truss 0.11.10rc3__py3-none-any.whl → 0.11.10rc5__py3-none-any.whl

truss/base/trt_llm_config.py

@@ -577,6 +577,7 @@ class TRTLLMConfigurationV2(PydanticTrTBaseModel):
             "quantization_type",
             "quantization_config",
             "skip_build_result",
+            "num_builder_gpus",
         ]
 
         build_settings = self.build.model_dump(exclude_unset=True)

truss/cli/train/core.py

@@ -16,7 +16,11 @@ from rich.text import Text
 
 from truss.cli.train import common, deploy_checkpoints
 from truss.cli.train.metrics_watcher import MetricsWatcher
-from truss.cli.train.types import DeployCheckpointArgs, DeployCheckpointsConfigComplete
+from truss.cli.train.types import (
+    DeployCheckpointArgs,
+    DeployCheckpointsConfigComplete,
+    DeploySuccessResult,
+)
 from truss.cli.utils import common as cli_common
 from truss.cli.utils.output import console
 from truss.remote.baseten.custom_types import (
@@ -244,17 +248,25 @@ def view_training_job_metrics(
 
 def create_model_version_from_inference_template(
     remote_provider: BasetenRemote, args: DeployCheckpointArgs
-) -> DeployCheckpointsConfigComplete:
+) -> DeploySuccessResult:
     if not args.deploy_config_path:
         return deploy_checkpoints.create_model_version_from_inference_template(
-            remote_provider, DeployCheckpointsConfig(), args.project_id, args.job_id
+            remote_provider,
+            DeployCheckpointsConfig(),
+            args.project_id,
+            args.job_id,
+            args.dry_run,
         )
     #### User provided a checkpoint deploy config file
     with loader.import_deploy_checkpoints_config(
         Path(args.deploy_config_path)
     ) as checkpoint_deploy:
         return deploy_checkpoints.create_model_version_from_inference_template(
-            remote_provider, checkpoint_deploy, args.project_id, args.job_id
+            remote_provider,
+            checkpoint_deploy,
+            args.project_id,
+            args.job_id,
+            args.dry_run,
         )
 
 

truss/cli/train/deploy_checkpoints/deploy_checkpoints.py

@@ -1,3 +1,4 @@
+import json
 import re
 from collections import OrderedDict
 from typing import List, Optional, Union
@@ -7,7 +8,11 @@ from InquirerPy import inquirer
 
 from truss.base import truss_config
 from truss.cli.train import common
-from truss.cli.train.types import DeployCheckpointsConfigComplete
+from truss.cli.train.types import (
+    DeployCheckpointsConfigComplete,
+    DeploySuccessModelVersion,
+    DeploySuccessResult,
+)
 from truss.cli.utils.output import console
 from truss.remote.baseten.remote import BasetenRemote
 from truss_train.definitions import (
@@ -35,13 +40,16 @@ def create_model_version_from_inference_template(
     checkpoint_deploy_config: DeployCheckpointsConfig,
     project_id: Optional[str],
     job_id: Optional[str],
-) -> DeployCheckpointsConfigComplete:
+    dry_run: bool,
+) -> DeploySuccessResult:
     checkpoint_deploy_config = _hydrate_deploy_config(
         checkpoint_deploy_config, remote_provider, project_id, job_id
     )
 
     request_data = _build_inference_template_request(
-        checkpoint_deploy_config, remote_provider
+        checkpoint_deploy_config=checkpoint_deploy_config,
+        remote_provider=remote_provider,
+        dry_run=dry_run,
     )
 
     # Call the GraphQL mutation to create model version from inference template
@@ -49,7 +57,9 @@ def create_model_version_from_inference_template(
         result = remote_provider.api.create_model_version_from_inference_template(
             request_data
         )
+        truss_config_result = _get_truss_config_from_result(result)
 
