truss 0.11.8rc2__py3-none-any.whl → 0.11.8rc4__py3-none-any.whl

truss/cli/train/core.py

@@ -156,7 +156,7 @@ def display_training_projects(projects: list[dict], remote_url: str) -> None:
         latest_job = project.get("latest_job") or {}
         if latest_job_id := latest_job.get("id", ""):
             latest_job_link = cli_common.format_link(
-                status_page_url(remote_url, latest_job_id), "link"
+                status_page_url(remote_url, project["id"], latest_job_id), "link"
             )
         else:
             latest_job_link = ""
@@ -305,8 +305,11 @@ def display_training_job(
     table.add_row("Created", cli_common.format_localized_time(job["created_at"]))
     table.add_row("Last Modified", cli_common.format_localized_time(job["updated_at"]))
     table.add_row(
-        "Status Page",
-        cli_common.format_link(status_page_url(remote_url, job["id"]), "link"),
+        "Job Page",
+        cli_common.format_link(
+            status_page_url(remote_url, job["training_project"]["id"], job["id"]),
+            "link",
+        ),
     )
 
     # Add error message if present
@@ -417,8 +420,8 @@ def download_checkpoint_artifacts(
     return urls_file
 
 
-def status_page_url(remote_url: str, training_job_id: str) -> str:
-    return f"{remote_url}/training/jobs/{training_job_id}"
+def status_page_url(remote_url: str, project_id: str, training_job_id: str) -> str:
+    return f"{remote_url}/training/{project_id}/logs/{training_job_id}"
 
 
 def _get_all_train_init_example_options(

truss/cli/train_commands.py

@@ -41,13 +41,14 @@ truss_cli.add_command(train)
 
 def _print_training_job_success_message(
     job_id: str,
+    project_id: str,
     project_name: str,
-    job_object: TrainingJob,
+    job_object: Optional[TrainingJob],
     remote_provider: BasetenRemote,
 ) -> None:
     """Print success message and helpful commands for a training job."""
     console.print("✨ Training job successfully created!", style="green")
-    should_print_cache_summary = (
+    should_print_cache_summary = job_object and (
         job_object.runtime.enable_cache
         or job_object.runtime.cache_config
         and job_object.runtime.cache_config.enabled
@@ -64,7 +65,7 @@ def _print_training_job_success_message(
         f"🔍 View metrics for your job via "
         f"[cyan]'truss train metrics --job-id {job_id}'[/cyan]\n"
         f"{cache_summary_snippet}"
-        f"🌐 Status page: {common.format_link(core.status_page_url(remote_provider.remote_url, job_id))}"
+        f"🌐 View job in the UI: {common.format_link(core.status_page_url(remote_provider.remote_url, project_id, job_id))}"
     )
 
 
@@ -80,8 +81,13 @@ def _handle_post_create_logic(
             style="green",
         )
     else:
+        # recreate currently doesn't pass back a job object.
         _print_training_job_success_message(
-            job_id, project_name, job_resp["job_object"], remote_provider
+            job_id,
+            project_id,
+            project_name,
+            job_resp.get("job_object"),
+            remote_provider,
         )
 
     if tail:
@@ -156,11 +162,16 @@ def recreate_training_job(job_id: Optional[str], remote: Optional[str], tail: bo
 @train.command(name="logs")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--tail", is_flag=True, help="Tail for ongoing logs.")
 @common.common_options()
 def get_job_logs(
-    remote: Optional[str], project_id: Optional[str], job_id: Optional[str], tail: bool
+    remote: Optional[str],
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    tail: bool,
 ):
     """Fetch logs for a training job"""
 
@@ -170,6 +181,10 @@ def get_job_logs(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
+
     project_id, job_id = train_common.get_most_recent_job(
         remote_provider, project_id, job_id
     )
@@ -188,12 +203,17 @@ def get_job_logs(
 
 @train.command(name="stop")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--all", is_flag=True, help="Stop all running jobs.")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def stop_job(
-    project_id: Optional[str], job_id: Optional[str], all: bool, remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    all: bool,
+    remote: Optional[str],
 ):
     """Stop a training job"""
 
