feat: Add watsonx inference adapter (#1895)

# What does this PR do?
IBM watsonx ai added as the inference [#1741
](https://github.com/meta-llama/llama-stack/issues/1741)

[//]: # (If resolving an issue, uncomment and update the line below)
[//]: # (Closes #[issue-number])

---------

Co-authored-by: Sajikumar JS <sajikumar.js@ibm.com>

README.md

@@ -119,6 +119,7 @@ Here is a list of the various API providers and available distributions that can
 |          OpenAI          |         Hosted         |            |       ✅       |            |            |               |
 |        Anthropic         |         Hosted         |            |       ✅       |            |            |               |
 |          Gemini          |         Hosted         |            |       ✅       |            |            |               |
+|          watsonx         |         Hosted         |            |       ✅       |            |            |               |
 
 
 ### Distributions

docs/source/distributions/remote_hosted_distro/watsonx.md

@@ -0,0 +1,88 @@
+---
+orphan: true
+---
+<!-- This file was auto-generated by distro_codegen.py, please edit source -->
+# watsonx Distribution
+
+```{toctree}
+:maxdepth: 2
+:hidden:
+
+self
+```
+
+The `llamastack/distribution-watsonx` distribution consists of the following provider configurations.
+
+| API | Provider(s) |
+|-----|-------------|
+| agents | `inline::meta-reference` |
+| datasetio | `remote::huggingface`, `inline::localfs` |
+| eval | `inline::meta-reference` |
+| inference | `remote::watsonx` |
+| safety | `inline::llama-guard` |
+| scoring | `inline::basic`, `inline::llm-as-judge`, `inline::braintrust` |
+| telemetry | `inline::meta-reference` |
+| tool_runtime | `remote::brave-search`, `remote::tavily-search`, `inline::code-interpreter`, `inline::rag-runtime`, `remote::model-context-protocol` |
+| vector_io | `inline::faiss` |
+
+
+
+### Environment Variables
+
+The following environment variables can be configured:
+
+- `LLAMASTACK_PORT`: Port for the Llama Stack distribution server (default: `5001`)
+- `WATSONX_API_KEY`: watsonx API Key (default: ``)
+- `WATSONX_PROJECT_ID`: watsonx Project ID (default: ``)
+
+### Models
+
+The following models are available by default:
+
+- `meta-llama/llama-3-3-70b-instruct (aliases: meta-llama/Llama-3.3-70B-Instruct)`
+- `meta-llama/llama-2-13b-chat (aliases: meta-llama/Llama-2-13b)`
+- `meta-llama/llama-3-1-70b-instruct (aliases: meta-llama/Llama-3.1-70B-Instruct)`
+- `meta-llama/llama-3-1-8b-instruct (aliases: meta-llama/Llama-3.1-8B-Instruct)`
+- `meta-llama/llama-3-2-11b-vision-instruct (aliases: meta-llama/Llama-3.2-11B-Vision-Instruct)`
+- `meta-llama/llama-3-2-1b-instruct (aliases: meta-llama/Llama-3.2-1B-Instruct)`
+- `meta-llama/llama-3-2-3b-instruct (aliases: meta-llama/Llama-3.2-3B-Instruct)`
+- `meta-llama/llama-3-2-90b-vision-instruct (aliases: meta-llama/Llama-3.2-90B-Vision-Instruct)`
+- `meta-llama/llama-guard-3-11b-vision (aliases: meta-llama/Llama-Guard-3-11B-Vision)`
+
+
+### Prerequisite: API Keys
+
+Make sure you have access to a watsonx API Key. You can get one by referring [watsonx.ai](https://www.ibm.com/docs/en/masv-and-l/maximo-manage/continuous-delivery?topic=setup-create-watsonx-api-key).
+
+
+## Running Llama Stack with watsonx
+
+You can do this via Conda (build code), venv or Docker which has a pre-built image.
+
+### Via Docker
+
+This method allows you to get started quickly without having to build the distribution code.
+
+```bash
+LLAMA_STACK_PORT=5001
+docker run \
+  -it \
+  -p $LLAMA_STACK_PORT:$LLAMA_STACK_PORT \
+  -v ./run.yaml:/root/my-run.yaml \
+  llamastack/distribution-watsonx \
+  --yaml-config /root/my-run.yaml \
+  --port $LLAMA_STACK_PORT \
+  --env WATSONX_API_KEY=$WATSONX_API_KEY \
+  --env WATSONX_PROJECT_ID=$WATSONX_PROJECT_ID \
+  --env WATSONX_BASE_URL=$WATSONX_BASE_URL
+```
+
+### Via Conda
+
+```bash
+llama stack build --template watsonx --image-type conda
+llama stack run ./run.yaml \
+  --port $LLAMA_STACK_PORT \
+  --env WATSONX_API_KEY=$WATSONX_API_KEY \
+  --env WATSONX_PROJECT_ID=$WATSONX_PROJECT_ID
+```

