feat: add RD-Agent-Quant scenario (#838)

* fix model input shape bug and costeer_model bug * fix a bug * fix a bug in docker result extraction * a system-level optimization * add a filter of stdout * update * add stdout to model * model training_hyperparameters update * quant scenario * update some quant settings * llm choose action * Thompson Sampling Bandit for action choosing * refine both scens * add trace messages for quant scen * fix some bugs * fix some bugs * update * update * update * fix * fix * fix * update for merge * fix ci * fix some bugs * fix ci * fix ci * fix ci * fix ci * refactor * default qlib4rdagent local env downloading * fix ci * fix ci * fix a bug * fix ci * fix: align all prompts on template (#908) * use template to render all prompts * fix CI --------- Co-authored-by: Xu Yang <xuyang1@microsoft.com> * add fin_quant in cli * fix a bug * fix ci * fix some bugs * refactor * remove the columns in hypothesis if no value generated in this column * fix a bug * fix ci * fix conda env * add qlib gitignore * remove existed qlib folder & install torch in qlib conda * fix workspace ui in feedback * align model config in coder and runner in docker or conda * fix CI * fix CI --------- Co-authored-by: Xu Yang <peteryang@vip.qq.com> Co-authored-by: Xu Yang <xuyang1@microsoft.com>
2026-08-07 12:07:43 +00:00 · 2025-05-29 16:16:51 +08:00
parent b0e88c7375
commit d1019cb568
64 changed files with 2427 additions and 1025 deletions
@@ -1,7 +1,6 @@
 from __future__ import annotations

 from abc import abstractmethod
-from pathlib import Path

 from rdagent.components.coder.CoSTEER.config import CoSTEERSettings
 from rdagent.components.coder.CoSTEER.evaluators import (
@@ -15,12 +14,9 @@ from rdagent.components.coder.CoSTEER.knowledge_management import (
 from rdagent.core.conf import RD_AGENT_SETTINGS
 from rdagent.core.evolving_framework import EvolvingStrategy, EvoStep, QueriedKnowledge
 from rdagent.core.experiment import FBWorkspace, Task
-from rdagent.core.prompts import Prompts
 from rdagent.core.scenario import Scenario
 from rdagent.core.utils import multiprocessing_wrapper

-implement_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
-

 class MultiProcessEvolvingStrategy(EvolvingStrategy):
    def __init__(self, scen: Scenario, settings: CoSTEERSettings):
@@ -8,8 +8,6 @@ from itertools import combinations
 from pathlib import Path
 from typing import List, Union

-from jinja2 import Environment, StrictUndefined
-
 from rdagent.components.coder.CoSTEER.config import CoSTEERSettings
 from rdagent.components.coder.CoSTEER.evaluators import CoSTEERSingleFeedback
 from rdagent.components.knowledge_management.graph import (
@@ -26,12 +24,12 @@ from rdagent.core.evolving_framework import (
    RAGStrategy,
 )
 from rdagent.core.experiment import FBWorkspace, Task
-from rdagent.core.prompts import Prompts
 from rdagent.log import rdagent_logger as logger
 from rdagent.oai.llm_utils import (
    APIBackend,
    calculate_embedding_distance_between_str_list,
 )
+from rdagent.utils.agent.tpl import T


 class CoSTEERKnowledge(Knowledge):
@@ -216,8 +214,6 @@ class CoSTEERQueriedKnowledgeV2(CoSTEERQueriedKnowledgeV1):


 class CoSTEERRAGStrategyV2(RAGStrategy):
-    prompt = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
-
    def __init__(self, knowledgebase: CoSTEERKnowledgeBaseV2, settings: CoSTEERSettings) -> None:
        super().__init__(knowledgebase)
        self.current_generated_trace_count = 0
@@ -324,12 +320,8 @@ class CoSTEERRAGStrategyV2(RAGStrategy):
        all_component_content = ""
        for _, component_node in enumerate(all_component_nodes):
            all_component_content += f"{component_node.content}, \n"
-        analyze_component_system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(self.prompt["analyze_component_prompt_v1_system"])
-            .render(
-                all_component_content=all_component_content,
-            )
+        analyze_component_system_prompt = T(".prompts:analyze_component_prompt_v1_system").r(
+            all_component_content=all_component_content,
        )

        analyze_component_user_prompt = target_task_information
@@ -11,9 +11,7 @@ File structure
    - Each coder could be tested.
 """

