Save/Load config as yaml

Makefile

@@ -27,6 +27,9 @@ clean-env:
 clean-data:
 	rm -rf data
 
+clean-output:
+	rm -rf output
+
 reports:
 	mkdir -p reports
 

README.md

@@ -166,5 +166,15 @@ python prompt.py -e examples/output_oasst1
 
 All open-source datasets and models are posted on [H2O.ai's Hugging Face page](https://huggingface.co/h2oai/).
 
+
+## Changelog
+The field is rapidly evolving, and we are constantly adding new features and fixing bugs.
+While we are striving to converge to a stable framework, at this early point of development certain changes may break your existing experiments. 
+We thus recommend to pin the version of the framework to the one you used for your experiments. 
+Below, we list a summary of the changes that may affect older experiments:
+- [PR 12](https://github.com/h2oai/h2o-llmstudio/pull/12). Experiment configurations are now stored in yaml format,
+allowing for more flexibility in the configuration while making it much easier to be backward compatible. Old experiment configurations that are stored in pickle format will be converted to yaml format automatically.
+- [PR 40](https://github.com/h2oai/h2o-llmstudio/pull/40). Datasets can now use past chat history as input. This feature can be enabled by setting `Parent Id Column` in the dataset configuration.
+
 ## License
 H2O LLM Studio is licensed under the Apache 2.0 license. Please see the [LICENSE](LICENSE) file for more information.

app.py

@@ -29,7 +29,6 @@ async def serve(q: Q):
     copy_expando(q.args, q.client)
 
     await initialize_client(q)
-
     await handle(q)
 
     if not q.args["experiment/display/chat/chatbot"]:

app_utils/db.py

@@ -149,3 +149,6 @@ def rename_experiment(self, id: int, new_name: str, new_path: str) -> None:
         experiment.name = new_name
         experiment.path = new_path
         self._session.commit()
+
+    def update(self) -> None:
+        self._session.commit()

app_utils/initializers.py

@@ -7,17 +7,19 @@
 from h2o_wave import Q
 
 from app_utils.sections.common import interface
-from llm_studio.src.utils.config_utils import load_config
-
+from llm_studio.src.utils.config_utils import (
+    load_config_py,
+    save_config_yaml,
+)
 from .config import default_cfg
 from .db import Database, Dataset
+from .migration import migrate_app
 from .utils import (
     get_data_dir,
     get_db_path,
     get_user_name,
     load_user_settings,
     prepare_default_dataset,
-    save_dill,
 )
 
 logger = logging.getLogger(__name__)
@@ -38,21 +40,21 @@ def import_data(q: Q):
 
             df = prepare_default_dataset(path)
 
-            cfg = load_config(
+            cfg = load_config_py(
                 config_path=os.path.join(
                     "llm_studio/python_configs", default_cfg.cfg_file
                 ),
                 config_name="ConfigProblemBase",
             )
 
             cfg.dataset.train_dataframe = os.path.join(path, "train_full.pq")
-            cfg.dataset.prompt_column = "instruction"
+            cfg.dataset.prompt_column = ("instruction",)
             cfg.dataset.answer_column = "output"
             cfg.dataset.parent_id_column = "None"
 
-            cfg_path = os.path.join(path, f"{default_cfg.cfg_file}.p")
+            cfg_path = os.path.join(path, f"{default_cfg.cfg_file}.yaml")
 
-            save_dill(cfg_path, cfg)
+            save_config_yaml(cfg_path, cfg)
 
             dataset = Dataset(
                 id=1,
@@ -96,6 +98,7 @@ async def initialize_client(q: Q) -> None:
 
         load_user_settings(q)
 
+        await migrate_app(q)
         await interface(q)
 
         q.args[default_cfg.start_page] = True
@@ -128,7 +131,6 @@ async def initialize_app(q: Q) -> None:
         script_sources.append(url)
 
     q.app["script_sources"] = script_sources
-
     q.app["initialized"] = True
 
     logger.info("Initializing app ... done")

app_utils/migration.py

@@ -0,0 +1,62 @@
+import logging
+import os
+
+import dill
+from h2o_wave import Q
+
+from app_utils.db import Database
+from app_utils.utils import get_data_dir, get_output_dir
+from llm_studio.src.utils.config_utils import save_config_yaml
+
+logger = logging.getLogger(__name__)
+
+
+async def migrate_app(q: Q) -> None:
+    """
+    Migration scripts for the app.
+    """
+    migrate_pickle_to_yaml(q)
+    migrate_database_pickle_to_yaml(q)
+
+
+def migrate_pickle_to_yaml(q: Q) -> None:
+    """
+    Change from pickle -> yaml for config files.
+    introduced in https://github.com/h2oai/h2o-llmstudio/pull/12
+    """
+    data_dir = get_data_dir(q)
+    output_dir = get_output_dir(q)
+
+    for dir in [data_dir, output_dir]:
+        if os.path.exists(dir):
+            for root, dirs, files in os.walk(dir):
+                for file in files:
+                    if file.endswith(".p") and not os.path.exists(
+                        os.path.join(root, file.replace(".p", ".yaml"))
+                    ):
+                        try:
+                            with open(os.path.join(root, file), "rb") as f:
+                                cfg = dill.load(f)
+                                save_config_yaml(
+                                    os.path.join(root, file.replace(".p", ".yaml")), cfg
+                                )
+                                logger.info(
+                                    f"migrated {os.path.join(root, file)} to yaml"
+                                )
+                        except Exception as e:
+                            logger.error(
+                                f"Could not migrate {os.path.join(root, file)} "
+                                f"to yaml: {e}"
+                            )
+
+
+def migrate_database_pickle_to_yaml(q: Q) -> None:
+    """
+    Change from pickle -> yaml for config files.
+    introduced in https://github.com/h2oai/h2o-llmstudio/pull/12
+    """
+    db: Database = q.client.app_db
+    for dataset_id in db.get_datasets_df()["id"]:
+        dataset = db.get_dataset(dataset_id)
+        dataset.config_file = dataset.config_file.replace(".p", ".yaml")
+        db.update()

