camel-ai 0.2.59__py3-none-any.whl → 0.2.82__py3-none-any.whl

camel/datagen/cot_datagen.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import json
 from datetime import datetime
@@ -39,7 +39,7 @@ class AgentResponse(BaseModel):
 
     score: Annotated[float, confloat(ge=0, le=1)] = Field(
         ...,
-        description="""Similarity score between 0 and 1 
+        description="""Similarity score between 0 and 1
         comparing current answer to correct answer""",
     )
 
@@ -204,26 +204,23 @@ class CoTDataGenerator:
         logger.info("Answer verification result: %s", is_correct)
         return is_correct
 
-    def monte_carlo_tree_search(
+    def evaluate_partial_solution(
         self, question: str, partial_solution: str = ""
     ) -> float:
-        r"""Perform Monte Carlo Tree Search to find the best solution.
+        r"""Evaluate the quality of a partial solution against the
+        golden answer.
 
-        Process:
-        a. Selection: Choose promising partial solutions based on previous
-        scores
-        b. Expansion: Generate new solution steps using the generator agent
-        c. Simulation: Evaluate solution quality using similarity scores
-        d. Backpropagation: Update solution tree with new findings
+        This function generates a similarity score between the given partial
+        solution and the correct answer (golden answer).
 
         Args:
-            question (str): The question to solve.
-            partial_solution (str): The current partial solution.
+            question (str): The question being solved.
+            partial_solution (str): The partial solution generated so far.
                 (default::obj:`""`)
 
         Returns:
-            float: The similarity score between the current
-                solution and golden answer.
+            float: A similarity score between 0 and 1, indicating how close the
+                partial solution is to the golden answer.
         """
         if question not in self.golden_answers:
             raise ValueError(
@@ -293,10 +290,21 @@ class CoTDataGenerator:
         r"""Solve a question using a multi-step approach.
 
         The solution process follows these steps:
-        1. Try to solve directly - if correct, return the solution
-        2. If not correct, use Monte Carlo Tree Search to find a good solution
-        3. If the solution isn't perfect, use binary search to locate errors
-        4. Generate a new solution based on the correct part
+        1. Try to solve directly - if correct, return the solution.
+        2. If not correct, perform a search by iteratively generating
+        new solutions and evaluating their similarity scores to
+        find a good solution. The search process involves:
+            a. Generation: Generate new solution candidates using
+            the generator agent.
+            b. Evaluation: Score each solution candidate for similarity
+            to the golden answer.
+            c. Selection: Keep the best-scoring candidate found so far.
+            d. Early stopping: If a sufficiently high-scoring solution
+            is found (score > 0.9), stop early.
+        3. If the solution isn't perfect, use binary search to locate
+        errors.
+        4. Generate a new solution based on the correct part of the
+        initial solution.
 
         Args:
             question (str): The question to solve.
@@ -304,14 +312,14 @@ class CoTDataGenerator:
         Returns:
             str: The best solution found.
         """
+
         # 1. Try direct solution first
         solution = self.get_answer(question)
         if self.verify_answer(question, solution):
             logger.info("Initial solution is correct")
             return solution
 
-        # 2. If direct solution fails, try Monte Carlo Tree Search
-        # to find a solution with high similarity score
+        # 2. If direct solution fails, iteratively search for a better solution
         best_solution = ""
         best_score: float = 0.0
         for i in range(self.search_limit):
@@ -319,23 +327,10 @@ class CoTDataGenerator:
             current_solution = self.get_answer(question, best_solution)
 
             # Evaluate solution similarity score
-            prompt = (
-                f"Please evaluate this solution and "
-                f"give a score between 0-1:\n"
-                f"Question: {question}\n"
-                f"Solution: {current_solution}\n"
-                f"Correct answer: {self.golden_answers.get(question, '')}\n"
-                f"Return a JSON object with a single field 'score' containing "
-                f"a float between 0 and 1, like this: {{'score': 0.85}}\n"
-            )
-            self.generator_agent.reset()
-            response = self.generator_agent.step(prompt)
             try:
-                response = self.generator_agent.step(
-                    prompt, response_format=AgentResponse
+                score = self.evaluate_partial_solution(
+                    question, current_solution
                 )
-                agent_response = response.msgs[0].parsed.score  # type: ignore [union-attr]
-                score = agent_response
 
                 # Exit early if we find a very good solution (score > 0.9)
                 if score > 0.9:
@@ -357,7 +352,7 @@ class CoTDataGenerator:
                     best_score,
                 )
             except Exception as e:
-                logger.error("Error parsing agent response: %s", str(e))
+                logger.error("Error evaluating partial solution: %s", str(e))
                 continue
 
