feat: enable finetune llm (#1055)

* feat: start with previous workspace

* feat: finetune llm

* add PrevModelLoadEvaluator

---------

Co-authored-by: Young <afe.young@gmail.com>
Co-authored-by: v-jianwan <v-jianwan@microsoft.com>
Co-authored-by: you-n-g <you-n-g@users.noreply.github.com>

rdagent/components/coder/data_science/conf.py

@@ -22,6 +22,8 @@ class DSCoderCoSTEERSettings(CoSTEERSettings):
     max_seconds_multiplier: int = 4
     env_type: str = "docker"
     # TODO: extract a function for env and conf.
+    extra_evaluator: list[str] = []
+    """Extra evaluators to use"""
 
     extra_eval: list[str] = []
     """

rdagent/components/coder/data_science/pipeline/__init__.py

@@ -141,6 +141,8 @@ class PipelineCoSTEER(DSCoSTEER):
         eval_l = [PipelineCoSTEEREvaluator(scen=scen)]
         if DS_RD_SETTING.enable_model_dump:
             eval_l.append(ModelDumpEvaluator(scen=scen, data_type="sample"))
+        for evaluator in settings.extra_evaluator:
+            eval_l.append(import_class(evaluator)(scen=scen))
 
         for extra_eval in DSCoderCoSTEERSettings().extra_eval:
             kls = import_class(extra_eval)

rdagent/components/coder/data_science/pipeline/prompts.yaml

@@ -216,7 +216,6 @@ pipeline_coder:
     {% endif %}
     {% endif %}
 
-
 pipeline_eval:
   system: |-
     {% include "scenarios.data_science.share:scen.role" %}