llama_stack/providers/registry/inference.py

@@ -288,4 +288,14 @@ def available_providers() -> List[ProviderSpec]:
                 provider_data_validator="llama_stack.providers.remote.inference.passthrough.PassthroughProviderDataValidator",
             ),
         ),
+        remote_provider_spec(
+            api=Api.inference,
+            adapter=AdapterSpec(
+                adapter_type="watsonx",
+                pip_packages=["ibm_watson_machine_learning"],
+                module="llama_stack.providers.remote.inference.watsonx",
+                config_class="llama_stack.providers.remote.inference.watsonx.WatsonXConfig",
+                provider_data_validator="llama_stack.providers.remote.inference.watsonx.WatsonXProviderDataValidator",
+            ),
+        ),
     ]

llama_stack/providers/remote/inference/watsonx/__init__.py

@@ -0,0 +1,22 @@
+# 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 llama_stack.apis.inference import Inference
+
+from .config import WatsonXConfig
+
+
+async def get_adapter_impl(config: WatsonXConfig, _deps) -> Inference:
+    # import dynamically so `llama stack build` does not fail due to missing dependencies
+    from .watsonx import WatsonXInferenceAdapter
+
+    if not isinstance(config, WatsonXConfig):
+        raise RuntimeError(f"Unexpected config type: {type(config)}")
+    adapter = WatsonXInferenceAdapter(config)
+    return adapter
+
+
+__all__ = ["get_adapter_impl", "WatsonXConfig"]

llama_stack/providers/remote/inference/watsonx/config.py

@@ -0,0 +1,46 @@
+# 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 os
+from typing import Any, Dict, Optional
+
+from pydantic import BaseModel, Field, SecretStr
+
+from llama_stack.schema_utils import json_schema_type
+
+
+class WatsonXProviderDataValidator(BaseModel):
+    url: str
+    api_key: str
+    project_id: str
+
+
+@json_schema_type
+class WatsonXConfig(BaseModel):
+    url: str = Field(
+        default_factory=lambda: os.getenv("WATSONX_BASE_URL", "https://us-south.ml.cloud.ibm.com"),
+        description="A base url for accessing the watsonx.ai",
+    )
+    api_key: Optional[SecretStr] = Field(
+        default_factory=lambda: os.getenv("WATSONX_API_KEY"),
+        description="The watsonx API key, only needed of using the hosted service",
+    )
+    project_id: Optional[str] = Field(
+        default_factory=lambda: os.getenv("WATSONX_PROJECT_ID"),
+        description="The Project ID key, only needed of using the hosted service",
+    )
+    timeout: int = Field(
+        default=60,
+        description="Timeout for the HTTP requests",
+    )
+
+    @classmethod
+    def sample_run_config(cls, **kwargs) -> Dict[str, Any]:
+        return {
+            "url": "${env.WATSONX_BASE_URL:https://us-south.ml.cloud.ibm.com}",
+            "api_key": "${env.WATSONX_API_KEY:}",
+            "project_id": "${env.WATSONX_PROJECT_ID:}",
+        }

llama_stack/providers/remote/inference/watsonx/models.py

@@ -0,0 +1,47 @@
+# 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 llama_stack.models.llama.sku_types import CoreModelId
+from llama_stack.providers.utils.inference.model_registry import build_hf_repo_model_entry
+
+MODEL_ENTRIES = [
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-3-70b-instruct",
+        CoreModelId.llama3_3_70b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-2-13b-chat",
+        CoreModelId.llama2_13b.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-1-70b-instruct",
+        CoreModelId.llama3_1_70b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-1-8b-instruct",
+        CoreModelId.llama3_1_8b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-2-11b-vision-instruct",
+        CoreModelId.llama3_2_11b_vision_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-2-1b-instruct",
+        CoreModelId.llama3_2_1b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-2-3b-instruct",
+        CoreModelId.llama3_2_3b_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-3-2-90b-vision-instruct",
+        CoreModelId.llama3_2_90b_vision_instruct.value,
+    ),
+    build_hf_repo_model_entry(
+        "meta-llama/llama-guard-3-11b-vision",
+        CoreModelId.llama_guard_3_11b_vision.value,
+    ),
+]