         # 3. If the answer is not completely correct,

camel/datagen/evol_instruct/__init__.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 from .evol_instruct import EvolInstructPipeline
 

camel/datagen/evol_instruct/evol_instruct.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import random
 import time

camel/datagen/evol_instruct/scorer.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import json
 from abc import ABC, abstractmethod
@@ -38,24 +38,27 @@ class BaseScorer(ABC):
 
 class MathScorer(BaseScorer):
     def __init__(self, agent: Optional[ChatAgent] = None):
-        self.system_msg = (
-            "You are an evaluator for math problems. Your task is to compare "
-            "a new math problem against a reference math problem, and rate it "
-            "in **four dimensions**, each scored from 1 to 5.\n\n"
-            "1. Diversity (1-5): How novel is the new problem compared to the "
-            "reference? 1 = very similar, 5 = completely different.\n"
-            "2. Difficulty (1-5): Rate the relative difficulty compared to the"
-            " reference problem. 1 = much less difficult, "
-            "3 = similar difficulty, 5 = much more difficult.\n"
-            "3. Validity (1-5): How well-defined and sound is the problem?"
-            "1 = very vague or flawed, 5 = very clear and rigorous.\n"
-            "4. Solvability (1-5): How likely is the problem solvable using "
-            "standard math techniques? 1 = very unsolvable or ambiguous, "
-            "5 = very clearly solvable.\n\n"
-            "Respond with a JSON object like: "
-            "{ \"diversity\": ..., \"difficulty\": ..., "
-            "\"validity\": ..., \"solvability\": ... }"
-        )
+        self.system_msg = """
+You are an evaluator for math problems. Your task is to compare a new math
+problem against a reference math problem by trying to solve it, and rate it
+in **three dimensions**.
+
+1. Diversity (1-5): How novel is the new problem compared to the
+reference? 1 = almost the same, 5 = completely different.
+
+2. Difficulty (1-10): Rate the relative difficulty compared to the reference
+problem. 1 = much less difficult, 5 = similar difficulty, 10 = much more
+difficult. The difficulty should be based on the complexity of reasoning—i.e.,
+problems that require multi-step reasoning or clever methods to solve.
+
+3. Solvability (1-10): How likely is the problem solvable using standard math
+techniques and only contain one question that could be answered by a number or
+a formula? 1 = very unsolvable or ambiguous, 10 = solvable and could be
+answered by a number or a formula.
+
+Respond with a JSON object like:
+{ "solution": ..., "diversity": ..., "difficulty": ..., "solvability": ... }
+"""
         self.agent = agent or ChatAgent(self.system_msg)
 
     class MathScoreSchema(BaseModel):
@@ -69,10 +72,6 @@ class MathScorer(BaseScorer):
         difficulty: int = Field(
             ..., description="Score for the relative difficulty"
         )
-        validity: int = Field(
-            ...,
-            description="Score for how well-defined and sound the problem is",
-        )
         solvability: int = Field(
             ...,
             description="Score for the solvability of the problem",
@@ -95,7 +94,7 @@ class MathScorer(BaseScorer):
         query = (
             f"Reference problem:\n{reference_problem}\n\n"
             f"New problem:\n{new_problem}\n\n"
-            "Provide scores in JSON format."
+            "Try to solve the new problem. Then provide scores in JSON format."
         )
         response = self.agent.step(query, response_format=self.MathScoreSchema)
         score_data = json.loads(response.msg.content)

camel/datagen/evol_instruct/templates.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
@@ -208,54 +208,54 @@ class MathEvolInstructTemplates(BaseEvolInstructTemplates):
     r"""Contains templates for MathEvolInstruct prompt transformations."""
 
     # Meta-instructions for in-depth evolving
-    INST_IN_DEPTH = (
-        "Please act as a math expert. Your objective is to create a new math "
-        "problem that is more challenging yet concise than the given math "
-        "problem. Ensure that the mathematical content (including any "
-        "equations or figures) is preserved, and rephrase the problem to "
-        "increase its complexity and depth. The generated problem should be "
-        "clearly stated, strictly mathematical, and suitable for solving with "
-        "symbolic computation (e.g., using sympy). You will be given a method "
-        "to guide your creation. Make sure to follow the method strictly. "
-        "Consolidate any multiple parts into one integrated question that "
-        "ask for one definitive answer. Respond with your generated problem "
-        "directly. "
-        "#Original Problem#:\n{prompt}\n"
-        "#Generated Problem#:\n"
-    ).lstrip()
+    INST_IN_DEPTH = """
+Please act as a math expert. Your objective is to create a new math problem
+that is more challenging yet concise than the given math problem. Modify the
+problem to increase its complexity and depth. The generated problem should be
+clearly stated, strictly mathematical, and suitable for solving with symbolic
+computation (e.g., using sympy). You will be given a method to guide your
+creation. Make sure to follow the method strictly. Consolidate any multiple
+parts into one integrated question that ask for one definitive answer. Do not
+include multiple-choice, true/false, or proof-based questions. The final
+answer should be a number or a formula. Respond with your generated problem
+directly. The difficulty should be based on the complexity of reasoning—i.e.,
+problems that require multi-step reasoning or clever methods to solve. The
+challenge of a problem should not stem purely from computational complexity;
+while complex calculations may be involved, a problem should not be considered
+difficult solely because lengthy computations increase solving time.
+#Original Problem#:
+{prompt}
+#Generated Problem#:
+"""
 
