Introduce Llama stack distributions (#22)

* Add distribution CLI scaffolding

* More progress towards `llama distribution install`

* getting closer to a distro definition, distro install + configure works

* Distribution server now functioning

* read existing configuration, save enums properly

* Remove inference uvicorn server entrypoint and llama inference CLI command

* updated dependency and client model name

* Improved exception handling

* local imports for faster cli

* undo a typo, add a passthrough distribution

* implement full-passthrough in the server

* add safety adapters, configuration handling, server + clients

* cleanup, moving stuff to common, nuke utils

* Add a Path() wrapper at the earliest place

* fixes

* Bring agentic system api to toolchain

Add adapter dependencies and resolve adapters using a topological sort

* refactor to reduce size of `agentic_system`

* move straggler files and fix some important existing bugs

* ApiSurface -> Api

* refactor a method out

* Adapter -> Provider

* Make each inference provider into its own subdirectory

* installation fixes

* Rename Distribution -> DistributionSpec, simplify RemoteProviders

* dict key instead of attr

* update inference config to take model and not model_dir

* Fix passthrough streaming, send headers properly not part of body :facepalm

* update safety to use model sku ids and not model dirs

* Update cli_reference.md

* minor fixes

* add DistributionConfig, fix a bug in model download

* Make install + start scripts do proper configuration automatically

* Update CLI_reference

* Nuke fp8_requirements, fold fbgemm into common requirements

* Update README, add newline between API surface configurations

* Refactor download functionality out of the Command so can be reused

* Add `llama model download` alias for `llama download`

* Show message about checksum file so users can check themselves

* Simpler intro statements

* get ollama working

* Reduce a bunch of dependencies from toolchain

Some improvements to the distribution install script

* Avoid using `conda run` since it buffers everything

* update dependencies and rely on LLAMA_TOOLCHAIN_DIR for dev purposes

* add validation for configuration input

* resort imports

* make optional subclasses default to yes for configuration

* Remove additional_pip_packages; move deps to providers

* for inline make 8b model the default

* Add scripts to MANIFEST

* allow installing from test.pypi.org

* Fix #2 to help with testing packages

* Must install llama-models at that same version first

* fix PIP_ARGS

---------

Co-authored-by: Hardik Shah <hjshah@fb.com>
Co-authored-by: Hardik Shah <hjshah@meta.com>

llama_toolchain/cli/distribution/__init__.py

@@ -0,0 +1,7 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+from .distribution import DistributionParser  # noqa

llama_toolchain/cli/distribution/configure.py

@@ -0,0 +1,106 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+import json
+import shlex
+
+import yaml
+
+from llama_toolchain.cli.subcommand import Subcommand
+from llama_toolchain.common.config_dirs import DISTRIBS_BASE_DIR
+from termcolor import cprint
+
+
+class DistributionConfigure(Subcommand):
+    """Llama cli for configuring llama toolchain configs"""
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "configure",
+            prog="llama distribution configure",
+            description="configure a llama stack distribution",
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_distribution_configure_cmd)
+
+    def _add_arguments(self):
+        self.parser.add_argument(
+            "--name",
+            type=str,
+            help="Name of the distribution to configure",
+            required=True,
+        )
+
+    def _run_distribution_configure_cmd(self, args: argparse.Namespace) -> None:
+        from llama_toolchain.distribution.datatypes import DistributionConfig
+        from llama_toolchain.distribution.registry import resolve_distribution_spec
+
+        config_file = DISTRIBS_BASE_DIR / args.name / "config.yaml"
+        if not config_file.exists():
+            self.parser.error(
+                f"Could not find {config_file}. Please run `llama distribution install` first"
+            )
+            return
+
+        # we need to find the spec from the name
+        with open(config_file, "r") as f:
+            config = DistributionConfig(**yaml.safe_load(f))
+
+        dist = resolve_distribution_spec(config.spec)
+        if dist is None:
+            raise ValueError(f"Could not find any registered spec `{config.spec}`")
+
+        configure_llama_distribution(dist, config)
+
+
+def configure_llama_distribution(dist: "Distribution", config: "DistributionConfig"):
+    from llama_toolchain.common.exec import run_command
+    from llama_toolchain.common.prompt_for_config import prompt_for_config
+    from llama_toolchain.common.serialize import EnumEncoder
+    from llama_toolchain.distribution.dynamic import instantiate_class_type
+
+    python_exe = run_command(shlex.split("which python"))
+    # simple check
+    conda_env = config.conda_env
+    if conda_env not in python_exe:
+        raise ValueError(
+            f"Please re-run configure by activating the `{conda_env}` conda environment"
+        )
+
+    if config.providers:
+        cprint(
+            f"Configuration already exists for {config.name}. Will overwrite...",
+            "yellow",
+            attrs=["bold"],
+        )
+
+    for api, provider_spec in dist.provider_specs.items():
+        cprint(f"Configuring API surface: {api.value}", "white", attrs=["bold"])
+        config_type = instantiate_class_type(provider_spec.config_class)
+        provider_config = prompt_for_config(
+            config_type,
+            (
+                config_type(**config.providers[api.value])
+                if api.value in config.providers
+                else None
+            ),
+        )
+        print("")
+
+        config.providers[api.value] = {
+            "provider_id": provider_spec.provider_id,
+            **provider_config.dict(),
+        }
+
+    config_path = DISTRIBS_BASE_DIR / config.name / "config.yaml"
+    with open(config_path, "w") as fp:
+        dist_config = json.loads(json.dumps(config.dict(), cls=EnumEncoder))
+        fp.write(yaml.dump(dist_config, sort_keys=False))
+
+    print(f"YAML configuration has been written to {config_path}")

