Merge pull request #1646 from basetenlabs/bump-version-0.9.91

Release 0.9.91

Author: marius-baseten

pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "truss"
-version = "0.9.90"
+version = "0.9.91"
 description = "A seamless bridge from model development to model delivery"
 license = "MIT"
 readme = "README.md"

truss-train/tests/import/deploy_checkpoints_config.py

@@ -12,7 +12,7 @@
             "HF_TOKEN": definitions.SecretReference(name="hf_access_token")
         }
     ),
-    checkpoint_details=definitions.CheckpointDetails(
+    checkpoint_details=definitions.CheckpointList(
         base_model_id="unsloth/gemma-3-1b-it",
         checkpoints=[
             definitions.Checkpoint(

truss-train/tests/import/project_with_deploy_checkpoints_config.py

@@ -12,7 +12,7 @@
             "HF_TOKEN": definitions.SecretReference(name="hf_access_token")
         }
     ),
-    checkpoint_details=definitions.CheckpointDetails(
+    checkpoint_details=definitions.CheckpointList(
         base_model_id="unsloth/gemma-3-1b-it",
         checkpoints=[
             definitions.Checkpoint(

truss-train/truss_train/definitions.py

@@ -87,14 +87,14 @@ def to_truss_config(self) -> truss_config.Checkpoint:
         )
 
 
-class CheckpointDetails(custom_types.SafeModel):
+class CheckpointList(custom_types.SafeModel):
     download_folder: str = truss_config.DEFAULT_TRAINING_CHECKPOINT_FOLDER
     base_model_id: Optional[str] = None
     checkpoints: List[Checkpoint] = []
 
-    def to_truss_config(self) -> truss_config.CheckpointConfiguration:
+    def to_truss_config(self) -> truss_config.CheckpointList:
         checkpoints = [checkpoint.to_truss_config() for checkpoint in self.checkpoints]
-        return truss_config.CheckpointConfiguration(
+        return truss_config.CheckpointList(
             checkpoints=checkpoints, download_folder=self.download_folder
         )
 
@@ -104,7 +104,7 @@ class DeployCheckpointsRuntime(custom_types.SafeModel):
 
 
 class DeployCheckpointsConfig(custom_types.SafeModel):
-    checkpoint_details: Optional[CheckpointDetails] = None
+    checkpoint_details: Optional[CheckpointList] = None
     model_name: Optional[str] = None
     deployment_name: Optional[str] = None
     runtime: Optional[DeployCheckpointsRuntime] = None

truss/base/truss_config.py

@@ -545,7 +545,7 @@ class Checkpoint(custom_types.ConfigModel):
     name: str
 
 
-class CheckpointConfiguration(custom_types.ConfigModel):
+class CheckpointList(custom_types.ConfigModel):
     download_folder: str = DEFAULT_TRAINING_CHECKPOINT_FOLDER
     checkpoints: list[Checkpoint] = pydantic.Field(default_factory=list)
 
@@ -586,7 +586,7 @@ class TrussConfig(custom_types.ConfigModel):
     trt_llm: Optional[trt_llm_config.TRTLLMConfiguration] = None
 
     # deploying from checkpoint
-    training_checkpoints: Optional[CheckpointConfiguration] = None
+    training_checkpoints: Optional[CheckpointList] = None
 
