Merge pull request #874 from basetenlabs/bump-version-0.9.6

Release 0.9.6
basetenlabs · Mar 21, 2024 · b225023 · b225023
2 parents 90af43f + 93d1f97
commit b225023
Show file tree

Hide file tree

Showing 7 changed files with 467 additions and 466 deletions.
diff --git a/poetry.lock b/poetry.lock
diff --git a/pyproject.toml b/pyproject.toml
@@ -1,13 +1,19 @@
 [tool.poetry]
 name = "truss"
-version = "0.9.5"
+version = "0.9.6"
 description = "A seamless bridge from model development to model delivery"
 license = "MIT"
 readme = "README.md"
 authors = ["Pankaj Gupta <pankaj@baseten.co>", "Phil Howes <phil@baseten.co>"]
 include = ["*.txt", "*.Dockerfile", "*.md"]
 repository = "https://github.com/basetenlabs/truss"
-keywords = ["MLOps", "AI", "Model Serving", "Model Deployment", "Machine Learning"]
+keywords = [
+    "MLOps",
+    "AI",
+    "Model Serving",
+    "Model Deployment",
+    "Machine Learning",
+]
 
 [tool.poetry.urls]
 "Homepage" = "https://truss.baseten.co"
@@ -16,90 +22,90 @@ keywords = ["MLOps", "AI", "Model Serving", "Model Deployment", "Machine Learnin
 "Baseten" = "https://baseten.co"
 
 [tool.poetry.dependencies]
-python = ">=3.8,<3.12"
-numpy = ">=1.23.5"
-msgpack = ">=1.0.2"
-msgpack-numpy = ">=0.4.7.1"
-packaging = ">=20.9"
-python-json-logger = ">=2.0.2"
-PyYAML = "^6.0"
-Jinja2 = "^3.1.2"
-python-on-whales = "^0.68.0"
-tenacity = "^8.0.1"
-single-source = "^0.3.0"
 blake3 = "^0.3.3"
+boto3 = "^1.26.157"
 fastapi = "^0.109.1"
-uvicorn = "^0.24.0"
+google-cloud-storage = "2.10.0"
 httpx = "^0.24.1"
-psutil = "^5.9.4"
-pydantic = "^2.6.1"
-boto3 = "^1.26.157"
-rich = "^13.4.2"
-watchfiles = "^0.19.0"
 huggingface_hub = ">=0.19.4"
-rich-click = "^1.6.1"
 inquirerpy = "^0.3.4"
-google-cloud-storage = "2.10.0"
+Jinja2 = "^3.1.2"
 loguru = ">=0.7.2"
-uvloop = "^0.19.0"
+msgpack = ">=1.0.2"
+msgpack-numpy = ">=0.4.7.1"
+numpy = ">=1.23.5"
+packaging = ">=20.9"
 pathspec = ">=0.9.0"
-
-[tool.poetry.group.builder.dependencies]
+psutil = "^5.9.4"
+pydantic = ">=1.10.0"
 python = ">=3.8,<3.12"
-packaging = ">=20.9"
 python-json-logger = ">=2.0.2"
+python-on-whales = "^0.68.0"
 PyYAML = "^6.0"
-Jinja2 = "^3.1.2"
-tenacity = "^8.0.1"
+rich = "^13.4.2"
+rich-click = "^1.6.1"
 single-source = "^0.3.0"
-click = "^8.0.3"
-requests = "^2.28.1"
+tenacity = "^8.0.1"
+uvicorn = "^0.24.0"
+uvloop = "^0.19.0"
+watchfiles = "^0.19.0"
+
+[tool.poetry.group.builder.dependencies]
 blake3 = "^0.3.3"
+boto3 = "^1.26.157"
+click = "^8.0.3"
 fastapi = "^0.109.1"
-uvicorn = "^0.24.0"
+google-cloud-storage = "2.10.0"
 httpx = "^0.24.1"
-psutil = "^5.9.4"
 huggingface_hub = ">=0.19.4"
-google-cloud-storage = "2.10.0"
-boto3 = "^1.26.157"
+Jinja2 = "^3.1.2"
 loguru = ">=0.7.2"
+packaging = ">=20.9"
 pathspec = ">=0.9.0"
+psutil = "^5.9.4"
+python = ">=3.8,<3.12"
+python-json-logger = ">=2.0.2"
+PyYAML = "^6.0"
+requests = "^2.28.1"
+single-source = "^0.3.0"
+tenacity = "^8.0.1"
+uvicorn = "^0.24.0"
 