llama_stack/providers/remote/inference/watsonx/watsonx.py

@@ -0,0 +1,260 @@
+# 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 typing import AsyncGenerator, List, Optional, Union
+
+from ibm_watson_machine_learning.foundation_models import Model
+from ibm_watson_machine_learning.metanames import GenTextParamsMetaNames as GenParams
+
+from llama_stack.apis.common.content_types import InterleavedContent, InterleavedContentItem
+from llama_stack.apis.inference import (
+    ChatCompletionRequest,
+    ChatCompletionResponse,
+    CompletionRequest,
+    EmbeddingsResponse,
+    EmbeddingTaskType,
+    Inference,
+    LogProbConfig,
+    Message,
+    ResponseFormat,
+    SamplingParams,
+    TextTruncation,
+    ToolChoice,
+    ToolConfig,
+    ToolDefinition,
+    ToolPromptFormat,
+)
+from llama_stack.providers.utils.inference.model_registry import ModelRegistryHelper
+from llama_stack.providers.utils.inference.openai_compat import (
+    OpenAICompatCompletionChoice,
+    OpenAICompatCompletionResponse,
+    process_chat_completion_response,
+    process_chat_completion_stream_response,
+    process_completion_response,
+    process_completion_stream_response,
+)
+from llama_stack.providers.utils.inference.prompt_adapter import (
+    chat_completion_request_to_prompt,
+    completion_request_to_prompt,
+    request_has_media,
+)
+
+from . import WatsonXConfig
+from .models import MODEL_ENTRIES
+
+
+class WatsonXInferenceAdapter(Inference, ModelRegistryHelper):
+    def __init__(self, config: WatsonXConfig) -> None:
+        ModelRegistryHelper.__init__(self, MODEL_ENTRIES)
+
+        print(f"Initializing watsonx InferenceAdapter({config.url})...")
+
+        self._config = config
+
+        self._project_id = self._config.project_id
+
+    async def initialize(self) -> None:
+        pass
+
+    async def shutdown(self) -> None:
+        pass
+
+    async def completion(
+        self,
+        model_id: str,
+        content: InterleavedContent,
+        sampling_params: Optional[SamplingParams] = None,
+        response_format: Optional[ResponseFormat] = None,
+        stream: Optional[bool] = False,
+        logprobs: Optional[LogProbConfig] = None,
+    ) -> AsyncGenerator:
+        if sampling_params is None:
+            sampling_params = SamplingParams()
+        model = await self.model_store.get_model(model_id)
+        request = CompletionRequest(
+            model=model.provider_resource_id,
+            content=content,
+            sampling_params=sampling_params,
+            response_format=response_format,
+            stream=stream,
+            logprobs=logprobs,
+        )
+        if stream:
+            return self._stream_completion(request)
+        else:
+            return await self._nonstream_completion(request)
+
+    def _get_client(self, model_id) -> Model:
+        config_api_key = self._config.api_key.get_secret_value() if self._config.api_key else None
+        config_url = self._config.url
+        project_id = self._config.project_id
+        credentials = {"url": config_url, "apikey": config_api_key}
+
+        return Model(model_id=model_id, credentials=credentials, project_id=project_id)
+
+    async def _nonstream_completion(self, request: CompletionRequest) -> ChatCompletionResponse:
+        params = await self._get_params(request)
+        r = self._get_client(request.model).generate(**params)
+        choices = []
+        if "results" in r:
+            for result in r["results"]:
+                choice = OpenAICompatCompletionChoice(
+                    finish_reason=result["stop_reason"] if result["stop_reason"] else None,
+                    text=result["generated_text"],
+                )
+                choices.append(choice)
+        response = OpenAICompatCompletionResponse(
+            choices=choices,
+        )
+        return process_completion_response(response)
+
+    async def _stream_completion(self, request: CompletionRequest) -> AsyncGenerator:
+        params = await self._get_params(request)
+
+        async def _generate_and_convert_to_openai_compat():
+            s = self._get_client(request.model).generate_text_stream(**params)
+            for chunk in s:
+                choice = OpenAICompatCompletionChoice(
+                    finish_reason=None,
+                    text=chunk,
+                )
+                yield OpenAICompatCompletionResponse(
+                    choices=[choice],
+                )
+
+        stream = _generate_and_convert_to_openai_compat()
+        async for chunk in process_completion_stream_response(stream):
+            yield chunk
+
+    async def chat_completion(
+        self,
+        model_id: str,
+        messages: List[Message],
+        sampling_params: Optional[SamplingParams] = None,
+        tools: Optional[List[ToolDefinition]] = None,
+        tool_choice: Optional[ToolChoice] = ToolChoice.auto,
+        tool_prompt_format: Optional[ToolPromptFormat] = None,
+        response_format: Optional[ResponseFormat] = None,
+        stream: Optional[bool] = False,
+        logprobs: Optional[LogProbConfig] = None,
+        tool_config: Optional[ToolConfig] = None,
+    ) -> AsyncGenerator:
+        if sampling_params is None:
+            sampling_params = SamplingParams()
+        model = await self.model_store.get_model(model_id)
+        request = ChatCompletionRequest(
+            model=model.provider_resource_id,
+            messages=messages,
+            sampling_params=sampling_params,
+            tools=tools or [],
+            response_format=response_format,
