
     prompts = optimizer.optimize(n_steps=config.n_steps)
 
     if hasattr(config, "prepend_exemplars") and config.prepend_exemplars:
@@ -80,7 +84,7 @@ def run_evaluation(df: pd.DataFrame, config: ExperimentConfig, prompts: List[str
 
     llm = get_llm(config=config)
     predictor = get_predictor(llm, config=config)
-
+    logger.info("Starting evaluation...")
     scores = task.evaluate(prompts, predictor)
     df = pd.DataFrame(dict(prompt=prompts, score=scores))
     df = df.sort_values("score", ascending=False, ignore_index=True)
 
@@ -10,7 +10,6 @@
 logger = Logger(__name__)
 
 try:
-    import torch
     from transformers import AutoTokenizer
     from vllm import LLM, SamplingParams
 
@@ -75,7 +74,7 @@ def __init__(
         """
         if not imports_successful:
             raise ImportError(
-                "Could not import at least one of the required libraries: torch, transformers, vllm. "
+                "Could not import at least one of the required libraries: transformers, vllm. "
                 "Please ensure they are installed in your environment."
             )
 
@@ -110,7 +109,9 @@ def __init__(
 
         if batch_size is None:
             gpu_blocks = self.llm.llm_engine.model_executor.cache_config.num_gpu_blocks
-            block_size = self.llm.llm_engine.model_executor.cache_config.block_size
+            block_size = (
+                self.llm.llm_engine.model_executor.cache_config.block_size
+            )  # TODO rename, block_size is misleading
             self.batch_size = int((gpu_blocks * block_size / self.max_model_len) * 0.95)
             logger.info(f"Batch size set to {self.batch_size} based on GPU memory.")
         else:
 
@@ -7,6 +7,8 @@
 from promptolution.predictors.base_predictor import BasePredictor
 from promptolution.tasks.base_task import BaseTask
 from promptolution.templates import (
+    CAPO_CROSSOVER_TEMPLATE,
+    CAPO_MUTATION_TEMPLATE,
     EVOPROMPT_DE_TEMPLATE,
     EVOPROMPT_DE_TEMPLATE_TD,
     EVOPROMPT_GA_TEMPLATE,
@@ -16,6 +18,7 @@
 )
 
 from .base_optimizer import DummyOptimizer
+from .capo import CAPO
 from .evoprompt_de import EvoPromptDE
 from .evoprompt_ga import EvoPromptGA
 from .opro import Opro
@@ -54,8 +57,26 @@ def get_optimizer(
     if meta_prompt is None and hasattr(config, "meta_prompt"):
         meta_prompt = config.meta_prompt
 
-    if optimizer == "dummy":
-        return DummyOptimizer(predictor=predictor, config=config)
+    if config.optimizer == "capo":
+        crossover_template = (
+            CAPO_CROSSOVER_TEMPLATE.replace("<task_desc>", task_description)
+            if task_description
+            else CAPO_CROSSOVER_TEMPLATE
+        )
+        mutation_template = (
+            CAPO_MUTATION_TEMPLATE.replace("<task_desc>", task_description)
+            if task_description
+            else CAPO_MUTATION_TEMPLATE
+        )
+
+        return CAPO(
+            predictor=predictor,
+            meta_llm=meta_llm,
+            task=task,
+            crossover_template=crossover_template,
+            mutation_template=mutation_template,
+            config=config,
+        )
 
     if config.optimizer == "evopromptde":
         template = (
@@ -77,4 +98,7 @@ def get_optimizer(
         template = OPRO_TEMPLATE_TD.replace("<task_desc>", task_description) if task_description else OPRO_TEMPLATE
         return Opro(predictor=predictor, meta_llm=meta_llm, task=task, prompt_template=template, config=config)
 
+    if optimizer == "dummy":
+        return DummyOptimizer(predictor=predictor, config=config)
+
     raise ValueError(f"Unknown optimizer: {config.optimizer}")
@@ -45,7 +45,6 @@ def __init__(
         self.task = task
         self.callbacks = callbacks or []
         self.predictor = predictor
-        self.verbosity = 0
 
         if config is not None:
             config.apply_to(self)
@@ -97,7 +96,7 @@ def _pre_optimization_loop(self):
         pass
 
     @abstractmethod
-    def _step(self):
+    def _step(self) -> List[str]:
         """Perform a single optimization step.
 
         This method should be implemented by concrete optimizer classes to define