app_utils/sections/dataset.py

@@ -25,17 +25,19 @@
     get_problem_types,
     get_unique_dataset_name,
     kaggle_download,
-    load_dill,
     local_download,
     make_label,
     parse_ui_elements,
     remove_temp_files,
     s3_download,
     s3_file_options,
-    save_dill,
 )
 from app_utils.wave_utils import ui_table_from_df
-from llm_studio.src.utils.config_utils import load_config
+from llm_studio.src.utils.config_utils import (
+    load_config_py,
+    load_config_yaml,
+    save_config_yaml,
+)
 from llm_studio.src.utils.data_utils import (
     get_fill_columns,
     read_dataframe,
@@ -417,7 +419,7 @@ async def dataset_import(
                     q.client["dataset/import/cfg_file"], model_types[0][0]
                 )
             if not edit:
-                q.client["dataset/import/cfg"] = load_config(
+                q.client["dataset/import/cfg"] = load_config_py(
                     config_path=(
                         f"llm_studio/python_configs/"
                         f"{q.client['dataset/import/cfg_file']}"
@@ -586,7 +588,9 @@ async def dataset_import(
                 # change the default validation strategy if validation df set
                 if cfg.dataset.validation_dataframe != "None":
                     cfg.dataset.validation_strategy = "custom"
-                save_dill(f"{new_path}/{q.client['dataset/import/cfg_file']}.p", cfg)
+                save_config_yaml(
+                    f"{new_path}/{q.client['dataset/import/cfg_file']}.yaml", cfg
+                )
 
                 train_rows = None
                 if os.path.exists(cfg.dataset.train_dataframe):
@@ -603,7 +607,7 @@ async def dataset_import(
                     id=q.client["dataset/import/id"],
                     name=q.client["dataset/import/name"],
                     path=new_path,
-                    config_file=f"{new_path}/{q.client['dataset/import/cfg_file']}.p",
+                    config_file=f"{new_path}/{q.client['dataset/import/cfg_file']}.yaml",
                     train_rows=train_rows,
                     validation_rows=validation_rows,
                 )
@@ -681,7 +685,7 @@ async def dataset_merge(q: Q, step, error=""):
             has_experiment = False
 
         current_files = os.listdir(current_dir)
-        current_files = [x for x in current_files if not x.endswith(".p")]
+        current_files = [x for x in current_files if not x.endswith(".yaml")]
         target_files = os.listdir(target_dir)
         overlapping_files = list(set(current_files).intersection(set(target_files)))
         rename_map = {}
@@ -834,7 +838,7 @@ async def dataset_newexperiment(q: Q, dataset_id: int):
     dataset = q.client.app_db.get_dataset(dataset_id)
 
     q.client["experiment/start/cfg_file"] = dataset.config_file.split("/")[-1].replace(
-        ".p", ""
+        ".yaml", ""
     )
     q.client["experiment/start/cfg_category"] = q.client[
         "experiment/start/cfg_file"
@@ -877,15 +881,15 @@ async def dataset_edit(
     q.client["dataset/import/id"] = dataset_id
 
     q.client["dataset/import/cfg_file"] = dataset.config_file.split("/")[-1].replace(
-        ".p", ""
+        ".yaml", ""
     )
     q.client["dataset/import/cfg_category"] = q.client["dataset/import/cfg_file"].split(
         "_"
     )[0]
     q.client["dataset/import/path"] = dataset.path
     q.client["dataset/import/name"] = dataset.name
     q.client["dataset/import/original_name"] = dataset.name
-    q.client["dataset/import/cfg"] = load_dill(dataset.config_file)
+    q.client["dataset/import/cfg"] = load_config_yaml(dataset.config_file)
 
     if allow_merge and experiments_df.shape[0]:
         allow_merge = False
@@ -961,8 +965,7 @@ async def dataset_display(q: Q) -> None:
         q.client["dataset/display/id"]
     ]
     dataset = q.client.app_db.get_dataset(dataset_id)
-    config_file = dataset.config_file
-    cfg = load_dill(config_file)
+    cfg = load_config_yaml(dataset.config_file)
 
     has_train_df = cfg.dataset.train_dataframe != "None"