+        model_version = None
         if result and result.get("model_version"):
             console.print(
                 f"Successfully created model version: {result['model_version']['name']}",
@@ -58,7 +68,10 @@ def create_model_version_from_inference_template(
             console.print(
                 f"Model version ID: {result['model_version']['id']}", style="yellow"
             )
-        else:
+            model_version = DeploySuccessModelVersion.model_validate(
+                result["model_version"]
+            )
+        elif not dry_run:
             console.print(
                 "Warning: Unexpected response format from server", style="yellow"
             )
@@ -68,12 +81,31 @@ def create_model_version_from_inference_template(
         console.print(f"Error creating model version: {e}", style="red")
         raise
 
-    return checkpoint_deploy_config
+    return DeploySuccessResult(
+        deploy_config=checkpoint_deploy_config,
+        truss_config=truss_config_result,
+        model_version=model_version,
+    )
+
+
+def _get_truss_config_from_result(result: dict) -> Optional[truss_config.TrussConfig]:
+    if result and result.get("truss_config"):
+        truss_config_dict = json.loads(result["truss_config"])
+        return truss_config.TrussConfig.from_dict(truss_config_dict)
+    # Although this should never happen, we defensively allow ourselves to return None
+    # because we need a failure to handle the truss config doesn't necessarily mean we failed to deploy
+    # the model version.
+    console.print(
+        "No truss config returned. Reach out to Baseten for support if this persists.",
+        style="red",
+    )
+    return None
 
 
 def _build_inference_template_request(
     checkpoint_deploy_config: DeployCheckpointsConfigComplete,
     remote_provider: BasetenRemote,
+    dry_run: bool,
 ) -> dict:
     """
     Build the GraphQL request data structure for createModelVersionFromInferenceTemplate mutation.
@@ -126,6 +158,7 @@ def _build_inference_template_request(
         "weights_sources": weights_sources,
         "inference_stack": inference_stack,
         "instance_type_id": instance_type_id,
+        "dry_run": dry_run,
     }
 
     return request_data

truss/cli/train/types.py

@@ -1,6 +1,9 @@
 from dataclasses import dataclass
 from typing import Optional
 
+from pydantic import BaseModel
+
+from truss.base import truss_config
 from truss_train.definitions import (
     CheckpointList,
     Compute,
@@ -11,6 +14,7 @@ from truss_train.definitions import (
 
 @dataclass
 class DeployCheckpointArgs:
+    dry_run: bool
     project_id: Optional[str]
     job_id: Optional[str]
     deploy_config_path: Optional[str]
@@ -26,3 +30,18 @@ class DeployCheckpointsConfigComplete(DeployCheckpointsConfig):
     model_name: str
     runtime: DeployCheckpointsRuntime
     compute: Compute
+
+
+class DeploySuccessModelVersion(BaseModel):
+    # allow extra fields to be forwards compatible with server
+    class Config:
+        extra = "allow"
+
+    name: str
+    id: str
+
+
+class DeploySuccessResult(BaseModel):
+    deploy_config: DeployCheckpointsConfigComplete
+    truss_config: Optional[truss_config.TrussConfig]
+    model_version: Optional[DeploySuccessModelVersion]

truss/cli/train_commands.py

@@ -1,5 +1,6 @@
 import os
 import sys
+from datetime import datetime
 from pathlib import Path
 from typing import Optional, cast
 
@@ -22,6 +23,7 @@ from truss.cli.train.core import (
     SORT_ORDER_ASC,
     SORT_ORDER_DESC,
 )
+from truss.cli.train.types import DeploySuccessResult
 from truss.cli.utils import common
 from truss.cli.utils.output import console, error_console
 from truss.remote.baseten.core import get_training_job_logs_with_pagination
@@ -306,6 +308,12 @@ def get_job_metrics(
 @click.option(
     "--dry-run", is_flag=True, help="Generate a truss config without deploying"
 )
+@click.option(
+    "--truss-config-output-dir",
+    type=str,
+    required=False,
+    help="Path to output the truss config to. If not provided, will output to truss_configs/<model_version_name>_<model_version_id> or truss_configs/dry_run_<timestamp> if dry run.",
+)
 @click.option("--remote", type=str, required=False, help="Remote to use")
 @common.common_options()
 def deploy_checkpoints(
@@ -315,6 +323,7 @@ def deploy_checkpoints(
     config: Optional[str],
     remote: Optional[str],
     dry_run: bool,
+    truss_config_output_dir: Optional[str],
 ):
     """
     Deploy a LoRA checkpoint via vLLM.
@@ -332,13 +341,46 @@ def deploy_checkpoints(
     result = train_cli.create_model_version_from_inference_template(
         remote_provider,
         train_cli.DeployCheckpointArgs(
-            project_id=project_id, job_id=job_id, deploy_config_path=config
+            project_id=project_id,
+            job_id=job_id,
+            deploy_config_path=config,
+            dry_run=dry_run,
         ),
     )
 
