truss 0.11.13rc3__py3-none-any.whl → 0.11.13rc500__py3-none-any.whl

truss/base/trt_llm_config.py

@@ -68,6 +68,7 @@ class TrussTRTLLMQuantizationType(str, Enum):
     FP8_KV = "fp8_kv"
     FP4 = "fp4"
     FP4_KV = "fp4_kv"
+    FP4_MLP_ONLY = "fp4_mlp_only"
 
 
 class TrussTRTLLMPluginConfiguration(PydanticTrTBaseModel):
@@ -713,7 +714,9 @@ def trt_llm_common_validation(config: "TrussConfig"):
             "accelerators or newer (CUDA_COMPUTE>=89)"
         )
     elif trt_llm_config.build.quantization_type in [
-        TrussTRTLLMQuantizationType.FP4
+        TrussTRTLLMQuantizationType.FP4,
+        TrussTRTLLMQuantizationType.FP4_KV,
+        TrussTRTLLMQuantizationType.FP4_MLP_ONLY,
     ] and config.resources.accelerator.accelerator in [
         truss_config.Accelerator.H100,
         truss_config.Accelerator.L4,

truss/templates/base.Dockerfile.jinja

@@ -115,9 +115,11 @@ WORKDIR $APP_HOME
 {% endblock %}
 
 
+{% set packages_dir = "/packages" %}
+RUN mkdir -p {{ packages_dir }}
 {% block bundled_packages_copy %}
     {%- if bundled_packages_dir_exists %}
-COPY --chown={{ default_owner }} ./{{ config.bundled_packages_dir }} /packages
+COPY --chown={{ default_owner }} ./{{ config.bundled_packages_dir }} {{ packages_dir }}
 {%- endif %}
 {% endblock %}
 

truss/templates/server.Dockerfile.jinja

@@ -69,7 +69,7 @@ COPY --chown={{ default_owner }} ./{{ config.data_dir }} ${APP_HOME}/data
 
 {%- if model_cache_v2 %}
 {# v0.0.9, keep synced with server_requirements.txt #}
-RUN curl -sSL --fail --retry 5 --retry-delay 2 -o /usr/local/bin/truss-transfer-cli https://github.com/basetenlabs/truss/releases/download/v0.11.12rc4/truss-transfer-cli-v0.11.12rc4-linux-x86_64-unknown-linux-musl
+RUN curl -sSL --fail --retry 5 --retry-delay 2 -o /usr/local/bin/truss-transfer-cli https://github.com/basetenlabs/truss/releases/download/v0.11.13rc3/truss-transfer-cli-v0.11.13rc3-linux-x86_64-unknown-linux-musl
 RUN chmod +x /usr/local/bin/truss-transfer-cli
 RUN mkdir /static-bptr
 RUN echo "hash {{model_cache_hash}}"
@@ -104,7 +104,7 @@ COPY --chown={{ default_owner }} ./{{ config.model_module_dir }} ${APP_HOME}/mod
 {# Macro to change ownership of directories and switch to regular user #}
 {%- macro chown_and_switch_to_regular_user_if_enabled(additional_chown_dirs=[]) -%}
 {%- if non_root_user %}
-RUN chown -R {{ app_username }}:{{ app_username }} {% for dir in additional_chown_dirs %}{{ dir }} {% endfor %}${HOME} ${APP_HOME}
+RUN chown -R {{ app_username }}:{{ app_username }} ${HOME} ${APP_HOME} {{ packages_dir }} {% for dir in additional_chown_dirs %}{{ dir }} {% endfor %}
 USER {{ app_username }}
 {%- endif %} {#- endif non_root_user #}
 {%- endmacro -%}

truss/templates/shared/lazy_data_resolver.py

@@ -1,13 +1,12 @@
 import atexit
 import json
 import logging
-import os
 import time
 from dataclasses import dataclass
 from functools import lru_cache
 from pathlib import Path
 from threading import Lock, Thread
-from typing import Optional, Union
+from typing import List, Optional, Union
 
 try:
     from prometheus_client import Counter, Gauge, Histogram
@@ -31,7 +30,7 @@ class TrussTransferStats:
     total_manifest_size_bytes: int
     total_download_time_secs: float
     total_aggregated_mb_s: Optional[float]
-    file_downloads: list[FileDownloadMetric]
+    file_downloads: List[FileDownloadMetric]
     b10fs_read_speed_mbps: Optional[float]
     b10fs_decision_to_use: bool
     b10fs_enabled: bool
@@ -70,26 +69,35 @@ class TrussTransferStats:
         except Exception:
             return None
 