     EVOL_METHODS = {
-        "constraints": (
-            "Add one or more significant constraints or requirements into the "
-            "'#Given Prompt#'. The added constraints must meaningfully alter "
-            "how the model would respond. For example, specify additional "
-            "rules, contexts, or limitations that demand creative adjustments."
-        ),
-        "deepening": (
-            "Increase the difficulty of the #Given Prompt# by integrating "
-            "additional layers of reasoning and rigor. Refine the problem so "
-            "that all added difficulty is consolidated into a single coherent "
-            "question requiring one final answer, avoiding fragmentation into "
-            "multiple sub-problems."
-        ),
-        "expansion": (
-            "Expand the #Given Prompt# by incorporating additional "
-            "perspectives or layers of complexity into the problem statement. "
-            "Ensure that the revised problem remains a single, unified "
-            "question with one final answer, rather than a series of separate "
-            "sub-questions."
-        ),
-        "condense": (
-            "Reformulate the given math problem into a well-structured and "
-            "formally stated mathematical question.\n"
-            "- Present the problem in a structured and rigorous mathematical "
-            "format.\n"
-            "- Removing unnecessary instructions, explanations, or hints.\n"
-            "- If the given problem contains several sub-questions, make "
-            "necessary changes to let the problem could be answered with one "
-            "number or expression by removing the sub-questions or combining "
-            "them into one."
-        ),
+        "constraints": """
+Add one or more significant constraints or requirements into the
+'#Given Prompt#'. The added constraints must meaningfully alter how the model
+would respond. For example, specify additional rules, contexts, or limitations
+that demand creative adjustments. This method should make the problem more
+challenging in the reasoning and the solution of it should be clever and
+elegant.
+""",
+        "deepening": """
+Increase the difficulty of the #Given Prompt# by integrating additional layers
+of reasoning and rigor. Refine the problem so that all added difficulty is
+consolidated into a single coherent question requiring one final answer,
+avoiding fragmentation into multiple sub-problems.
+""",
+        "expansion": """
+Expand the #Given Prompt# by incorporating additional perspectives or layers
+of complexity into the problem statement. Ensure that the revised problem
+remains a single, unified question with one final answer, rather than a
+series of separate sub-questions.
+""",
+        "condense": """
+Reformulate the given math problem into a well-structured and formally stated
+mathematical question. Remove unnecessary instructions, explanations, or hints.
+If the given problem contains several sub-questions, make necessary changes
+to let the problem could be answered with one number or one expression by
+removing the sub-questions or combining them into one.
+""",
     }
 
     IN_DEPTH_KEYS = ['constraints', 'deepening', 'expansion']

camel/datagen/self_improving_cot.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import asyncio
 import json
@@ -116,7 +116,7 @@ class SelfImprovingCoTPipeline:
                 samples to be drawn using the rejection sampling
                 method, where samples are accepted or rejected based on
                 a predefined condition to achieve a desired distribution.
-                (default: :obj: `None`)
+                (default: :obj:`None`)
             evaluate_agent (Optional[ChatAgent]): The chat agent used for
                 evaluating reasoning traces. (default: :obj:`None`)
             reward_model (BaseRewardModel, optional): Model used to evaluate
@@ -861,7 +861,7 @@ Problem: {problem}
 
 Please show your complete reasoning process."""
 
-    EVALUATION_TEMPLATE = """Please evaluate this reasoning trace and 
+    EVALUATION_TEMPLATE = """Please evaluate this reasoning trace and
 provide scores and feedback in valid JSON format.
 
 Problem: {problem}
@@ -884,7 +884,7 @@ Respond ONLY with a JSON object in this exact format:
     "feedback": "<specific feedback for improvement>"
 }}"""
 
-    IMPROVEMENT_TEMPLATE = """Based on this feedback, generate an 
+    IMPROVEMENT_TEMPLATE = """Based on this feedback, generate an
 improved reasoning trace:
 Problem: {problem}
 

camel/datagen/self_instruct/__init__.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 from .filter import (
     FILTER_REGISTRY,
     FilterFunction,

camel/datagen/self_instruct/filter/__init__.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 from .filter_function import (
     FilterFunction,
     KeywordFilter,

camel/datagen/self_instruct/filter/filter_function.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import re
 from abc import ABC, abstractmethod

camel/datagen/self_instruct/filter/filter_registry.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 from typing import Any, Callable, Dict
 
 from .filter_function import (

camel/datagen/self_instruct/filter/instruction_filter.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 from typing import Any, Dict, List, Tuple, Union
 
 from camel.logger import get_logger

camel/datagen/self_instruct/self_instruct.py

@@ -1,4 +1,4 @@
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
 # You may obtain a copy of the License at
@@ -10,7 +10,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-# ========= Copyright 2023-2024 @ CAMEL-AI.org. All Rights Reserved. =========
+# ========= Copyright 2023-2025 @ CAMEL-AI.org. All Rights Reserved. =========
 
 import json
 import os