+            stream=stream,
+            logprobs=logprobs,
+            tool_config=tool_config,
+        )
+
+        if stream:
+            return self._stream_chat_completion(request)
+        else:
+            return await self._nonstream_chat_completion(request)
+
+    async def _nonstream_chat_completion(self, request: ChatCompletionRequest) -> ChatCompletionResponse:
+        params = await self._get_params(request)
+        r = self._get_client(request.model).generate(**params)
+        choices = []
+        if "results" in r:
+            for result in r["results"]:
+                choice = OpenAICompatCompletionChoice(
+                    finish_reason=result["stop_reason"] if result["stop_reason"] else None,
+                    text=result["generated_text"],
+                )
+                choices.append(choice)
+        response = OpenAICompatCompletionResponse(
+            choices=choices,
+        )
+        return process_chat_completion_response(response, request)
+
+    async def _stream_chat_completion(self, request: ChatCompletionRequest) -> AsyncGenerator:
+        params = await self._get_params(request)
+        model_id = request.model
+
+        # if we shift to TogetherAsyncClient, we won't need this wrapper
+        async def _to_async_generator():
+            s = self._get_client(model_id).generate_text_stream(**params)
+            for chunk in s:
+                choice = OpenAICompatCompletionChoice(
+                    finish_reason=None,
+                    text=chunk,
+                )
+                yield OpenAICompatCompletionResponse(
+                    choices=[choice],
+                )
+
+        stream = _to_async_generator()
+        async for chunk in process_chat_completion_stream_response(stream, request):
+            yield chunk
+
+    async def _get_params(self, request: Union[ChatCompletionRequest, CompletionRequest]) -> dict:
+        input_dict = {"params": {}}
+        media_present = request_has_media(request)
+        llama_model = self.get_llama_model(request.model)
+        if isinstance(request, ChatCompletionRequest):
+            input_dict["prompt"] = await chat_completion_request_to_prompt(request, llama_model)
+        else:
+            assert not media_present, "Together does not support media for Completion requests"
+            input_dict["prompt"] = await completion_request_to_prompt(request)
+        if request.sampling_params:
+            if request.sampling_params.strategy:
+                input_dict["params"][GenParams.DECODING_METHOD] = request.sampling_params.strategy.type
+            if request.sampling_params.max_tokens:
+                input_dict["params"][GenParams.MAX_NEW_TOKENS] = request.sampling_params.max_tokens
+            if request.sampling_params.repetition_penalty:
+                input_dict["params"][GenParams.REPETITION_PENALTY] = request.sampling_params.repetition_penalty
+            if request.sampling_params.additional_params.get("top_p"):
+                input_dict["params"][GenParams.TOP_P] = request.sampling_params.additional_params["top_p"]
+            if request.sampling_params.additional_params.get("top_k"):
+                input_dict["params"][GenParams.TOP_K] = request.sampling_params.additional_params["top_k"]
+            if request.sampling_params.additional_params.get("temperature"):
+                input_dict["params"][GenParams.TEMPERATURE] = request.sampling_params.additional_params["temperature"]
+            if request.sampling_params.additional_params.get("length_penalty"):
+                input_dict["params"][GenParams.LENGTH_PENALTY] = request.sampling_params.additional_params[
+                    "length_penalty"
+                ]
+            if request.sampling_params.additional_params.get("random_seed"):
+                input_dict["params"][GenParams.RANDOM_SEED] = request.sampling_params.additional_params["random_seed"]
+            if request.sampling_params.additional_params.get("min_new_tokens"):
+                input_dict["params"][GenParams.MIN_NEW_TOKENS] = request.sampling_params.additional_params[
+                    "min_new_tokens"
+                ]
+            if request.sampling_params.additional_params.get("stop_sequences"):
+                input_dict["params"][GenParams.STOP_SEQUENCES] = request.sampling_params.additional_params[
+                    "stop_sequences"
+                ]
+            if request.sampling_params.additional_params.get("time_limit"):
+                input_dict["params"][GenParams.TIME_LIMIT] = request.sampling_params.additional_params["time_limit"]
+            if request.sampling_params.additional_params.get("truncate_input_tokens"):
+                input_dict["params"][GenParams.TRUNCATE_INPUT_TOKENS] = request.sampling_params.additional_params[
+                    "truncate_input_tokens"
+                ]
+            if request.sampling_params.additional_params.get("return_options"):
+                input_dict["params"][GenParams.RETURN_OPTIONS] = request.sampling_params.additional_params[
+                    "return_options"
+                ]
+
+        params = {
+            **input_dict,
+        }
+        return params
+
+    async def embeddings(
+        self,
+        model_id: str,
+        contents: List[str] | List[InterleavedContentItem],
+        text_truncation: Optional[TextTruncation] = TextTruncation.none,
+        output_dimension: Optional[int] = None,
+        task_type: Optional[EmbeddingTaskType] = None,
+    ) -> EmbeddingsResponse:
+        pass

