truefoundry 0.11.5rc2__tar.gz → 0.11.6__tar.gz

{truefoundry-0.11.5rc2 → truefoundry-0.11.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truefoundry
-Version: 0.11.5rc2
+Version: 0.11.6
 Summary: TrueFoundry CLI
 Author-email: TrueFoundry Team <abhishek@truefoundry.com>
 Requires-Python: <3.14,>=3.8.1

{truefoundry-0.11.5rc2 → truefoundry-0.11.6}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "truefoundry"
-version = "0.11.5rc2"  # This is a placeholder version, it will be replaced when releasing
+version = "0.11.6"  # This is a placeholder version, it will be replaced when releasing
 requires-python = ">=3.8.1,<3.14"
 description = "TrueFoundry CLI"
 authors = [{ name = "TrueFoundry Team", email = "abhishek@truefoundry.com" }]

{truefoundry-0.11.5rc2 → truefoundry-0.11.6}/truefoundry/deploy/python_deploy_codegen.py

@@ -187,7 +187,10 @@ def convert_deployment_config_to_python(
 
 
 def generate_python_snippet_for_trigger_job(
-    application_fqn: str, command: Optional[str], params: Optional[Dict[str, str]]
+    application_fqn: str,
+    command: Optional[str],
+    params: Optional[Dict[str, str]],
+    run_name_alias: Optional[str],
 ):
     job_run_python_template = """\
 from truefoundry.deploy import trigger_job
@@ -197,6 +200,7 @@ response = trigger_job(
   # You can pass command or params, but not both
   # command={{command}}
   # params={{params}}
+  # run_name_alias={{run_name_alias}}
 )
 
 print(response.jobRunName)
@@ -220,6 +224,17 @@ print(response.jobRunName)
         output_python_str = output_python_str.replace(
             "{{params}}", "<Enter Params(key-value pairs) here as python dict>"
         )
+    if run_name_alias is not None:
+        output_python_str = output_python_str.replace(
+            "# run_name_alias", "run_name_alias"
+        )
+        output_python_str = output_python_str.replace(
+            "{{run_name_alias}}", repr(run_name_alias)
+        )
+    else:
+        output_python_str = output_python_str.replace(
+            "{{run_name_alias}}", "<Enter Run Name Alias here>"
+        )
 
     return output_python_str
 
@@ -229,6 +244,7 @@ def generate_curl_snippet_for_trigger_job(
     application_id: str,
     command: Optional[str],
     params: Optional[Dict[str, str]],
+    run_name_alias: Optional[str],
 ):
     job_run_curl_request_template = """curl -X 'POST' \\
   '{{control_plane_url}}/api/svc/v1/jobs/trigger' \\
@@ -237,7 +253,10 @@ def generate_curl_snippet_for_trigger_job(
   -H 'Content-Type: application/json' \\
   -d '{
   "applicationId": "{{application_id}}",
-  "input": {{input}}
+  "input": {{input}},
+  "metadata": {
+    "job_run_name_alias": "{{run_name_alias}}"
+  }
 }'
 """
     output_curl_str = job_run_curl_request_template.replace(
@@ -247,4 +266,10 @@ def generate_curl_snippet_for_trigger_job(
     output_curl_str = output_curl_str.replace(
         "{{input}}", json.dumps({"command": command, "params": params}, indent=2)
     )
+    if run_name_alias is not None:
+        output_curl_str = output_curl_str.replace("{{run_name_alias}}", run_name_alias)
+    else:
+        output_curl_str = output_curl_str.replace(
+            "{{run_name_alias}}", "<Enter Run Name Alias here>"
+        )
     return output_curl_str

{truefoundry-0.11.5rc2 → truefoundry-0.11.6}/truefoundry/ml/integrations/huggingface/trainer_callback.py

@@ -31,6 +31,7 @@ class TrueFoundryMLCallback(TrainerCallback):
         log_checkpoints: bool = True,
         checkpoint_artifact_name: Optional[str] = None,
         auto_end_run_on_train_end: bool = False,
+        log_training_arguments: bool = True,
     ):
         """
         Args:
@@ -38,7 +39,7 @@ class TrueFoundryMLCallback(TrainerCallback):
             log_checkpoints: Whether to log checkpoints or not, defaults to True.
             checkpoint_artifact_name: The name of the artifact to log checkpoints to, required if log_checkpoints is True.
             auto_end_run_on_train_end: Whether to end the run automatically when training ends, defaults to False.
-
+            log_training_arguments: Whether to log the training arguments or not, defaults to True.
             Usage:
                 from transformers import Trainer
                 from truefoundry.ml.integrations.huggingface.trainer_callback import TrueFoundryMLCallback
@@ -67,6 +68,7 @@ class TrueFoundryMLCallback(TrainerCallback):
             )
         self._checkpoint_artifact_name = checkpoint_artifact_name
         self._auto_end_run_on_train_end = auto_end_run_on_train_end
+        self._log_training_arguments = log_training_arguments
 
     @classmethod
     def with_managed_run(
@@ -76,6 +78,7 @@ class TrueFoundryMLCallback(TrainerCallback):
         log_checkpoints: bool = True,
         checkpoint_artifact_name: Optional[str] = None,
         auto_end_run_on_train_end: bool = True,
+        log_training_arguments: bool = True,
     ) -> "TrueFoundryMLCallback":
         """
         Args:
@@ -84,7 +87,7 @@ class TrueFoundryMLCallback(TrainerCallback):
             log_checkpoints: Whether to log checkpoints or not, defaults to True.
             checkpoint_artifact_name: The name of the artifact to log checkpoints to, required if log_checkpoints is True.
             auto_end_run_on_train_end: Whether to end the run automatically when training ends, defaults to True.
-
+            log_training_arguments: Whether to log the training arguments or not, defaults to True.
         Usage:
             from transformers import Trainer
             from truefoundry.ml.integrations.huggingface.trainer_callback import TrueFoundryMLCallback
@@ -109,6 +112,7 @@ class TrueFoundryMLCallback(TrainerCallback):
             log_checkpoints=log_checkpoints,
             checkpoint_artifact_name=checkpoint_artifact_name,
             auto_end_run_on_train_end=auto_end_run_on_train_end,
+            log_training_arguments=log_training_arguments,
         )
 
     def _drop_non_finite_values(self, dct: Dict[str, Any]) -> Dict[str, Any]:
@@ -184,6 +188,21 @@ class TrueFoundryMLCallback(TrainerCallback):
             description=description,
         )
 
+    def on_train_begin(
+        self,
+        args: "TrainingArguments",
+        state: "TrainerState",
+        control: "TrainerControl",
+        **kwargs,
+    ):
+        if self._log_training_arguments:
+            training_arguments = args.to_sanitized_dict()
+            valid_types = (bool, int, float, str)
+            training_arguments = {
+                k: v for k, v in training_arguments.items() if type(v) in valid_types
+            }
+            self._run.log_params(training_arguments)
+
     def on_train_end(
         self,
         args: "TrainingArguments",