-import json
 from pathlib import Path
-from typing import Dict

 from jinja2 import Environment, StrictUndefined

@@ -15,7 +15,7 @@ class FactorCoSTEERSettings(CoSTEERSettings):
    simple_background: bool = False
    """Whether to use simple background information for code feedback"""

-    file_based_execution_timeout: int = 120
+    file_based_execution_timeout: int = 3600
    """Timeout in seconds for each factor implementation execution"""

    select_method: str = "random"
@@ -1,20 +1,16 @@
 import io
 import json
 from abc import abstractmethod
-from pathlib import Path
 from typing import Dict, Tuple

 import pandas as pd
-from jinja2 import Environment, StrictUndefined

 from rdagent.components.coder.factor_coder.config import FACTOR_COSTEER_SETTINGS
 from rdagent.components.coder.factor_coder.factor import FactorTask
 from rdagent.core.experiment import Task, Workspace
-from rdagent.core.prompts import Prompts
 from rdagent.oai.llm_conf import LLM_SETTINGS
 from rdagent.oai.llm_utils import APIBackend
-
-evaluate_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+from rdagent.utils.agent.tpl import T


 class FactorEvaluator:
@@ -81,36 +77,26 @@ class FactorCodeEvaluator(FactorEvaluator):
        factor_information = target_task.get_task_information()
        code = implementation.all_codes

-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(evaluate_prompts["evaluator_code_feedback_v1_system"])
-            .render(
-                scenario=(
-                    self.scen.get_scenario_all_desc(
-                        target_task,
-                        filtered_tag="feature",
-                        simple_background=FACTOR_COSTEER_SETTINGS.simple_background,
-                    )
-                    if self.scen is not None
-                    else "No scenario description."
+        system_prompt = T(".prompts:evaluator_code_feedback_v1_system").r(
+            scenario=(
+                self.scen.get_scenario_all_desc(
+                    target_task,
+                    filtered_tag="feature",
+                    simple_background=FACTOR_COSTEER_SETTINGS.simple_background,
                )
+                if self.scen is not None
+                else "No scenario description."
            )
        )

        execution_feedback_to_render = execution_feedback
        for _ in range(10):  # 10 times to split the content is enough
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    evaluate_prompts["evaluator_code_feedback_v1_user"],
-                )
-                .render(
-                    factor_information=factor_information,
-                    code=code,
-                    execution_feedback=execution_feedback_to_render,
-                    value_feedback=value_feedback,
-                    gt_code=gt_implementation.code if gt_implementation else None,
-                )
+            user_prompt = T(".prompts:evaluator_code_feedback_v1_user").r(
+                factor_information=factor_information,
+                code=code,
+                execution_feedback=execution_feedback_to_render,
+                value_feedback=value_feedback,
+                gt_code=gt_implementation.code if gt_implementation else None,
            )
            if (
                APIBackend().build_messages_and_calculate_token(
@@ -189,17 +175,11 @@ class FactorOutputFormatEvaluator(FactorEvaluator):
        buffer = io.StringIO()
        gen_df.info(buf=buffer)
        gen_df_info_str = f"The user is currently working on a feature related task.\nThe output dataframe info is:\n{buffer.getvalue()}"
-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(
-                evaluate_prompts["evaluator_output_format_system"],
-            )
-            .render(
-                scenario=(
-                    self.scen.get_scenario_all_desc(implementation.target_task, filtered_tag="feature")
-                    if self.scen is not None
-                    else "No scenario description."
-                )
+        system_prompt = T(".prompts:evaluator_output_format_system").r(
+            scenario=(
+                self.scen.get_scenario_all_desc(implementation.target_task, filtered_tag="feature")
+                if self.scen is not None
+                else "No scenario description."
            )
        )

