Reduce a bunch of dependencies from toolchain

Some improvements to the distribution install script

CONTRIBUTING.md

@@ -26,11 +26,14 @@ Meta has a [bounty program](http://facebook.com/whitehat/info) for the safe
 disclosure of security bugs. In those cases, please go through the process
 outlined on that page and do not file a public issue.
 
-## Coding Style  
+## Coding Style
 * 2 spaces for indentation rather than tabs
 * 80 character line length
 * ...
 
+## Tips
+* If you are developing with a llama-models repository checked out and need your distribution to reflect changes from there, set `LLAMA_MODELS_DIR` to that dir when running any of the `llama` CLI commands.
+
 ## License
 By contributing to Llama, you agree that your contributions will be licensed
 under the LICENSE file in the root directory of this source tree.

llama_toolchain/agentic_system/api/datatypes.py

@@ -8,8 +8,9 @@ from datetime import datetime
 from enum import Enum
 from typing import Any, Dict, List, Literal, Optional, Union
 
+from llama_models.schema_utils import json_schema_type
+
 from pydantic import BaseModel, Field
-from strong_typing.schema import json_schema_type
 from typing_extensions import Annotated
 
 from llama_toolchain.common.deployment_types import *  # noqa: F403

llama_toolchain/agentic_system/api/endpoints.py

@@ -8,9 +8,7 @@ from .datatypes import *  # noqa: F403
 from typing import Protocol
 
 # this dependency is annoying and we need a forked up version anyway
-from pyopenapi import webmethod
+from llama_models.schema_utils import json_schema_type, webmethod
-
-from strong_typing.schema import json_schema_type
 
 
 @json_schema_type

llama_toolchain/common/deployment_types.py

@@ -9,9 +9,9 @@ from typing import Dict, Optional
 
 from llama_models.llama3_1.api.datatypes import URL
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 
 @json_schema_type

llama_toolchain/common/training_types.py

@@ -5,8 +5,8 @@
 # the root directory of this source tree.
 
 from llama_models.llama3_1.api.datatypes import URL
+from llama_models.schema_utils import json_schema_type
 from pydantic import BaseModel
-from strong_typing.schema import json_schema_type
 
 
 @json_schema_type(schema={"description": "Checkpoint created during training runs"})

llama_toolchain/dataset/api/datatypes.py

@@ -9,9 +9,9 @@ from typing import Any, Dict, Optional
 
 from llama_models.llama3_1.api.datatypes import URL
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 
 @json_schema_type

llama_toolchain/dataset/api/endpoints.py

@@ -6,10 +6,9 @@
 
 from typing import Protocol
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type, webmethod
 
-from pyopenapi import webmethod
+from pydantic import BaseModel
-from strong_typing.schema import json_schema_type
 
 from .datatypes import *  # noqa: F403
 

llama_toolchain/distribution/datatypes.py

@@ -7,8 +7,9 @@
 from enum import Enum
 from typing import Any, Dict, List, Optional
 
+from llama_models.schema_utils import json_schema_type
+
 from pydantic import BaseModel, Field
-from strong_typing.schema import json_schema_type
 
 
 @json_schema_type

llama_toolchain/distribution/install_distribution.sh

@@ -50,20 +50,30 @@ ensure_conda_env_python310() {
     conda create -n "${env_name}" python="${python_version}" -y
   fi
 
-  # Install pip dependencies
-  if [ -n "$pip_dependencies" ]; then
-    echo "Installing pip dependencies: $pip_dependencies"
-    conda run -n "${env_name}" pip install $pip_dependencies
-  fi
-
   # Re-installing llama-toolchain in the new conda environment
-  if git rev-parse --is-inside-work-tree &> /dev/null; then
+  if git rev-parse --is-inside-work-tree &>/dev/null; then
     repo_root=$(git rev-parse --show-toplevel)
     cd "$repo_root"
     conda run -n "${env_name}" pip install -e .
   else
-    echo -e "${RED}Not inside a Git repository. Please re-run from within llama-toolchain repository.${NC}"
+    conda run -n "${env_name}" pip install llama-toolchain
-    exit 1
+  fi
+
+  if [ -n "$LLAMA_MODELS_DIR" ]; then
+    if [ ! -d "$LLAMA_MODELS_DIR" ]; then
+      echo -e "${RED}Warning: LLAMA_MODELS_DIR is set but directory does not exist: $LLAMA_MODELS_DIR${NC}" >&2
+      exit 1
+    fi
+
+    echo "Installing from LLAMA_MODELS_DIR: $LLAMA_MODELS_DIR"
+    conda run -n "${env_name}" pip uninstall -y llama-models
+    conda run -n "${env_name}" pip install -e "$LLAMA_MODELS_DIR"
+  fi
+
+  # Install pip dependencies
+  if [ -n "$pip_dependencies" ]; then
+    echo "Installing pip dependencies: $pip_dependencies"
+    conda run -n "${env_name}" pip install $pip_dependencies
   fi
 }
 