     if dry_run:
-        console.print("--dry-run flag provided, did not deploy", style="yellow")
-    train_cli.print_deploy_checkpoints_success_message(result)
+        console.print("did not deploy because --dry-run flag provided", style="yellow")
+
+    _write_truss_config(result, truss_config_output_dir, dry_run)
+
+    if not dry_run:
+        train_cli.print_deploy_checkpoints_success_message(result.deploy_config)
+
+
+def _write_truss_config(
+    result: DeploySuccessResult, truss_config_output_dir: Optional[str], dry_run: bool
+) -> None:
+    if not result.truss_config:
+        return
+    # format: 20251006_123456
+    datestamp = datetime.now().strftime("%Y%m%d_%H%M%S")
+    folder_name = (
+        f"{result.model_version.name}_{result.model_version.id}"
+        if result.model_version
+        else f"dry_run_{datestamp}"
+    )
+    output_dir_str = truss_config_output_dir or f"truss_configs/{folder_name}"
+    output_dir = Path(output_dir_str)
+    output_path = output_dir / "config.yaml"
+    os.makedirs(output_dir, exist_ok=True)
+    console.print(f"Writing truss config to {output_path}", style="yellow")
+    console.print(f"👀 Run `cat {output_path}` to view the truss config", style="green")
+    if dry_run:
+        console.print(
+            f"🚀 Run `cd {output_dir} && truss push --publish` to deploy the truss",
+            style="green",
+        )
+    result.truss_config.write_to_yaml_file(output_path)
 
 
 @train.command(name="download")

truss/remote/baseten/api.py

@@ -800,6 +800,7 @@ class BasetenApi:
                     id
                     name
                 }
+                truss_config
             }
         }
         """

truss/templates/control/control/application.py

@@ -38,15 +38,17 @@ class SanitizedExceptionMiddleware(BaseHTTPMiddleware):
         try:
             return await call_next(request)
         except Exception as exc:
-            sanitized_traceback = self._create_sanitized_traceback(exc)
-            request.app.state.logger.error(sanitized_traceback)
-            request.app.state.logger.error(f"Request path: {request.url.path}")
-
+            # NB(nikhil): Intentionally bypass error logging for ModelLoadFailed, since health checks
+            # are noisy. The underlying model logs for why the load failed will still be visible.
             if isinstance(exc, ModelLoadFailed):
                 return JSONResponse(
                     {"error": str(exc)}, status_code=http.HTTPStatus.BAD_GATEWAY.value
                 )
-            elif isinstance(exc, PatchApplicatonError):
+
+            sanitized_traceback = self._create_sanitized_traceback(exc)
+            request.app.state.logger.error(sanitized_traceback)
+
+            if isinstance(exc, PatchApplicatonError):
                 error_type = _camel_to_snake_case(type(exc).__name__)
                 return JSONResponse({"error": {"type": error_type, "msg": str(exc)}})
             else:

{truss-0.11.10rc3.dist-info → truss-0.11.10rc5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.10rc3
+Version: 0.11.10rc5
 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
@@ -37,7 +37,7 @@ Requires-Dist: rich<14,>=13.4.2
 Requires-Dist: ruff>=0.4.8
 Requires-Dist: tenacity>=8.0.1
 Requires-Dist: tomlkit>=0.13.2
-Requires-Dist: truss-transfer<0.0.35,>=0.0.30
+Requires-Dist: truss-transfer<0.0.36,>=0.0.32
 Requires-Dist: watchfiles<0.20,>=0.19.0
 Description-Content-Type: text/markdown
 