@@ -504,35 +484,25 @@ class FactorFinalDecisionEvaluator(FactorEvaluator):
        code_feedback: str,
        **kwargs,
    ) -> Tuple:
-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(evaluate_prompts["evaluator_final_decision_v1_system"])
-            .render(
-                scenario=(
-                    self.scen.get_scenario_all_desc(target_task, filtered_tag="feature")
-                    if self.scen is not None
-                    else "No scenario description."
-                )
+        system_prompt = T(".prompts:evaluator_final_decision_v1_system").r(
+            scenario=(
+                self.scen.get_scenario_all_desc(target_task, filtered_tag="feature")
+                if self.scen is not None
+                else "No scenario description."
            )
        )
        execution_feedback_to_render = execution_feedback

        for _ in range(10):  # 10 times to split the content is enough
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    evaluate_prompts["evaluator_final_decision_v1_user"],
-                )
-                .render(
-                    factor_information=target_task.get_task_information(),
-                    execution_feedback=execution_feedback_to_render,
-                    code_feedback=code_feedback,
-                    value_feedback=(
-                        value_feedback
-                        if value_feedback is not None
-                        else "No Ground Truth Value provided, so no evaluation on value is performed."
-                    ),
-                )
+            user_prompt = T(".prompts:evaluator_final_decision_v1_user").r(
+                factor_information=target_task.get_task_information(),
+                execution_feedback=execution_feedback_to_render,
+                code_feedback=code_feedback,
+                value_feedback=(
+                    value_feedback
+                    if value_feedback is not None
+                    else "No Ground Truth Value provided, so no evaluation on value is performed."
+                ),
            )
            if (
                APIBackend().build_messages_and_calculate_token(
@@ -1,11 +1,8 @@
 from __future__ import annotations

 import json
-from pathlib import Path
 from typing import Dict

-from jinja2 import Environment, StrictUndefined
-
 from rdagent.components.coder.CoSTEER.evaluators import CoSTEERSingleFeedback
 from rdagent.components.coder.CoSTEER.evolving_strategy import (
    MultiProcessEvolvingStrategy,
@@ -17,11 +14,9 @@ from rdagent.components.coder.CoSTEER.knowledge_management import (
 from rdagent.components.coder.factor_coder.config import FACTOR_COSTEER_SETTINGS
 from rdagent.components.coder.factor_coder.factor import FactorFBWorkspace, FactorTask
 from rdagent.core.experiment import FBWorkspace
-from rdagent.core.prompts import Prompts
 from rdagent.oai.llm_conf import LLM_SETTINGS
 from rdagent.oai.llm_utils import APIBackend
-
-implement_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+from rdagent.utils.agent.tpl import T