@@ -203,6 +223,9 @@ def stop_job(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     if all:
         train_cli.stop_all_jobs(remote_provider, project_id)
     else:
@@ -217,13 +240,17 @@ def stop_job(
 @click.option(
     "--project-id", type=str, required=False, help="View training jobs for a project."
 )
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option(
     "--job-id", type=str, required=False, help="View a specific training job."
 )
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def view_training(
-    project_id: Optional[str], job_id: Optional[str], remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    remote: Optional[str],
 ):
     """List all training jobs for a project"""
 
@@ -233,16 +260,24 @@ def view_training(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
+
     train_cli.view_training_details(remote_provider, project_id, job_id)
 
 
 @train.command(name="metrics")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def get_job_metrics(
-    project_id: Optional[str], job_id: Optional[str], remote: Optional[str]
+    project_id: Optional[str],
+    project: Optional[str],
+    job_id: Optional[str],
+    remote: Optional[str],
 ):
     """Get metrics for a training job"""
 
@@ -252,11 +287,15 @@ def get_job_metrics(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     train_cli.view_training_job_metrics(remote_provider, project_id, job_id)
 
 
 @train.command(name="deploy_checkpoints")
 @click.option("--project-id", type=str, required=False, help="Project ID.")
+@click.option("--project", type=str, required=False, help="Project name or project id.")
 @click.option("--job-id", type=str, required=False, help="Job ID.")
 @click.option(
     "--config",
@@ -271,6 +310,7 @@ def get_job_metrics(
 @common.common_options()
 def deploy_checkpoints(
     project_id: Optional[str],
+    project: Optional[str],
     job_id: Optional[str],
     config: Optional[str],
     remote: Optional[str],
@@ -286,6 +326,9 @@ def deploy_checkpoints(
     remote_provider: BasetenRemote = cast(
         BasetenRemote, RemoteFactory.create(remote=remote)
     )
+    project_id = _maybe_resolve_project_id_from_id_or_name(
+        remote_provider, project_id=project_id, project=project
+    )
     prepare_checkpoint_result = train_cli.prepare_checkpoint_deploy(
         remote_provider,
         train_cli.PrepareCheckpointArgs(
@@ -492,3 +535,15 @@ def view_cache_summary(project: str, remote: Optional[str], sort: str, order: st
     )
 
     train_cli.view_cache_summary_by_project(remote_provider, project, sort, order)
+
+
+def _maybe_resolve_project_id_from_id_or_name(
+    remote_provider: BasetenRemote, project_id: Optional[str], project: Optional[str]
+) -> Optional[str]:
+    """resolve the project_id or project. `project` can be name or id"""
+    if project and project_id:
+        console.print("Both `project-id` and `project` provided. Using `project`.")
+    project_str = project or project_id
+    if not project_str:
+        return None
+    return train_cli.fetch_project_by_name_or_id(remote_provider, project_str)["id"]

truss/templates/base.Dockerfile.jinja

@@ -33,7 +33,7 @@ RUN useradd -u {{ app_user_uid }} -ms /bin/bash {{ app_username }}
 ENV DEBIAN_FRONTEND=noninteractive
 
 
-{%- set UV_VERSION = "0.7.19" %}
+{%- set UV_VERSION = "0.8.22" %}
 {#
 NB(nikhil): We use a semi-complex uv installation command across the board:
 - A generous UV_HTTP_TIMEOUT (5m) for packages that take a long time to install.
@@ -59,10 +59,12 @@ RUN {{ python_exec_path }} -c "import sys; \
 {% endblock %}
 