{truss-0.11.10rc3.dist-info → truss-0.11.10rc5.dist-info}/RECORD

@@ -5,26 +5,26 @@ truss/base/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 truss/base/constants.py,sha256=sExArdnuGg83z83XMgaQ4b8SS3V_j_bJEpOATDGJzpE,3600
 truss/base/custom_types.py,sha256=FUSIT2lPOQb6gfg6IzT63YBV8r8L6NIZ0D74Fp3e_jQ,2835
 truss/base/errors.py,sha256=zDVLEvseTChdPP0oNhBBQCtQUtZJUaof5zeWMIjqz6o,691
-truss/base/trt_llm_config.py,sha256=-hRpRsbxnfDaKS-5112yT0iP6R0evOtoTvnn557cwvc,32926
+truss/base/trt_llm_config.py,sha256=81ZZxRQF3o29HLCX6nlXtPwALejcdns6c4mbrExwASk,32958
 truss/base/truss_config.py,sha256=7CtiJIwMHtDU8Wzn8UTJUVVunD0pWFl4QUVycK2aIpY,28055
 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=Cfr9-TDE-esQI_R8az5OpLoQyz3Qv38mLsSNwy9znmI,18873
+truss/cli/train_commands.py,sha256=CrVqWsdkmSxgi3i2sSEyiE4QdfD0Z96F2Ib-PMZJjm8,20444
 truss/cli/logs/base_watcher.py,sha256=vuqteoaMVGX34cgKcETf4X_gOkvnSnDaWz1_pbeFhqs,3343
 truss/cli/logs/model_log_watcher.py,sha256=38vQCcNItfDrTKucvdJ10ZYLOcbGa5ZAKUqUnV4nH34,1971
 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=bJ40Jr4oXsG-blr3LGwd02KlYds6YbOS2x5twj_nyiI,26344
+truss/cli/train/core.py,sha256=Tiflb7aqonFl93QLCs7WO1nsnl3LjfiUHSfUcllBuT0,26496
 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
 truss/cli/train/poller.py,sha256=TGRzELxsicga0bEXewSX1ujw6lfPmDnHd6nr8zvOFO8,3550
-truss/cli/train/types.py,sha256=aGqZ5yOLv5WZL5_rmhl3s9NoXhjXOkeSsHD3QEH0en4,700
+truss/cli/train/types.py,sha256=0tfgInTm1Dmq7p8Gh_0RwxGViN6Tpp0YjsiYeMUNiPE,1163
 truss/cli/train/deploy_checkpoints/__init__.py,sha256=HuiDD6-qfwJ7fbRVX4s9Fxn6rmO6nwTLb0fVxwcMKls,137
-truss/cli/train/deploy_checkpoints/deploy_checkpoints.py,sha256=T3ss6qZE7iwbxXcHrNRCmt1e7VwOyVnWwwFZGv6222I,21398
+truss/cli/train/deploy_checkpoints/deploy_checkpoints.py,sha256=WSv2olpsnXCEcutqFszCKU36fMo3MZ-zIVM6ihVYF_w,22599
 truss/cli/train/deploy_checkpoints/deploy_checkpoints_helpers.py,sha256=r_IKMlqejMwIU6gsfxDIRzfmltfDf6Sz9-vnKrP_10s,83
 truss/cli/train/deploy_checkpoints/deploy_full_checkpoints.py,sha256=BEgn0p-LUE_enXJKrbHtuwEkLi3B1qcNJVzExB60Fzg,1199
 truss/cli/train/deploy_checkpoints/deploy_lora_checkpoints.py,sha256=xP403ny-7YGPZCkEFRqi-Awt3hYVteGMBnZNWyY_8Pc,1138
@@ -52,7 +52,7 @@ truss/patch/truss_dir_patch_applier.py,sha256=ALnaVnu96g0kF2UmGuBFTua3lrXpwAy4sG
 truss/remote/remote_factory.py,sha256=-0gLh_yIyNDgD48Q6sR8Yo5dOMQg84lrHRvn_XR0n4s,3585
 truss/remote/truss_remote.py,sha256=TEe6h6by5-JLy7PMFsDN2QxIY5FmdIYN3bKvHHl02xM,8440
 truss/remote/baseten/__init__.py,sha256=XNqJW1zyp143XQc6-7XVwsUA_Q_ZJv_ausn1_Ohtw9Y,176