 class FactorMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
@@ -36,24 +31,14 @@ class FactorMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
        queried_former_failed_knowledge_to_render: list,
        queried_similar_error_knowledge_to_render: list,
    ) -> str:
-        error_summary_system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(implement_prompts["evolving_strategy_error_summary_v2_system"])
-            .render(
-                scenario=self.scen.get_scenario_all_desc(target_task),
-                factor_information_str=target_task.get_task_information(),
-                code_and_feedback=queried_former_failed_knowledge_to_render[-1].get_implementation_and_feedback_str(),
-            )
-            .strip("\n")
+        error_summary_system_prompt = T(".prompts:evolving_strategy_error_summary_v2_system").r(
+            scenario=self.scen.get_scenario_all_desc(target_task),
+            factor_information_str=target_task.get_task_information(),
+            code_and_feedback=queried_former_failed_knowledge_to_render[-1].get_implementation_and_feedback_str(),
        )
        for _ in range(10):  # max attempt to reduce the length of error_summary_user_prompt
-            error_summary_user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(implement_prompts["evolving_strategy_error_summary_v2_user"])
-                .render(
-                    queried_similar_error_knowledge=queried_similar_error_knowledge_to_render,
-                )
-                .strip("\n")
+            error_summary_user_prompt = T(".prompts:evolving_strategy_error_summary_v2_user").r(
+                queried_similar_error_knowledge=queried_similar_error_knowledge_to_render,
            )
            if (
                APIBackend().build_messages_and_calculate_token(
@@ -106,16 +91,9 @@ class FactorMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
        latest_attempt_to_latest_successful_execution = queried_knowledge.task_to_former_failed_traces[
            target_factor_task_information
        ][1]
-
-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(
-                implement_prompts["evolving_strategy_factor_implementation_v1_system"],
-            )
-            .render(
-                scenario=self.scen.get_scenario_all_desc(target_task, filtered_tag="feature"),
-                queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
-            )
+        system_prompt = T(".prompts:evolving_strategy_factor_implementation_v1_system").r(
+            scenario=self.scen.get_scenario_all_desc(target_task, filtered_tag="feature"),
+            queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
        )
        queried_similar_successful_knowledge_to_render = queried_similar_successful_knowledge
        queried_similar_error_knowledge_to_render = queried_similar_error_knowledge
@@ -136,19 +114,12 @@ class FactorMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
            else:
                error_summary_critics = None
            # 构建user_prompt。开始写代码
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    implement_prompts["evolving_strategy_factor_implementation_v2_user"],
-                )
-                .render(
-                    factor_information_str=target_factor_task_information,
-                    queried_similar_successful_knowledge=queried_similar_successful_knowledge_to_render,
-                    queried_similar_error_knowledge=queried_similar_error_knowledge_to_render,
-                    error_summary_critics=error_summary_critics,
-                    latest_attempt_to_latest_successful_execution=latest_attempt_to_latest_successful_execution,
-                )
-                .strip("\n")
+            user_prompt = T(".prompts:evolving_strategy_factor_implementation_v2_user").r(
+                factor_information_str=target_factor_task_information,
+                queried_similar_successful_knowledge=queried_similar_successful_knowledge_to_render,
+                queried_similar_error_knowledge=queried_similar_error_knowledge_to_render,
+                error_summary_critics=error_summary_critics,
+                latest_attempt_to_latest_successful_execution=latest_attempt_to_latest_successful_execution,
            )
            if (
                APIBackend().build_messages_and_calculate_token(user_prompt=user_prompt, system_prompt=system_prompt)
@@ -49,6 +49,12 @@ class FactorTask(CoSTEERTask):
        return f"""factor_name: {self.factor_name}
 factor_description: {self.factor_description}
 factor_formulation: {self.factor_formulation}
+variables: {str(self.variables)}"""
+
+    def get_task_brief_information(self):
+        return f"""factor_name: {self.factor_name}
+factor_description: {self.factor_description}
+factor_formulation: {self.factor_formulation}
 variables: {str(self.variables)}"""

    def get_task_information_and_implementation_result(self):
@@ -116,6 +116,9 @@ evolving_strategy_error_summary_v2_system: |-

  You suggestion should not include any code, just some clear and short suggestions. Please point out very critical issues in your response, ignore non-important issues to avoid confusion. If no big issue found in the code, you can response "No critics found".

+  [NOTE]
+  1. When processing data, avoid time leakage.
+
  Please response the critic in the following format. Here is an example structure for the output:
  critic 1: The critic message to critic 1
  critic 2: The critic message to critic 2
@@ -0,0 +1,13 @@
+from pydantic_settings import SettingsConfigDict
+
+from rdagent.components.coder.CoSTEER.config import CoSTEERSettings
+
+
+class ModelCoSTEERSettings(CoSTEERSettings):
+    model_config = SettingsConfigDict(env_prefix="MODEL_CoSTEER_")
+
+    env_type: str = "conda"  # or "docker"
+    """Environment to run model code in coder and runner: 'conda' for local conda env, 'docker' for Docker container"""
+
+
+MODEL_COSTEER_SETTINGS = ModelCoSTEERSettings()
@@ -1,18 +1,14 @@
 import json
-from pathlib import Path
 from typing import Dict, Tuple

 import numpy as np
-from jinja2 import Environment, StrictUndefined

 from rdagent.components.coder.CoSTEER.evaluators import CoSTEEREvaluator
 from rdagent.components.coder.model_coder.model import ModelFBWorkspace, ModelTask
 from rdagent.core.experiment import Task, Workspace
-from rdagent.core.prompts import Prompts
 from rdagent.oai.llm_conf import LLM_SETTINGS
 from rdagent.oai.llm_utils import APIBackend
-
-evaluate_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+from rdagent.utils.agent.tpl import T