@@ -79,10 +89,11 @@ pip_dependencies="$3"
 
 ensure_conda_env_python310 "$env_name" "$pip_dependencies"
 
-echo -e "${GREEN}Successfully setup distribution environment. Starting to configure ....${NC}"
+echo -e "${GREEN}Successfully setup distribution environment. Configuring...${NC}"
 
 eval "$(conda shell.bash hook)"
 conda deactivate && conda activate "$env_name"
 
 python_interp=$(conda run -n "$env_name" which python)
+
 $python_interp -m llama_toolchain.cli.llama distribution configure --name "$distribution_name"

llama_toolchain/distribution/registry.py

@@ -10,32 +10,11 @@ from typing import List, Optional
 from .datatypes import Api, DistributionSpec, RemoteProviderSpec
 from .distribution import api_providers
 
-# This is currently duplicated from `requirements.txt` with a few minor changes
+# These are the dependencies needed by the distribution server.
-# dev-dependencies like "ufmt" etc. are nuked. A few specialized dependencies
+# `llama-toolchain` is automatically installed by the installation script.
-# are moved to the appropriate distribution.
 COMMON_DEPENDENCIES = [
-    "accelerate",
-    "black==24.4.2",
-    "blobfile",
-    "codeshield",
-    "fairscale",
     "fastapi",
-    "fire",
-    "flake8",
-    "httpx",
-    "huggingface-hub",
-    "json-strong-typing",
-    "llama-models",
-    "pandas",
-    "Pillow",
-    "pydantic==1.10.13",
-    "pydantic_core==2.18.2",
     "python-dotenv",
-    "python-openapi",
-    "requests",
-    "tiktoken",
-    "torch",
-    "transformers",
     "uvicorn",
 ]
 
@@ -59,10 +38,22 @@ def available_distribution_specs() -> List[DistributionSpec]:
         DistributionSpec(
             spec_id="inline",
             description="Use code from `llama_toolchain` itself to serve all llama stack APIs",
-            additional_pip_packages=COMMON_DEPENDENCIES
+            additional_pip_packages=(
-            + [
+                COMMON_DEPENDENCIES
-                "fbgemm-gpu==0.8.0",
+                # why do we need any of these? they should be completely covered
-            ],
+                # by the provider dependencies themselves
+                + [
+                    "accelerate",
+                    "blobfile",
+                    "codeshield",
+                    "fairscale",
+                    "pandas",
+                    "Pillow",
+                    "torch",
+                    "transformers",
+                    "fbgemm-gpu==0.8.0",
+                ]
+            ),
             provider_specs={
                 Api.inference: providers[Api.inference]["meta-reference"],
                 Api.safety: providers[Api.safety]["meta-reference"],
@@ -72,20 +63,7 @@ def available_distribution_specs() -> List[DistributionSpec]:
         DistributionSpec(
             spec_id="remote",
             description="Point to remote services for all llama stack APIs",
-            additional_pip_packages=[
+            additional_pip_packages=COMMON_DEPENDENCIES,
-                "python-dotenv",
-                "blobfile",
-                "fairscale",
-                "fastapi",
-                "fire",
-                "httpx",
-                "huggingface-hub",
-                "json-strong-typing",
-                "pydantic==1.10.13",
-                "pydantic_core==2.18.2",
-                "tiktoken",
-                "uvicorn",
-            ],
             provider_specs={x: remote_spec(x) for x in providers},
         ),
         DistributionSpec(

llama_toolchain/evaluations/api/endpoints.py

@@ -6,9 +6,9 @@
 
 from typing import List, Protocol
 
-from pydantic import BaseModel
+from llama_models.schema_utils import webmethod
 
-from pyopenapi import webmethod
+from pydantic import BaseModel
 
 from llama_models.llama3_1.api.datatypes import *  # noqa: F403
 from .datatypes import *  # noqa: F403

llama_toolchain/inference/api/datatypes.py

@@ -7,9 +7,9 @@
 from enum import Enum
 from typing import List, Literal, Optional, Union
 
-from pydantic import BaseModel, Field
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel, Field
 from typing_extensions import Annotated
 
 from llama_models.llama3_1.api.datatypes import *  # noqa: F403

llama_toolchain/inference/api/endpoints.py

@@ -8,7 +8,7 @@ from .datatypes import *  # noqa: F403
 from typing import Optional, Protocol
 
 # this dependency is annoying and we need a forked up version anyway
-from pyopenapi import webmethod
+from llama_models.schema_utils import webmethod
 