 [tool.poetry.dev-dependencies]
-ipython = "^7.16"
-pytest = "7.2.0"
-pre-commit = "^2.18.1"
-isort = "^5.10.1"
+black = "^22.6.0"
+coverage = "^6.4.1"
+dockerfile = "^3.2.0"
 flake8 = "^4.0.1"
 ipdb = "^0.13.9"
-coverage = "^6.4.1"
-pytest-cov = "^3.0.0"
-black = "^22.6.0"
-nbconvert = "^7.2.1"
 ipykernel = "^6.16.0"
-dockerfile = "^3.2.0"
+ipython = "^7.16"
+isort = "^5.10.1"
+nbconvert = "^7.2.1"
+pre-commit = "^2.18.1"
+pytest = "7.2.0"
+pytest-cov = "^3.0.0"
 types-PyYAML = "^6.0.12.12"
 types-setuptools = "^69.0.0.0"
 
 [tool.poetry.scripts]
 truss = 'truss.cli:truss_cli'
 
 [tool.poetry.group.dev.dependencies]
+flask = "^2.3.3"
+httpx = { extras = ["cli"], version = "^0.24.1" }
 mypy = "^1.0.0"
 pytest-split = "^0.8.1"
-httpx = {extras = ["cli"], version = "^0.24.1"}
 requests-mock = "^1.11.0"
-flask = "^2.3.3"
 types-requests = "2.31.0.2"
 
 [build-system]
-requires = ["poetry-core>=1.2.1"]
 build-backend = "poetry.core.masonry.api"
+requires = ["poetry-core>=1.2.1"]
 
 [tool.isort]
 profile = "black"
 src_paths = ["isort", "test"]
 
 [tool.mypy]
-python_version = 3.8
 ignore_missing_imports = true
+python_version = "3.8"
diff --git a/truss/templates/cache.Dockerfile.jinja b/truss/templates/cache.Dockerfile.jinja
@@ -1,5 +1,3 @@
-FROM python:3.11-slim as cache_warmer
-
 RUN mkdir -p /app/model_cache
 WORKDIR /app
 

diff --git a/truss/templates/copy_cache_files.Dockerfile.jinja b/truss/templates/copy_cache_files.Dockerfile.jinja
diff --git a/truss/templates/server.Dockerfile.jinja b/truss/templates/server.Dockerfile.jinja
@@ -1,7 +1,3 @@
-{%- if model_cache %}
-{%- include "cache.Dockerfile.jinja" %}
-{%- endif %}
-
 {% extends "base.Dockerfile.jinja" %}
 
 {% block base_image_patch %}
@@ -51,10 +47,9 @@ RUN pip install -r {{server_requirements_filename}} --no-cache-dir && rm -rf /ro
 
 
 {% block app_copy %}
-{%- if model_cache %}
-# Copy data before code for better caching
-    {%- include "copy_cache_files.Dockerfile.jinja"%}
-{%- endif %}
+    {%- if model_cache %}
+    {%- include "cache.Dockerfile.jinja" %}
+    {%- endif %}
 
 {%- if external_data_files %}
 {% for url, dst in external_data_files %}

diff --git a/truss/templates/trtllm/packages/schema.py b/truss/templates/trtllm/packages/schema.py
@@ -142,7 +142,10 @@ def cache_path(self) -> Path:
             return Path(f"sq_{self.sq_alpha}")
 
 
-class EngineBuildArgs(BaseModel, use_enum_values=True):
+class EngineBuildArgs(BaseModel):
+    class Config:
+        use_enum_values = True
+
     repo: Optional[str] = None
     args: Optional[ArgsConfig] = None
     quant: Optional[Quant] = None

diff --git a/truss/truss_config.py b/truss/truss_config.py
@@ -582,7 +582,7 @@ def obj_to_dict(obj, verbose: bool = False):
                 )
             elif isinstance(field_curr_value, TRTLLMConfiguration):
                 d["trt_llm"] = transform_optional(
-                    field_curr_value, lambda data: data.model_dump(mode="json")
+                    field_curr_value, lambda data: data.dict()
                 )
             else:
                 d[field_name] = field_curr_value