 # This shape evaluator is also used in data_science
@@ -70,32 +66,21 @@ class ModelCodeEvaluator(CoSTEEREvaluator):
        model_task_information = target_task.get_task_information()
        code = implementation.all_codes

-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(evaluate_prompts["evaluator_code_feedback"]["system"])
-            .render(
-                scenario=(
-                    self.scen.get_scenario_all_desc(target_task, filtered_tag=target_task.model_type)
-                    if self.scen is not None
-                    else "No scenario description."
-                )
+        system_prompt = T(".prompts:evaluator_code_feedback.system").r(
+            scenario=(
+                self.scen.get_scenario_all_desc(target_task, filtered_tag=target_task.model_type)
+                if self.scen is not None
+                else "No scenario description."
            )
        )
-
        execution_feedback_to_render = model_execution_feedback
        for _ in range(10):  # 10 times to split the content is enough
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    evaluate_prompts["evaluator_code_feedback"]["user"],
-                )
-                .render(
-                    model_information=model_task_information,
-                    code=code,
-                    model_execution_feedback=execution_feedback_to_render,
-                    model_value_feedback=model_value_feedback,
-                    gt_code=gt_implementation.all_codes if gt_implementation else None,
-                )
+            user_prompt = T(".prompts:evaluator_code_feedback.user").r(
+                model_information=model_task_information,
+                code=code,
+                model_execution_feedback=execution_feedback_to_render,
+                model_value_feedback=model_value_feedback,
+                gt_code=gt_implementation.all_codes if gt_implementation else None,
            )
            if (
                APIBackend().build_messages_and_calculate_token(
@@ -133,34 +118,25 @@ class ModelFinalEvaluator(CoSTEEREvaluator):
        if gt_implementation is not None:
            assert isinstance(gt_implementation, ModelFBWorkspace)

-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(evaluate_prompts["evaluator_final_feedback"]["system"])
-            .render(
-                scenario=(
-                    self.scen.get_scenario_all_desc(target_task, filtered_tag=target_task.model_type)
-                    if self.scen is not None
-                    else "No scenario description."
-                )
+        system_prompt = T(".prompts:evaluator_final_feedback.system").r(
+            scenario=(
+                self.scen.get_scenario_all_desc(target_task, filtered_tag=target_task.model_type)
+                if self.scen is not None
+                else "No scenario description."
            )
        )

        execution_feedback_to_render = model_execution_feedback

        for _ in range(10):  # 10 times to split the content is enough
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    evaluate_prompts["evaluator_final_feedback"]["user"],
-                )
-                .render(
-                    model_information=target_task.get_task_information(),
-                    model_execution_feedback=execution_feedback_to_render,
-                    model_shape_feedback=model_shape_feedback,
-                    model_code_feedback=model_code_feedback,
-                    model_value_feedback=model_value_feedback,
-                )
+            user_prompt = T(".prompts:evaluator_final_feedback.user").r(
+                model_information=target_task.get_task_information(),
+                model_execution_feedback=execution_feedback_to_render,
+                model_shape_feedback=model_shape_feedback,
+                model_code_feedback=model_code_feedback,
+                model_value_feedback=model_value_feedback,
            )
+
            if (
                APIBackend().build_messages_and_calculate_token(
                    user_prompt=user_prompt,
@@ -1,9 +1,6 @@
 import json
-from pathlib import Path
 from typing import Dict