-truss/remote/baseten/api.py,sha256=SJHy1JWD7-dlXJRuN9fxFsGzUk-Irtt_KOnBi7f7_4s,28235
+truss/remote/baseten/api.py,sha256=UxDjs0VOu_E_mDM6fa63HXE3TjTOjBR--aaHOwp11pQ,28264
 truss/remote/baseten/auth.py,sha256=tI7s6cI2EZgzpMIzrdbILHyGwiHDnmoKf_JBhJXT55E,776
 truss/remote/baseten/core.py,sha256=uxtmBI9RAVHu1glIEJb5Q4ccJYLeZM1Cp5Svb9W68Yw,21965
 truss/remote/baseten/custom_types.py,sha256=bYrfTzGgYr6FDoya0omyadCLSTcTc-83U2scQORyUj0,4715
@@ -73,7 +73,7 @@ truss/templates/copy_cache_files.Dockerfile.jinja,sha256=Os5zFdYLZ_AfCRGq4RcpVTO
 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=_xcJR-s1AEP_l6lQaTJ6JUNXLw_ZxMB13bmoOs82zZ0,5202
+truss/templates/control/control/application.py,sha256=5Kam6M-XtfKGaXQz8cc3d0bwDkB80o2MskABWROx1gk,5321
 truss/templates/control/control/endpoints.py,sha256=KzqsLVNJE6r6TCPW8D5FMCtsfHadTwR15A3z_viGxmM,11782
 truss/templates/control/control/server.py,sha256=R4Y219i1dcz0kkksN8obLoX-YXWGo9iW1igindyG50c,3128
 truss/templates/control/control/helpers/context_managers.py,sha256=W6dyFgLBhPa5meqrOb3w_phMtKfaJI-GhwUfpiycDc8,413
@@ -335,6 +335,7 @@ truss/truss_handle/patch/local_truss_patch_applier.py,sha256=fOHWKt3teYnbqeRsF63
 truss/truss_handle/patch/signature.py,sha256=8eas8gy6Japd1hrgdmtHmKTTxQmWsbmgKRQQGL2PVuA,858
 truss/truss_handle/patch/truss_dir_patch_applier.py,sha256=uhhHvKYHn_dpfz0xp4jwO9_qAej5sO3f8of_h-21PP4,3666
 truss/util/.truss_ignore,sha256=jpQA9ou-r_JEIcEHsUqGLHhir_m3d4IPGNyzKXtS-2g,3131
+truss/util/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 truss/util/docker.py,sha256=6PD7kMBBrOjsdvgkuSv7JMgZbe3NoJIeGasljMm2SwA,3934
 truss/util/download.py,sha256=1lfBwzyaNLEp7SAVrBd9BX5inZpkCVp8sBnS9RNoiJA,2521
 truss/util/env_vars.py,sha256=7Bv686eER71Barrs6fNamk_TrTJGmu9yV2TxaVmupn0,1232
@@ -369,8 +370,8 @@ 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-0.11.10rc3.dist-info/METADATA,sha256=UjBDuNvCsQnKDM5sKkeJL27F51Ldl-AdXNGhOK8usTk,6681
-truss-0.11.10rc3.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-truss-0.11.10rc3.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
-truss-0.11.10rc3.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
-truss-0.11.10rc3.dist-info/RECORD,,
+truss-0.11.10rc5.dist-info/METADATA,sha256=a_HXEXPbO3cg_SRpfjPUops_cJT4ArcED_amm8TRXvI,6681
+truss-0.11.10rc5.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+truss-0.11.10rc5.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
+truss-0.11.10rc5.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
+truss-0.11.10rc5.dist-info/RECORD,,