 {% block install_uv %}
-{# Install `uv` and `curl` if not already present in the image. #}
+{# Install `uv` and `curl` if not already present in the image. We validate the expected location for `uv` at the very end
+   due to limitations with `pipefail` in Docker context. #}
 RUN if ! command -v uv >/dev/null 2>&1; then \
     command -v curl >/dev/null 2>&1 || (apt update && apt install -y curl) && \
-    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/{{ UV_VERSION }}/install.sh | sh; \
+    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/{{ UV_VERSION }}/install.sh | sh && \
+    test -x ${HOME}/.local/bin/uv; \
 fi
 {# Add the user's local bin to the path, used by uv. #}
 ENV PATH=${PATH}:${HOME}/.local/bin

truss/templates/control/control/application.py

@@ -2,13 +2,15 @@ import asyncio
 import logging
 import logging.config
 import re
+import traceback
+from functools import wraps
 from pathlib import Path
-from typing import Dict
+from typing import Any, Callable, Coroutine, Dict
 
 import httpx
 from endpoints import control_app
-from fastapi import FastAPI
-from fastapi.responses import JSONResponse
+from fastapi import FastAPI, Request
+from fastapi.responses import JSONResponse, Response
 from helpers.errors import ModelLoadFailed, PatchApplicatonError
 from helpers.inference_server_controller import InferenceServerController
 from helpers.inference_server_process_controller import InferenceServerProcessController
@@ -18,17 +20,43 @@ from shared import log_config
 from starlette.datastructures import State
 
 
+def sanitized_exception_logging(
+    original_handler: Callable[[Request, Any], Coroutine[Any, Any, Response]],
+) -> Callable[[Request, Any], Coroutine[Any, Any, Response]]:
+    @wraps(original_handler)
+    async def wrapper(request: Request, error: Exception) -> Response:
+        logger = request.app.state.logger
+        traceback_error = traceback.extract_tb(error.__traceback__)
+
+        if traceback_error:
+            last_frame = traceback_error[-1]
+            concise_error = (
+                f'  File "{last_frame.filename}", line {last_frame.lineno}, in {last_frame.name}\n'
+                f"    {last_frame.line}"
+            )
+        else:
+            concise_error = f"{type(error).__name__}: {error}"
+
+        logger.error(concise_error)
+        return await original_handler(request, error)
+
+    return wrapper
+
+
+@sanitized_exception_logging
 async def handle_patch_error(_, exc):
     error_type = _camel_to_snake_case(type(exc).__name__)
     return JSONResponse(content={"error": {"type": error_type, "msg": str(exc)}})
 
 
+@sanitized_exception_logging
 async def generic_error_handler(_, exc):
     return JSONResponse(
         content={"error": {"type": "unknown", "msg": f"{type(exc)}: {exc}"}}
     )
 
 
+@sanitized_exception_logging
 async def handle_model_load_failed(_, error):
     # Model load failures should result in 503 status
     return JSONResponse({"error": str(error)}, 503)

truss/tests/test_data/server.Dockerfile

@@ -17,7 +17,8 @@ RUN /usr/local/bin/python3 -c "import sys; \
     || { echo "ERROR: Supplied base image does not have 3.8 <= python <= 3.13"; exit 1; }
 RUN if ! command -v uv >/dev/null 2>&1; then \
     command -v curl >/dev/null 2>&1 || (apt update && apt install -y curl) && \
-    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/0.7.19/install.sh | sh; \
+    curl -LsSf --retry 5 --retry-delay 5 https://astral.sh/uv/0.8.22/install.sh | sh && \
+    test -x ${HOME}/.local/bin/uv; \
 fi
 ENV PATH=${PATH}:${HOME}/.local/bin
 ENV PYTHONUNBUFFERED="True"

{truss-0.11.8rc2.dist-info → truss-0.11.8rc4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.8rc2
+Version: 0.11.8rc4
 Summary: A seamless bridge from model development to model delivery
 Project-URL: Repository, https://github.com/basetenlabs/truss
 Project-URL: Homepage, https://truss.baseten.co