-from jinja2 import Environment, StrictUndefined
-
 from rdagent.components.coder.CoSTEER.config import CoSTEER_SETTINGS
 from rdagent.components.coder.CoSTEER.evaluators import CoSTEERSingleFeedback
 from rdagent.components.coder.CoSTEER.evolving_strategy import (
@@ -14,16 +11,13 @@ from rdagent.components.coder.CoSTEER.knowledge_management import (
    CoSTEERQueriedKnowledgeV2,
 )
 from rdagent.components.coder.model_coder.model import (
-    ModelExperiment,
    ModelFBWorkspace,
    ModelTask,
 )
 from rdagent.core.experiment import FBWorkspace
-from rdagent.core.prompts import Prompts
 from rdagent.oai.llm_conf import LLM_SETTINGS
 from rdagent.oai.llm_utils import APIBackend
-
-coder_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+from rdagent.utils.agent.tpl import T


 class ModelMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
@@ -52,31 +46,18 @@ class ModelMultiProcessEvolvingStrategy(MultiProcessEvolvingStrategy):
            if isinstance(queried_knowledge, CoSTEERQueriedKnowledgeV2)
            else queried_former_failed_knowledge
        )
-        system_prompt = (
-            Environment(undefined=StrictUndefined)
-            .from_string(
-                coder_prompts["evolving_strategy_model_coder"]["system"],
-            )
-            .render(
-                scenario=self.scen.get_scenario_all_desc(filtered_tag=target_task.model_type),
-                queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
-                current_code=workspace.file_dict.get("model.py"),
-            )
+        system_prompt = T(".prompts:evolving_strategy_model_coder.system").r(
+            scenario=self.scen.get_scenario_all_desc(filtered_tag="model"),
+            queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
+            current_code=workspace.file_dict.get("model.py"),
        )

        queried_similar_successful_knowledge_to_render = queried_similar_successful_knowledge
        for _ in range(10):  # max attempt to reduce the length of user_prompt
-            user_prompt = (
-                Environment(undefined=StrictUndefined)
-                .from_string(
-                    coder_prompts["evolving_strategy_model_coder"]["user"],
-                )
-                .render(
-                    model_information_str=model_information_str,
-                    queried_similar_successful_knowledge=queried_similar_successful_knowledge_to_render,
-                    queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
-                )
-                .strip("\n")
+            user_prompt = T(".prompts:evolving_strategy_model_coder.user").r(
+                model_information_str=model_information_str,
+                queried_similar_successful_knowledge=queried_similar_successful_knowledge_to_render,
+                queried_former_failed_knowledge=queried_former_failed_knowledge_to_render,
            )
            if (
                APIBackend().build_messages_and_calculate_token(
@@ -5,10 +5,11 @@ from pathlib import Path
 from typing import Dict, Optional

 from rdagent.components.coder.CoSTEER.task import CoSTEERTask
+from rdagent.components.coder.model_coder.conf import MODEL_COSTEER_SETTINGS
 from rdagent.core.experiment import Experiment, FBWorkspace
 from rdagent.core.utils import cache_with_pickle
 from rdagent.oai.llm_utils import md5_hash
-from rdagent.utils.env import KGDockerEnv, QTDockerEnv
+from rdagent.utils.env import KGDockerEnv, QlibCondaConf, QlibCondaEnv, QTDockerEnv


 class ModelTask(CoSTEERTask):
@@ -19,6 +20,7 @@ class ModelTask(CoSTEERTask):
        architecture: str,
        *args,
        hyperparameters: Dict[str, str],
+        training_hyperparameters: Dict[str, str],
        formulation: str = None,
        variables: Dict[str, str] = None,
        model_type: Optional[str] = None,
@@ -28,6 +30,7 @@ class ModelTask(CoSTEERTask):
        self.architecture: str = architecture
        self.variables: str = variables
        self.hyperparameters: str = hyperparameters
+        self.training_hyperparameters: str = training_hyperparameters
        self.model_type: str = (
            model_type  # Tabular for tabular model, TimesSeries for time series model, Graph for graph model, XGBoost for XGBoost model
        )
@@ -41,6 +44,17 @@ description: {self.description}
        task_desc += f"architecture: {self.architecture}\n"
        task_desc += f"variables: {self.variables}\n" if self.variables else ""
        task_desc += f"hyperparameters: {self.hyperparameters}\n"
+        task_desc += f"training_hyperparameters: {self.training_hyperparameters}\n"
+        task_desc += f"model_type: {self.model_type}\n"
+        return task_desc
+
+    def get_task_brief_information(self):
+        task_desc = f"""name: {self.name}
+description: {self.description}
+"""
+        task_desc += f"architecture: {self.architecture}\n"
+        task_desc += f"hyperparameters: {self.hyperparameters}\n"
+        task_desc += f"training_hyperparameters: {self.training_hyperparameters}\n"
        task_desc += f"model_type: {self.model_type}\n"
        return task_desc