llama_stack/templates/dependencies.json

@@ -755,5 +755,41 @@
     "vllm",
     "sentence-transformers --no-deps",
     "torch torchvision --index-url https://download.pytorch.org/whl/cpu"
+  ],
+  "watsonx": [
+    "aiosqlite",
+    "autoevals",
+    "blobfile",
+    "chardet",
+    "datasets",
+    "emoji",
+    "faiss-cpu",
+    "fastapi",
+    "fire",
+    "httpx",
+    "ibm_watson_machine_learning",
+    "langdetect",
+    "matplotlib",
+    "mcp",
+    "nltk",
+    "numpy",
+    "openai",
+    "opentelemetry-exporter-otlp-proto-http",
+    "opentelemetry-sdk",
+    "pandas",
+    "pillow",
+    "psycopg2-binary",
+    "pymongo",
+    "pypdf",
+    "pythainlp",
+    "redis",
+    "requests",
+    "scikit-learn",
+    "scipy",
+    "sentencepiece",
+    "tqdm",
+    "transformers",
+    "tree_sitter",
+    "uvicorn"
   ]
 }

llama_stack/templates/watsonx/__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 .watsonx import get_distribution_template  # noqa: F401

llama_stack/templates/watsonx/build.yaml

@@ -0,0 +1,30 @@
+version: '2'
+distribution_spec:
+  description: Use watsonx for running LLM inference
+  providers:
+    inference:
+    - remote::watsonx
+    vector_io:
+    - inline::faiss
+    safety:
+    - inline::llama-guard
+    agents:
+    - inline::meta-reference
+    telemetry:
+    - inline::meta-reference
+    eval:
+    - inline::meta-reference
+    datasetio:
+    - remote::huggingface
+    - inline::localfs
+    scoring:
+    - inline::basic
+    - inline::llm-as-judge
+    - inline::braintrust
+    tool_runtime:
+    - remote::brave-search
+    - remote::tavily-search
+    - inline::code-interpreter
+    - inline::rag-runtime
+    - remote::model-context-protocol
+image_type: conda