{truss-0.11.8rc2.dist-info → truss-0.11.8rc4.dist-info}/RECORD

@@ -11,13 +11,13 @@ truss/base/truss_spec.py,sha256=jFVF79CXoEEspl2kXBAPyi-rwISReIGTdobGpaIhwJw,5979
 truss/cli/chains_commands.py,sha256=Kpa5mCg6URAJQE2ZmZfVQFhjBHEitKT28tKiW0H6XAI,17406
 truss/cli/cli.py,sha256=PaMkuwXZflkU7sa1tEoT_Zmy-iBkEZs1m4IVqcieaeo,30367
 truss/cli/remote_cli.py,sha256=G_xCKRXzgkCmkiZJhUFfsv5YSVgde1jLA5LPQitpZgI,1905
-truss/cli/train_commands.py,sha256=TZhtvofviWQF34pYppRCaQ6qayTsvPnx6afTrYbFpOM,17319
+truss/cli/train_commands.py,sha256=sJN_Qpq-ejHY7KZ555n37PzpYEF-gZiQCVHW5JwUPHo,19303
 truss/cli/logs/base_watcher.py,sha256=KKyd7lIrdaEeDVt8EtjMioSPGVpLyOcF0ewyzE_GGdQ,2785
 truss/cli/logs/model_log_watcher.py,sha256=NACcP-wkcaroYa2Cb9BZC7Yr0554WZa_FSM2LXOf4A8,1263
 truss/cli/logs/training_log_watcher.py,sha256=r6HRqrLnz-PiKTUXiDYYxg4ZnP8vYcXlEX1YmgHhzlo,1173
 truss/cli/logs/utils.py,sha256=z-U_FG4BUzdZLbE3BnXb4DZQ0zt3LSZ3PiQpLaDuc3o,1031
 truss/cli/train/common.py,sha256=xTR41U5FeSndXfNBBHF9wF5XwZH1sOIVFlv-XHjsKIU,1547
-truss/cli/train/core.py,sha256=Xyl0SgCXIyvoQJLnkIvX6R2gWUVyeDiFHLB1caWJZjs,26154
+truss/cli/train/core.py,sha256=cqE0iBEElqP3oMDpCimuwpCwFyqfOYEBhweCZat7GiI,26262
 truss/cli/train/deploy_from_checkpoint_config.yml,sha256=mktaVrfhN8Kjx1UveC4xr-gTW-kjwbHvq6bx_LpO-Wg,371
 truss/cli/train/deploy_from_checkpoint_config_whisper.yml,sha256=6GbOorYC8ml0UyOUvuBpFO_fuYtYE646JqsalR-D4oY,406
 truss/cli/train/metrics_watcher.py,sha256=smz-zrEsBj_-wJHI0pAZ-EAPrvfCWzq1eQjGiFNM-Mk,12755
@@ -66,14 +66,14 @@ truss/remote/baseten/utils/time.py,sha256=Ry9GMjYnbIGYVIGwtmv4V8ljWjvdcaCf5NOQzl
 truss/remote/baseten/utils/transfer.py,sha256=d3VptuQb6M1nyS6kz0BAfeOYDLkMKUjatJXpY-mp-As,1548
 truss/templates/README.md.jinja,sha256=N7CJdyldZuJamj5jLh47le0hFBdu9irVsTBqoxhPNPQ,2476
 truss/templates/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-truss/templates/base.Dockerfile.jinja,sha256=93UgFjjDzmRILsWjvQBMcWuHY2j-HlP8pkCLmvZz_0o,5098
+truss/templates/base.Dockerfile.jinja,sha256=DgvNqmkt5QRrqy7a9hw6HffmVTUrdGl06oVZ4eeha5Y,5253
 truss/templates/cache.Dockerfile.jinja,sha256=1qZqDo1phrcqi-Vwol-VafYJkADsBbQWU6huQ-_1x00,1146
 truss/templates/cache_requirements.txt,sha256=xoPoJ-OVnf1z6oq_RVM3vCr3ionByyqMLj7wGs61nUs,87
 truss/templates/copy_cache_files.Dockerfile.jinja,sha256=Os5zFdYLZ_AfCRGq4RcpVTObOTwL7zvmwYcvOzd_Zqo,126
 truss/templates/docker_server_requirements.txt,sha256=PyhOPKAmKW1N2vLvTfLMwsEtuGpoRrbWuNo7tT6v2Mc,18
 truss/templates/server.Dockerfile.jinja,sha256=CUYnF_hgxPGq2re7__0UPWlwzOHMoFkxp6NVKi3U16s,7071