-    def publish_to_prometheus(self):
+    def publish_to_prometheus(self, hidden_time: float = 0.0):
         """Publish transfer stats to Prometheus metrics. Only runs once."""
         if not PROMETHEUS_AVAILABLE:
             return
         global METRICS_REGISTERED
-        if not METRICS_REGISTERED:
+
+        if METRICS_REGISTERED:
+            logging.info(
+                "Model cache metrics already registered, skipping."
+            )  # this should never happen
+            return
+        else:
             # Ensure metrics are only registered once
             METRICS_REGISTERED = True
 
-        # Define metrics with model_cache label
+        # Define metrics with model_cache prefix
         manifest_size_gauge = Gauge(
             "model_cache_manifest_size_bytes", "Total manifest size in bytes"
         )
+        # histograms have intentially wide buckets to capture a variety of download times
         download_time_histogram = Histogram(
             "model_cache_download_time_seconds",
             "Total download time in seconds",
-            buckets=[
+            buckets=[0]
+            + [
                 2**i
                 for i in range(-3, 11)  # = [0.125, .. 2048] seconds
-            ],
+            ]
+            + [float("inf")],
         )
         download_speed_gauge = Gauge(
             "model_cache_download_speed_mbps", "Aggregated download speed in MB/s"
@@ -103,21 +111,29 @@ class TrussTransferStats:
             "model_cache_file_size_bytes_total",
             "Total size of downloaded files in bytes",
         )
+        file_download_hidden_time_gauge = Gauge(
+            "model_cache_file_download_hidden_time_seconds",
+            "Total time hidden from user by starting the import before user code (seconds)",
+        )
         file_download_time_histogram = Histogram(
             "model_cache_file_download_time_seconds",
             "File download time distribution",
-            buckets=[
+            buckets=[0]
+            + [
                 2**i
                 for i in range(-3, 11)  # = [0.125, .. 2048] seconds
-            ],
+            ]
+            + [float("inf")],
         )
         file_download_speed_histogram = Histogram(
             "model_cache_file_download_speed_mbps",
             "File download speed distribution",
-            buckets=[
+            buckets=[0]
+            + [
                 2**i
                 for i in range(-1, 12)  # = [0.5, .. 4096] MB/s
-            ],
+            ]
+            + [float("inf")],
         )
 
         # B10FS specific metrics
@@ -153,6 +169,7 @@ class TrussTransferStats:
         # Set main transfer metrics
         manifest_size_gauge.set(self.total_manifest_size_bytes)
         download_time_histogram.observe(self.total_download_time_secs)
+        file_download_hidden_time_gauge.set(hidden_time)
 
         if self.total_aggregated_mb_s is not None:
             download_speed_gauge.set(self.total_aggregated_mb_s)
@@ -331,10 +348,7 @@ class LazyDataResolverV2:
                 if stats and publish_stats:
                     self.logger.info(f"model_cache: {stats}")
                     # Publish stats to Prometheus
-                    if (
-                        os.getenv("TRUSS_MODEL_CACHE_PROMETHEUS", "0") == "1"
-                    ):  # Hide behind feature flag for core-product to enabled.
-                        stats.publish_to_prometheus()
+                    stats.publish_to_prometheus()
                 self.logger.info(
                     f"model_cache: Fetch took {fetch_t:.2f} seconds, of which {start_lock_t:.2f} seconds were spent blocking."
                 )

truss/tests/test_data/server.Dockerfile

@@ -36,6 +36,7 @@ COPY --chown= ./data ${APP_HOME}/data
 COPY --chown= ./server ${APP_HOME}
 COPY --chown= ./config.yaml ${APP_HOME}/config.yaml
 COPY --chown= ./model ${APP_HOME}/model
+RUN mkdir -p /packages
 COPY --chown= ./packages /packages
 ENV INFERENCE_SERVER_PORT="8080"
 ENV SERVER_START_CMD="/usr/local/bin/python3 /app/main.py"

{truss-0.11.13rc3.dist-info → truss-0.11.13rc500.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: truss
-Version: 0.11.13rc3
+Version: 0.11.13rc500
 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<0.0.40,>=0.0.36
+Requires-Dist: truss-transfer<0.0.40,>=0.0.37
 Requires-Dist: watchfiles<0.20,>=0.19.0
 Description-Content-Type: text/markdown
 