llama_stack/templates/watsonx/doc_template.md

@@ -0,0 +1,74 @@
+---
+orphan: true
+---
+# watsonx Distribution
+
+```{toctree}
+:maxdepth: 2
+:hidden:
+
+self
+```
+
+The `llamastack/distribution-{{ name }}` distribution consists of the following provider configurations.
+
+{{ providers_table }}
+
+{% if run_config_env_vars  %}
+
+### Environment Variables
+
+The following environment variables can be configured:
+
+{% for var, (default_value, description) in run_config_env_vars.items() %}
+- `{{ var }}`: {{ description }} (default: `{{ default_value }}`)
+{% endfor %}
+{% endif %}
+
+{% if default_models %}
+### Models
+
+The following models are available by default:
+
+{% for model in default_models %}
+- `{{ model.model_id }} {{ model.doc_string }}`
+{% endfor %}
+{% endif %}
+
+
+### Prerequisite: API Keys
+
+Make sure you have access to a watsonx API Key. You can get one by referring [watsonx.ai](https://www.ibm.com/docs/en/masv-and-l/maximo-manage/continuous-delivery?topic=setup-create-watsonx-api-key).
+
+
+## Running Llama Stack with watsonx
+
+You can do this via Conda (build code), venv or Docker which has a pre-built image.
+
+### Via Docker
+
+This method allows you to get started quickly without having to build the distribution code.
+
+```bash
+LLAMA_STACK_PORT=5001
+docker run \
+  -it \
+  -p $LLAMA_STACK_PORT:$LLAMA_STACK_PORT \
+  -v ./run.yaml:/root/my-run.yaml \
+  llamastack/distribution-{{ name }} \
+  --yaml-config /root/my-run.yaml \
+  --port $LLAMA_STACK_PORT \
+  --env WATSONX_API_KEY=$WATSONX_API_KEY \
+  --env WATSONX_PROJECT_ID=$WATSONX_PROJECT_ID \
+  --env WATSONX_BASE_URL=$WATSONX_BASE_URL
+```
+
+### Via Conda
+
+```bash
+llama stack build --template watsonx --image-type conda
+llama stack run ./run.yaml \
+  --port $LLAMA_STACK_PORT \
+  --env WATSONX_API_KEY=$WATSONX_API_KEY \
+  --env WATSONX_PROJECT_ID=$WATSONX_PROJECT_ID
+```

llama_stack/templates/watsonx/run.yaml

@@ -0,0 +1,210 @@
+version: '2'
+image_name: watsonx
+apis:
+- agents
+- datasetio
+- eval
+- inference
+- safety
+- scoring
+- telemetry
+- tool_runtime
+- vector_io
+providers:
+  inference:
+  - provider_id: watsonx
+    provider_type: remote::watsonx
+    config:
+      url: ${env.WATSONX_BASE_URL:https://us-south.ml.cloud.ibm.com}
+      api_key: ${env.WATSONX_API_KEY:}
+      project_id: ${env.WATSONX_PROJECT_ID:}
+  vector_io:
+  - provider_id: faiss
+    provider_type: inline::faiss
+    config:
+      kvstore:
+        type: sqlite
+        namespace: null
+        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/faiss_store.db
+  safety:
+  - provider_id: llama-guard
+    provider_type: inline::llama-guard
+    config:
+      excluded_categories: []
+  agents:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      persistence_store:
+        type: sqlite
+        namespace: null
+        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/agents_store.db
+  telemetry:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      service_name: "${env.OTEL_SERVICE_NAME:\u200B}"
+      sinks: ${env.TELEMETRY_SINKS:console,sqlite}
+      sqlite_db_path: ${env.SQLITE_DB_PATH:~/.llama/distributions/watsonx/trace_store.db}
+  eval:
+  - provider_id: meta-reference
+    provider_type: inline::meta-reference
+    config:
+      kvstore:
+        type: sqlite
+        namespace: null
+        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/meta_reference_eval.db
+  datasetio:
+  - provider_id: huggingface
+    provider_type: remote::huggingface
+    config:
+      kvstore:
+        type: sqlite
+        namespace: null
+        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/huggingface_datasetio.db
+  - provider_id: localfs
+    provider_type: inline::localfs
+    config:
+      kvstore:
+        type: sqlite
+        namespace: null
+        db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/localfs_datasetio.db
+  scoring:
+  - provider_id: basic
+    provider_type: inline::basic
+    config: {}
+  - provider_id: llm-as-judge
+    provider_type: inline::llm-as-judge
+    config: {}
+  - provider_id: braintrust
+    provider_type: inline::braintrust
+    config:
+      openai_api_key: ${env.OPENAI_API_KEY:}
+  tool_runtime:
+  - provider_id: brave-search
+    provider_type: remote::brave-search
+    config:
+      api_key: ${env.BRAVE_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: tavily-search
+    provider_type: remote::tavily-search
+    config:
+      api_key: ${env.TAVILY_SEARCH_API_KEY:}
+      max_results: 3
+  - provider_id: code-interpreter
+    provider_type: inline::code-interpreter
+    config: {}
+  - provider_id: rag-runtime
+    provider_type: inline::rag-runtime
+    config: {}
+  - provider_id: model-context-protocol
+    provider_type: remote::model-context-protocol
+    config: {}
+metadata_store:
+  type: sqlite
+  db_path: ${env.SQLITE_STORE_DIR:~/.llama/distributions/watsonx}/registry.db
+models:
+- metadata: {}
+  model_id: meta-llama/llama-3-3-70b-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-3-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.3-70B-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-3-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-2-13b-chat
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-2-13b-chat
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-2-13b
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-2-13b-chat
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-1-70b-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.1-70B-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-1-70b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-1-8b-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-1-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.1-8B-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-1-8b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-2-11b-vision-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-11b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.2-11B-Vision-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-11b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-2-1b-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-1b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.2-1B-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-1b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-2-3b-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-3b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.2-3B-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-3b-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-3-2-90b-vision-instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-90b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-3.2-90B-Vision-Instruct
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-3-2-90b-vision-instruct
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/llama-guard-3-11b-vision
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-guard-3-11b-vision
+  model_type: llm
+- metadata: {}
+  model_id: meta-llama/Llama-Guard-3-11B-Vision
+  provider_id: watsonx
+  provider_model_id: meta-llama/llama-guard-3-11b-vision
+  model_type: llm
+shields: []
+vector_dbs: []
+datasets: []
+scoring_fns: []
+benchmarks: []
+tool_groups:
+- toolgroup_id: builtin::websearch
+  provider_id: tavily-search
+- toolgroup_id: builtin::rag
+  provider_id: rag-runtime
+- toolgroup_id: builtin::code_interpreter
+  provider_id: code-interpreter
+server:
+  port: 8321