     # Internal / Legacy.
     input_type: str = "Any"

truss/cli/cli.py

@@ -361,7 +361,6 @@ def cleanup() -> None:
 @click.option("-n", "--name", type=click.STRING)
 @click.option(
     "--python-config/--no-python-config",
-    type=bool,
     default=False,
     help="Uses the code first tooling to build models.",
 )
@@ -453,7 +452,6 @@ def _extract_request_data(data: Optional[str], file: Optional[Path]):
 )
 @click.option(
     "--published",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -580,7 +578,6 @@ def run_python(script, target_directory):
 @click.option("--model-name", type=str, required=False, help="Name of the model")
 @click.option(
     "--publish",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -592,7 +589,6 @@ def run_python(script, target_directory):
 )
 @click.option(
     "--promote",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -613,7 +609,6 @@ def run_python(script, target_directory):
 )
 @click.option(
     "--preserve-previous-production-deployment",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -625,15 +620,13 @@ def run_python(script, target_directory):
 )
 @click.option(
     "--trusted",
-    type=bool,
     is_flag=True,
     required=False,
     default=None,
     help="[DEPRECATED] All models are trusted by default.",
 )
 @click.option(
     "--disable-truss-download",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -650,7 +643,6 @@ def run_python(script, target_directory):
 )
 @click.option(
     "--wait/--no-wait",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -667,16 +659,14 @@ def run_python(script, target_directory):
 )
 @click.option(
     "--include-git-info",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
     help=_INCLUDE_GIT_INFO_DOC,
 )
-@click.option("--tail", type=bool, is_flag=True)
+@click.option("--tail", is_flag=True)
 @click.option(
     "--preserve-env-instance-type/--no-preserve-env-instance-type",
-    type=bool,
     is_flag=True,
     required=False,
     default=None,
@@ -879,7 +869,7 @@ def push(
 @click.option("--remote", type=str, required=False)
 @click.option("--model-id", type=str, required=True)
 @click.option("--deployment-id", type=str, required=True)
-@click.option("--tail", type=bool, is_flag=True, help="Tail for ongoing logs.")
+@click.option("--tail", is_flag=True, help="Tail for ongoing logs.")
 @common_options()
 def model_logs(
     remote: Optional[str], model_id: str, deployment_id: str, tail: bool = False
@@ -1212,13 +1202,11 @@ def _create_chains_table(service) -> Tuple[rich.table.Table, List[str]]:
 )
 @click.option(
     "--publish/--no-publish",
-    type=bool,
     default=False,
     help="Create chainlets as published deployments.",
 )
 @click.option(
     "--promote/--no-promote",
-    type=bool,
     default=False,
     help="Replace production chainlets with newly deployed chainlets.",
 )
@@ -1233,13 +1221,11 @@ def _create_chains_table(service) -> Tuple[rich.table.Table, List[str]]:
 )
 @click.option(
     "--wait/--no-wait",
-    type=bool,
     default=True,
     help="Wait until all chainlets are ready (or deployment failed).",
 )
 @click.option(
     "--watch/--no-watch",
-    type=bool,
     default=False,
     help=(
         "Watches the chains source code and applies live patches. Using this option "
@@ -1250,7 +1236,6 @@ def _create_chains_table(service) -> Tuple[rich.table.Table, List[str]]:
 )
 @click.option(
     "--dryrun",
-    type=bool,
     default=False,
     is_flag=True,
     help="Produces only generated files, but doesn't deploy anything.",
@@ -1274,7 +1259,6 @@ def _create_chains_table(service) -> Tuple[rich.table.Table, List[str]]:
 )
 @click.option(
     "--include-git-info",
-    type=bool,
     is_flag=True,
     required=False,
     default=False,
@@ -1554,9 +1538,7 @@ def train():
 @train.command(name="push")
 @click.argument("config", type=Path, required=True)
 @click.option("--remote", type=str, required=False, help="Remote to use")
-@click.option(
-    "--tail", type=bool, is_flag=True, help="Tail for status + logs after push."
-)
+@click.option("--tail", is_flag=True, help="Tail for status + logs after push.")
 @common_options()
 def push_training_job(config: Path, remote: Optional[str], tail: bool):
     """Run a training job"""
@@ -1600,7 +1582,7 @@ def push_training_job(config: Path, remote: Optional[str], tail: bool):
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
-@click.option("--tail", type=bool, is_flag=True, help="Tail for ongoing logs.")
+@click.option("--tail", is_flag=True, help="Tail for ongoing logs.")
 @common_options()
 def get_job_logs(
     remote: Optional[str], project_id: Optional[str], job_id: Optional[str], tail: bool
@@ -1632,7 +1614,7 @@ def get_job_logs(
 @train.command(name="stop")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
-@click.option("--all", type=bool, is_flag=True, help="Stop all running jobs.")
+@click.option("--all", is_flag=True, help="Stop all running jobs.")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common_options()
 def stop_job(
@@ -1708,10 +1690,7 @@ def get_job_metrics(
     help="path to a python file that defines a DeployCheckpointsConfig",
 )
 @click.option(
-    "--dry-run",
-    type=bool,
-    is_flag=True,
-    help="Generate a truss config without deploying",
+    "--dry-run", is_flag=True, help="Generate a truss config without deploying"
 )
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common_options()

truss/cli/train/deploy_checkpoints.py

@@ -21,7 +21,7 @@
 from truss_train.definitions import (
     DEFAULT_LORA_RANK,
     Checkpoint,
-    CheckpointDetails,
+    CheckpointList,
     Compute,
     DeployCheckpointsConfig,
     DeployCheckpointsRuntime,
@@ -156,10 +156,10 @@ def _render_vllm_lora_truss_config(
 def _get_checkpoint_details(
     console: Console,
     remote_provider: BasetenRemote,
-    checkpoint_details: Optional[CheckpointDetails],
+    checkpoint_details: Optional[CheckpointList],
     project_id: Optional[str],
     job_id: Optional[str],
-) -> CheckpointDetails:
+) -> CheckpointList:
     if checkpoint_details and checkpoint_details.checkpoints:
         return _process_user_provided_checkpoints(checkpoint_details, remote_provider)
     else:
@@ -171,16 +171,16 @@ def _get_checkpoint_details(
 def _prompt_user_for_checkpoint_details(
     console: Console,
     remote_provider: BasetenRemote,
-    checkpoint_details: Optional[CheckpointDetails],
+    checkpoint_details: Optional[CheckpointList],
     project_id: Optional[str],
     job_id: Optional[str],
-) -> CheckpointDetails:
+) -> CheckpointList:
     project_id, job_id = get_most_recent_job(
         console, remote_provider, project_id, job_id
     )
     response_checkpoints = _fetch_checkpoints(remote_provider, project_id, job_id)
     if not checkpoint_details:
-        checkpoint_details = CheckpointDetails()
+        checkpoint_details = CheckpointList()
 