 truss/templates/control/requirements.txt,sha256=nqqNmlTwFeV8sV4fqwItwzzd_egADBP_e-cEopXBJ4k,358
-truss/templates/control/control/application.py,sha256=jYeta6hWe1SkfLL3W4IDmdYjg3ZuKqI_UagWYs5RB_E,3793
+truss/templates/control/control/application.py,sha256=Js6I1b5lxgi8Ylyj2-g3UeEqpKnSvvdYiXq-jSq2XX8,4803
 truss/templates/control/control/endpoints.py,sha256=VQ1lvZjFvR091yRkiFdvXw1Q7PiNGXT9rJwY7_sX6yg,11828
 truss/templates/control/control/server.py,sha256=R4Y219i1dcz0kkksN8obLoX-YXWGo9iW1igindyG50c,3128
 truss/templates/control/control/helpers/context_managers.py,sha256=W6dyFgLBhPa5meqrOb3w_phMtKfaJI-GhwUfpiycDc8,413
@@ -184,7 +184,7 @@ truss/tests/test_data/pima-indians-diabetes.csv,sha256=BvW3ws17ymhv2k-S6rX2Hn_2Q
 truss/tests/test_data/readme_int_example.md,sha256=fuHvpLtdkJy1f4NAR_djotVBdzusHYNXc-Fwh588XAE,1586
 truss/tests/test_data/readme_no_example.md,sha256=T2CzFMRvICXeX3_5XbFoqhHchcHGot-xM7izx34B3aQ,1607
 truss/tests/test_data/readme_str_example.md,sha256=fP4pvMqgLdIapaOf_BgRiV0H7pw4so0RNxrlq5lbROE,1726
-truss/tests/test_data/server.Dockerfile,sha256=64M9aAB-Ar5Yl0k9r5WFxVb-4iUShm_OR1w9pKJfDgg,1985
+truss/tests/test_data/server.Dockerfile,sha256=3rWiU3RxBh0PIh8pS23FnP8UA-ErmzxTlgf_J22mMQ8,2024
 truss/tests/test_data/annotated_types_truss/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 truss/tests/test_data/annotated_types_truss/config.yaml,sha256=B-ZyyjLLqtxGfXj2tkH68Hy7NOMB_coYvoWyWom61g0,147
 truss/tests/test_data/annotated_types_truss/model/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -368,8 +368,8 @@ truss_train/deployment.py,sha256=lWWANSuzBWu2M4oK4qD7n-oVR1JKdmw2Pn5BJQHg-Ck,307
 truss_train/loader.py,sha256=0o66EjBaHc2YY4syxxHVR4ordJWs13lNXnKjKq2wq0U,1630
 truss_train/public_api.py,sha256=9N_NstiUlmBuLUwH_fNG_1x7OhGCytZLNvqKXBlStrM,1220
 truss_train/restore_from_checkpoint.py,sha256=8hdPm-WSgkt74HDPjvCjZMBpvA9MwtoYsxVjOoa7BaM,1176
-truss-0.11.8rc2.dist-info/METADATA,sha256=NIPV3aMNwl9z0RIyWN2JObtaNyTFnxccFdJObD4T0Hc,6680
-truss-0.11.8rc2.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-truss-0.11.8rc2.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
-truss-0.11.8rc2.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
-truss-0.11.8rc2.dist-info/RECORD,,
+truss-0.11.8rc4.dist-info/METADATA,sha256=L17Z01_P_5zrhwF9d9sK1tGmAS-_vLI1YoVAVs0Vpvk,6680
+truss-0.11.8rc4.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+truss-0.11.8rc4.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
+truss-0.11.8rc4.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
+truss-0.11.8rc4.dist-info/RECORD,,