llama_toolchain/cli/distribution/create.py

@@ -0,0 +1,44 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+
+from llama_toolchain.cli.subcommand import Subcommand
+
+
+class DistributionCreate(Subcommand):
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "create",
+            prog="llama distribution create",
+            description="create a Llama stack distribution",
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_distribution_create_cmd)
+
+    def _add_arguments(self):
+        self.parser.add_argument(
+            "--name",
+            type=str,
+            help="Name of the distribution to create",
+            required=True,
+        )
+        # for each Api the user wants to support, we should
+        # get the list of available providers, ask which one the user
+        # wants to pick and then ask for their configuration.
+
+    def _run_distribution_create_cmd(self, args: argparse.Namespace) -> None:
+        from llama_toolchain.distribution.registry import resolve_distribution_spec
+
+        dist = resolve_distribution_spec(args.name)
+        if dist is not None:
+            self.parser.error(f"Distribution with name {args.name} already exists")
+            return
+
+        raise NotImplementedError()

llama_toolchain/cli/distribution/distribution.py

@@ -0,0 +1,35 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+
+from llama_toolchain.cli.subcommand import Subcommand
+
+from .configure import DistributionConfigure
+from .create import DistributionCreate
+from .install import DistributionInstall
+from .list import DistributionList
+from .start import DistributionStart
+
+
+class DistributionParser(Subcommand):
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "distribution",
+            prog="llama distribution",
+            description="Operate on llama stack distributions",
+        )
+
+        subparsers = self.parser.add_subparsers(title="distribution_subcommands")
+
+        # Add sub-commands
+        DistributionList.create(subparsers)
+        DistributionInstall.create(subparsers)
+        DistributionCreate.create(subparsers)
+        DistributionConfigure.create(subparsers)
+        DistributionStart.create(subparsers)

llama_toolchain/cli/distribution/install.py

@@ -0,0 +1,111 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+import os
+
+import pkg_resources
+import yaml
+
+from termcolor import cprint
+
+from llama_toolchain.cli.subcommand import Subcommand
+from llama_toolchain.common.config_dirs import DISTRIBS_BASE_DIR
+
+
+class DistributionInstall(Subcommand):
+    """Llama cli for configuring llama toolchain configs"""
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "install",
+            prog="llama distribution install",
+            description="Install a llama stack distribution",
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_distribution_install_cmd)
+
+    def _add_arguments(self):
+        from llama_toolchain.distribution.registry import available_distribution_specs
+
+        self.parser.add_argument(
+            "--spec",
+            type=str,
+            help="Distribution spec to install (try ollama-inline)",
+            required=True,
+            choices=[d.spec_id for d in available_distribution_specs()],
+        )
+        self.parser.add_argument(
+            "--name",
+            type=str,
+            help="What should the installation be called locally?",
+            required=True,
+        )
+        self.parser.add_argument(
+            "--conda-env",
+            type=str,
+            help="conda env in which this distribution will run (default = distribution name)",
+        )
+
+    def _run_distribution_install_cmd(self, args: argparse.Namespace) -> None:
+        from llama_toolchain.common.exec import run_with_pty
+        from llama_toolchain.distribution.datatypes import DistributionConfig
+        from llama_toolchain.distribution.distribution import distribution_dependencies
+        from llama_toolchain.distribution.registry import resolve_distribution_spec
+
+        os.makedirs(DISTRIBS_BASE_DIR, exist_ok=True)
+        script = pkg_resources.resource_filename(
+            "llama_toolchain",
+            "distribution/install_distribution.sh",
+        )
+
+        dist = resolve_distribution_spec(args.spec)
+        if dist is None:
+            self.parser.error(f"Could not find distribution {args.spec}")
+            return
+
+        distrib_dir = DISTRIBS_BASE_DIR / args.name
+        os.makedirs(distrib_dir, exist_ok=True)
+
+        deps = distribution_dependencies(dist)
+        if not args.conda_env:
+            print(f"Using {args.name} as the Conda environment for this distribution")
+
+        conda_env = args.conda_env or args.name
+
+        config_file = distrib_dir / "config.yaml"
+        if config_file.exists():
+            c = DistributionConfig(**yaml.safe_load(config_file.read_text()))
+            if c.spec != dist.spec_id:
+                self.parser.error(
+                    f"already installed distribution with `spec={c.spec}` does not match provided spec `{args.spec}`"
+                )
+                return
+            if c.conda_env != conda_env:
+                self.parser.error(
+                    f"already installed distribution has `conda_env={c.conda_env}` different from provided conda env `{conda_env}`"
+                )
+                return
+        else:
+            with open(config_file, "w") as f:
+                c = DistributionConfig(
+                    spec=dist.spec_id,
+                    name=args.name,
+                    conda_env=conda_env,
+                )
+                f.write(yaml.dump(c.dict(), sort_keys=False))
+
+        return_code = run_with_pty([script, conda_env, args.name, " ".join(deps)])
+
+        assert return_code == 0, cprint(
+            f"Failed to install distribution {dist.spec_id}", color="red"
+        )
+        cprint(
+            f"Distribution `{args.name}` (with spec {dist.spec_id}) has been installed successfully!",
+            color="green",
+        )