@@ -99,7 +113,15 @@ class ModelFBWorkspace(FBWorkspace):
    ):
        self.before_execute()
        try:
-            qtde = QTDockerEnv() if self.target_task.version == 1 else KGDockerEnv()
+            if self.target_task.version == 1:
+                if MODEL_COSTEER_SETTINGS.env_type == "docker":
+                    qtde = QTDockerEnv()
+                elif MODEL_COSTEER_SETTINGS.env_type == "conda":
+                    qtde = QlibCondaEnv(conf=QlibCondaConf())
+                else:
+                    raise ValueError(f"Unknown env_type: {MODEL_COSTEER_SETTINGS.env_type}")
+            else:
+                qtde = KGDockerEnv()
            qtde.prepare()

            if self.target_task.version == 1:
@@ -1,12 +1,10 @@
 import re
 from pathlib import Path

-from jinja2 import Environment, StrictUndefined
-
 from rdagent.components.coder.model_coder.model import ModelExperiment, ModelFBWorkspace
 from rdagent.core.developer import Developer
-from rdagent.core.prompts import Prompts
 from rdagent.oai.llm_utils import APIBackend
+from rdagent.utils.agent.tpl import T

 DIRNAME = Path(__file__).absolute().resolve().parent

@@ -17,18 +15,14 @@ class ModelCodeWriter(Developer[ModelExperiment]):
        for t in exp.sub_tasks:
            mti = ModelFBWorkspace(t)
            mti.prepare()
-            pr = Prompts(file_path=DIRNAME / "prompt.yaml")

-            user_prompt_tpl = Environment(undefined=StrictUndefined).from_string(pr["code_implement_user"])
-            sys_prompt_tpl = Environment(undefined=StrictUndefined).from_string(pr["code_implement_sys"])
-
-            user_prompt = user_prompt_tpl.render(
+            user_prompt = T(".prompts:code_implement_user").r(
                name=t.name,
                description=t.description,
                formulation=t.formulation,
                variables=t.variables,
            )
-            system_prompt = sys_prompt_tpl.render()
+            system_prompt = T(".prompts:code_implement_sys").r()

            resp = APIBackend().build_messages_and_create_chat_completion(user_prompt, system_prompt)

@@ -2,19 +2,16 @@ from __future__ import annotations

 import json
 import re
-from pathlib import Path

-from rdagent.components.coder.model_coder.model import ModelExperiment, ModelTask
+from rdagent.components.coder.model_coder.model import ModelTask
 from rdagent.components.document_reader.document_reader import (
    load_and_process_pdfs_by_langchain,
 )
 from rdagent.components.loader.task_loader import ModelTaskLoader
-from rdagent.core.prompts import Prompts
 from rdagent.log import rdagent_logger as logger
 from rdagent.oai.llm_utils import APIBackend
 from rdagent.scenarios.qlib.experiment.model_experiment import QlibModelExperiment
-
-document_process_prompts = Prompts(file_path=Path(__file__).parent / "prompts.yaml")
+from rdagent.utils.agent.tpl import T


 def extract_model_from_doc(doc_content: str) -> dict:
@@ -32,7 +29,7 @@ def extract_model_from_doc(doc_content: str) -> dict:
        {model_name: dict{description, formulation, variables}}
    """
    session = APIBackend().build_chat_session(
-        session_system_prompt=document_process_prompts["extract_model_formulation_system"],
+        session_system_prompt=T(".prompts:extract_model_formulation_system").r(),
    )
    current_user_prompt = doc_content