{truss-0.11.13rc3.dist-info → truss-0.11.13rc500.dist-info}/RECORD

@@ -5,7 +5,7 @@ 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=81ZZxRQF3o29HLCX6nlXtPwALejcdns6c4mbrExwASk,32958
+truss/base/trt_llm_config.py,sha256=rEtBVFg2QnNMxnaz11s5Z69dJB1w7Bpt48Wf6jSsVZI,33087
 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
@@ -66,12 +66,12 @@ 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=DgvNqmkt5QRrqy7a9hw6HffmVTUrdGl06oVZ4eeha5Y,5253
+truss/templates/base.Dockerfile.jinja,sha256=tdMmK5TeiQuYbz4gqbACM3R-l-mazqL9tAZtJ4sxC4g,5331
 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=BQpo2Mt_fBrdin1qD8HBKBo2N3Yr2lXrvV_a7J5WSzE,7071
+truss/templates/server.Dockerfile.jinja,sha256=Mu5_ZxuAknwaEOsF0l-XssA9pDg3pD3eLl6JBzNJ4rg,7091
 truss/templates/control/requirements.txt,sha256=tJGr83WoE0CZm2FrloZ9VScK84q-_FTuVXjDYrexhW0,250
 truss/templates/control/control/application.py,sha256=5Kam6M-XtfKGaXQz8cc3d0bwDkB80o2MskABWROx1gk,5321
 truss/templates/control/control/endpoints.py,sha256=KzqsLVNJE6r6TCPW8D5FMCtsfHadTwR15A3z_viGxmM,11782
@@ -107,7 +107,7 @@ truss/templates/server/common/tracing.py,sha256=XSTXNoRtV8vXwveJoX3H32go0JKnLmzn
 truss/templates/server/common/patches/whisper/patch.py,sha256=kDECQ-wmEpeAZFhUTQP457ofueeMsm7DgNy9tqinhJQ,2383
 truss/templates/shared/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 truss/templates/shared/dynamic_config_resolver.py,sha256=75s42NFhQI5jL7BqlJH_UkuQS7ptbtFh13f2nh6X5Wo,920
-truss/templates/shared/lazy_data_resolver.py,sha256=czfggu9DZ_qDnE2MxOdE2R8aZyJe2G1Cd-PL0AUGx-I,13561
+truss/templates/shared/lazy_data_resolver.py,sha256=HxrZz6X30j2LbsExYSqhuOGoYEffGpd7FPBtJexI7TQ,14064
 truss/templates/shared/log_config.py,sha256=l9udyu4VKHZePlfK9LQEd5TOUUodPuehypsXRSUL4Ac,5411
 truss/templates/shared/secrets_resolver.py,sha256=3prDe3Q06NTmUEe7KCW-W4TD1CzGck9lpDG789209z4,2110
 truss/templates/shared/serialization.py,sha256=_WC_2PPkRi-MdTwxwjG8LKQptnHi4sANfpOlKWevqWc,3736
@@ -185,7 +185,7 @@ 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=3rWiU3RxBh0PIh8pS23FnP8UA-ErmzxTlgf_J22mMQ8,2024
+truss/tests/test_data/server.Dockerfile,sha256=KoQN5qBpiXL93qbjbG76kfRVapwfV5CNJzQcpjPocz0,2047
 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
@@ -370,8 +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-0.11.13rc3.dist-info/METADATA,sha256=cPpD-bEoveXxM_dTFQAJPui0DFO940QYrTTWlO7aivc,6681
-truss-0.11.13rc3.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-truss-0.11.13rc3.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
-truss-0.11.13rc3.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
-truss-0.11.13rc3.dist-info/RECORD,,
+truss-0.11.13rc500.dist-info/METADATA,sha256=fO90Qd0TBo1ZirChnji7aZh2vHpqYmuUGsCYMZGe4Oo,6683
+truss-0.11.13rc500.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+truss-0.11.13rc500.dist-info/entry_points.txt,sha256=-MwKfHHQHQ6j0HqIgvxrz3CehCmczDLTD-OsRHnjjuU,130
+truss-0.11.13rc500.dist-info/licenses/LICENSE,sha256=FTqGzu85i-uw1Gi8E_o0oD60bH9yQ_XIGtZbA1QUYiw,1064
+truss-0.11.13rc500.dist-info/RECORD,,