 
 @json_schema_type

llama_toolchain/inference/meta_reference/config.py

@@ -6,8 +6,9 @@
 
 from typing import Optional
 
+from llama_models.schema_utils import json_schema_type
+
 from pydantic import BaseModel
-from strong_typing.schema import json_schema_type
 
 from llama_toolchain.inference.api import QuantizationConfig
 

llama_toolchain/inference/meta_reference/generation.py

@@ -25,8 +25,8 @@ from fairscale.nn.model_parallel.initialize import (
 from llama_models.llama3_1.api.args import ModelArgs
 from llama_models.llama3_1.api.chat_format import ChatFormat, ModelInput
 from llama_models.llama3_1.api.datatypes import Message
-from llama_models.llama3_1.api.model import Transformer
 from llama_models.llama3_1.api.tokenizer import Tokenizer
+from llama_models.llama3_1.reference_impl.model import Transformer
 from llama_models.sku_list import resolve_model
 from termcolor import cprint
 

llama_toolchain/inference/ollama/config.py

@@ -4,8 +4,8 @@
 # This source code is licensed under the terms described in the LICENSE file in
 # the root directory of this source tree.
 
+from llama_models.schema_utils import json_schema_type
 from pydantic import BaseModel, Field
-from strong_typing.schema import json_schema_type
 
 
 @json_schema_type

llama_toolchain/memory/api/datatypes.py

@@ -6,9 +6,9 @@
 
 from typing import Any, Dict
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 
 @json_schema_type

llama_toolchain/memory/api/endpoints.py

@@ -6,7 +6,7 @@
 
 from typing import List, Protocol
 
-from pyopenapi import webmethod
+from llama_models.schema_utils import webmethod
 
 from .datatypes import *  # noqa: F403
 

llama_toolchain/models/api/endpoints.py

@@ -6,9 +6,9 @@
 
 from typing import Protocol
 
-from pydantic import BaseModel  # noqa: F401
+from llama_models.schema_utils import webmethod  # noqa: F401
 
-from pyopenapi import webmethod  # noqa: F401
+from pydantic import BaseModel  # noqa: F401
 
 
 class Models(Protocol): ...

llama_toolchain/post_training/api/datatypes.py

@@ -7,9 +7,9 @@
 from enum import Enum
 from typing import List
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 
 class OptimizerType(Enum):

llama_toolchain/post_training/api/endpoints.py

@@ -8,10 +8,9 @@ from datetime import datetime
 
 from typing import Any, Dict, List, Optional, Protocol
 
-from pydantic import BaseModel, Field
+from llama_models.schema_utils import json_schema_type, webmethod
 
-from pyopenapi import webmethod
+from pydantic import BaseModel, Field
-from strong_typing.schema import json_schema_type
 
 from llama_models.llama3_1.api.datatypes import *  # noqa: F403
 from llama_toolchain.dataset.api.datatypes import *  # noqa: F403

llama_toolchain/reward_scoring/api/datatypes.py

@@ -6,9 +6,9 @@
 
 from typing import List
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 from llama_models.llama3_1.api.datatypes import *  # noqa: F403
 

llama_toolchain/reward_scoring/api/endpoints.py

@@ -7,7 +7,7 @@
 from typing import List, Protocol, Union
 from .datatypes import *  # noqa: F403
 
-from pyopenapi import webmethod
+from llama_models.schema_utils import webmethod
 
 
 @json_schema_type

llama_toolchain/safety/api/datatypes.py

@@ -9,9 +9,9 @@ from typing import Dict, Optional, Union
 
 from llama_models.llama3_1.api.datatypes import ToolParamDefinition
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type
 
-from strong_typing.schema import json_schema_type
+from pydantic import BaseModel
 
 from llama_toolchain.common.deployment_types import RestAPIExecutionConfig
 

llama_toolchain/safety/api/endpoints.py

@@ -10,7 +10,7 @@ from typing import List, Protocol
 from llama_models.llama3_1.api.datatypes import Message
 
 # this dependency is annoying and we need a forked up version anyway
-from pyopenapi import webmethod
+from llama_models.schema_utils import webmethod
 
 
 @json_schema_type

llama_toolchain/synthetic_data_generation/api/endpoints.py

@@ -6,10 +6,9 @@
 
 from typing import Any, Dict, List, Optional, Protocol
 
-from pydantic import BaseModel
+from llama_models.schema_utils import json_schema_type, webmethod
 
-from pyopenapi import webmethod
+from pydantic import BaseModel
-from strong_typing.schema import json_schema_type
 
 from llama_models.llama3_1.api.datatypes import *  # noqa: F403
 from llama_toolchain.reward_scoring.api.datatypes import *  # noqa: F403

requirements.txt

@@ -1,18 +1,7 @@
-black==24.4.2
-fastapi
-fbgemm-gpu==0.8.0
 fire
-flake8
 httpx
 huggingface-hub
-json-strong-typing
 llama-models
-omegaconf
-pre-commit
 pydantic==1.10.13
 pydantic_core==2.18.2
-python-openapi
 requests
-ufmt==2.7.0
-usort==1.0.8
-uvicorn