     # first, gather all checkpoint ids the user wants to deploy
     # allow the user to select a checkpoint
@@ -196,8 +196,8 @@ def _prompt_user_for_checkpoint_details(
 
 
 def _process_user_provided_checkpoints(
-    checkpoint_details: CheckpointDetails, remote_provider: BasetenRemote
-) -> CheckpointDetails:
+    checkpoint_details: CheckpointList, remote_provider: BasetenRemote
+) -> CheckpointList:
     # check if the user-provided checkpoint details are valid. Fill in missing values.
     checkpoints_by_training_job_id = {}
     for checkpoint in checkpoint_details.checkpoints:
@@ -342,9 +342,10 @@ def _get_runtime(
         hf_secret_name = _get_hf_secret_name(
             console, runtime.environment_variables.get(HF_TOKEN_ENVVAR_NAME)
         )
-        runtime.environment_variables[HF_TOKEN_ENVVAR_NAME] = SecretReference(
-            name=hf_secret_name
-        )
+        if hf_secret_name:
+            runtime.environment_variables[HF_TOKEN_ENVVAR_NAME] = SecretReference(
+                name=hf_secret_name
+            )
     return runtime
 
 

truss/cli/train/types.py

@@ -3,7 +3,7 @@
 from typing import Optional
 
 from truss_train.definitions import (
-    CheckpointDetails,
+    CheckpointList,
     Compute,
     DeployCheckpointsConfig,
     DeployCheckpointsRuntime,
@@ -23,7 +23,7 @@ class DeployCheckpointsConfigComplete(DeployCheckpointsConfig):
     removes the optional fileds. This helps provide type safety internal handling.
     """
 
-    checkpoint_details: CheckpointDetails
+    checkpoint_details: CheckpointList
     model_name: str
     deployment_name: str
     runtime: DeployCheckpointsRuntime

truss/remote/baseten/api.py

@@ -109,6 +109,8 @@ def _post_graphql_query(self, query: str, variables: Optional[dict] = None) -> d
         if errors:
             message = errors[0]["message"]
             error_code = errors[0].get("extensions", {}).get("code")
+            if errors[0].get("extensions", {}).get("description") is not None:
+                message = errors[0].get("extensions", {}).get("description")
             raise ApiError(message, error_code)
 
         return resp_dict

truss/tests/cli/train/test_deploy_checkpoints.py

@@ -83,7 +83,7 @@ def deploy_checkpoints_mock_checkbox(create_mock_prompt):
 
 def test_render_vllm_lora_truss_config():
     deploy_config = DeployCheckpointsConfigComplete(
-        checkpoint_details=definitions.CheckpointDetails(
+        checkpoint_details=definitions.CheckpointList(
             checkpoints=[
                 definitions.Checkpoint(
                     id="checkpoint-1",
@@ -181,7 +181,7 @@ def test_prepare_checkpoint_deploy_complete_config(
 ):
     # Create complete config with all fields specified
     complete_config = definitions.DeployCheckpointsConfig(
-        checkpoint_details=definitions.CheckpointDetails(
+        checkpoint_details=definitions.CheckpointList(
             checkpoints=[
                 definitions.Checkpoint(
                     id="checkpoint-1",

truss/tests/remote/baseten/test_api.py

@@ -34,6 +34,19 @@ def mock_graphql_error_response():
     return response
 
 
+def mock_graphql_error_response_with_description():
+    response = Response()
+    response.status_code = 200
+    response.json = mock.Mock(
+        return_value={
+            "errors": [
+                {"message": "error", "extensions": {"description": "descriptive_error"}}
+            ]
+        }
+    )
+    return response
+
+
 def mock_unsuccessful_response():
     response = Response()
     response.status_code = 400
@@ -141,6 +154,17 @@ def test_post_graphql_query_error(mock_post, baseten_api):
         baseten_api._post_graphql_query("sample_query_string")
 
 
+@mock.patch(
+    "requests.post", return_value=mock_graphql_error_response_with_description()
+)
+def test_post_graphql_query_error_with_description(mock_post, baseten_api):
+    with pytest.raises(ApiError) as exc_info:
+        baseten_api._post_graphql_query("sample_query_string")
+
+    exception = exc_info.value
+    assert str(exception) == "descriptive_error"
+
+
 @mock.patch("requests.post", return_value=mock_unsuccessful_response())
 def test_post_requests_error(mock_post, baseten_api):
     with pytest.raises(requests.exceptions.HTTPError):