llama_toolchain/cli/distribution/list.py

@@ -0,0 +1,54 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+import json
+
+from llama_toolchain.cli.subcommand import Subcommand
+
+
+class DistributionList(Subcommand):
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "list",
+            prog="llama distribution list",
+            description="Show available llama stack distributions",
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_distribution_list_cmd)
+
+    def _add_arguments(self):
+        pass
+
+    def _run_distribution_list_cmd(self, args: argparse.Namespace) -> None:
+        from llama_toolchain.cli.table import print_table
+        from llama_toolchain.distribution.registry import available_distribution_specs
+
+        # eventually, this should query a registry at llama.meta.com/llamastack/distributions
+        headers = [
+            "Spec ID",
+            "ProviderSpecs",
+            "Description",
+        ]
+
+        rows = []
+        for spec in available_distribution_specs():
+            providers = {k.value: v.provider_id for k, v in spec.provider_specs.items()}
+            rows.append(
+                [
+                    spec.spec_id,
+                    json.dumps(providers, indent=2),
+                    spec.description,
+                ]
+            )
+        print_table(
+            rows,
+            headers,
+            separate_rows=True,
+        )

llama_toolchain/cli/distribution/start.py

@@ -0,0 +1,82 @@
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the terms described in the LICENSE file in
+# the root directory of this source tree.
+
+import argparse
+
+import pkg_resources
+import yaml
+
+from llama_toolchain.cli.subcommand import Subcommand
+from llama_toolchain.common.config_dirs import DISTRIBS_BASE_DIR
+
+
+class DistributionStart(Subcommand):
+
+    def __init__(self, subparsers: argparse._SubParsersAction):
+        super().__init__()
+        self.parser = subparsers.add_parser(
+            "start",
+            prog="llama distribution start",
+            description="""start the server for a Llama stack distribution. you should have already installed and configured the distribution""",
+            formatter_class=argparse.RawTextHelpFormatter,
+        )
+        self._add_arguments()
+        self.parser.set_defaults(func=self._run_distribution_start_cmd)
+
+    def _add_arguments(self):
+        self.parser.add_argument(
+            "--name",
+            type=str,
+            help="Name of the distribution to start",
+            required=True,
+        )
+        self.parser.add_argument(
+            "--port",
+            type=int,
+            help="Port to run the server on. Defaults to 5000",
+            default=5000,
+        )
+        self.parser.add_argument(
+            "--disable-ipv6",
+            action="store_true",
+            help="Disable IPv6 support",
+            default=False,
+        )
+
+    def _run_distribution_start_cmd(self, args: argparse.Namespace) -> None:
+        from llama_toolchain.common.exec import run_with_pty
+        from llama_toolchain.distribution.registry import resolve_distribution_spec
+
+        config_file = DISTRIBS_BASE_DIR / args.name / "config.yaml"
+        if not config_file.exists():
+            self.parser.error(
+                f"Could not find {config_file}. Please run `llama distribution install` first"
+            )
+            return
+
+        # we need to find the spec from the name
+        with open(config_file, "r") as f:
+            config = yaml.safe_load(f)
+
+        dist = resolve_distribution_spec(config["spec"])
+        if dist is None:
+            raise ValueError(f"Could not find any registered spec `{config['spec']}`")
+
+        conda_env = config["conda_env"]
+        if not conda_env:
+            raise ValueError(
+                f"Could not find Conda environment for distribution `{args.name}`"
+            )
+
+        script = pkg_resources.resource_filename(
+            "llama_toolchain",
+            "distribution/start_distribution.sh",
+        )
+        args = [script, conda_env, config_file, "--port", str(args.port)] + (
+            ["--disable-ipv6"] if args.disable_ipv6 else []
+        )
+
+        run_with_pty(args)