llama_stack/templates/watsonx/watsonx.py

@@ -0,0 +1,90 @@
+# 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 pathlib import Path
+
+from llama_stack.distribution.datatypes import Provider, ToolGroupInput
+from llama_stack.providers.remote.inference.watsonx import WatsonXConfig
+from llama_stack.providers.remote.inference.watsonx.models import MODEL_ENTRIES
+from llama_stack.templates.template import DistributionTemplate, RunConfigSettings, get_model_registry
+
+
+def get_distribution_template() -> DistributionTemplate:
+    providers = {
+        "inference": ["remote::watsonx"],
+        "vector_io": ["inline::faiss"],
+        "safety": ["inline::llama-guard"],
+        "agents": ["inline::meta-reference"],
+        "telemetry": ["inline::meta-reference"],
+        "eval": ["inline::meta-reference"],
+        "datasetio": ["remote::huggingface", "inline::localfs"],
+        "scoring": ["inline::basic", "inline::llm-as-judge", "inline::braintrust"],
+        "tool_runtime": [
+            "remote::brave-search",
+            "remote::tavily-search",
+            "inline::code-interpreter",
+            "inline::rag-runtime",
+            "remote::model-context-protocol",
+        ],
+    }
+
+    inference_provider = Provider(
+        provider_id="watsonx",
+        provider_type="remote::watsonx",
+        config=WatsonXConfig.sample_run_config(),
+    )
+
+    available_models = {
+        "watsonx": MODEL_ENTRIES,
+    }
+    default_tool_groups = [
+        ToolGroupInput(
+            toolgroup_id="builtin::websearch",
+            provider_id="tavily-search",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::rag",
+            provider_id="rag-runtime",
+        ),
+        ToolGroupInput(
+            toolgroup_id="builtin::code_interpreter",
+            provider_id="code-interpreter",
+        ),
+    ]
+
+    default_models = get_model_registry(available_models)
+    return DistributionTemplate(
+        name="watsonx",
+        distro_type="remote_hosted",
+        description="Use watsonx for running LLM inference",
+        container_image=None,
+        template_path=Path(__file__).parent / "doc_template.md",
+        providers=providers,
+        available_models_by_provider=available_models,
+        run_configs={
+            "run.yaml": RunConfigSettings(
+                provider_overrides={
+                    "inference": [inference_provider],
+                },
+                default_models=default_models,
+                default_tool_groups=default_tool_groups,
+            ),
+        },
+        run_config_env_vars={
+            "LLAMASTACK_PORT": (
+                "5001",
+                "Port for the Llama Stack distribution server",
+            ),
+            "WATSONX_API_KEY": (
+                "",
+                "watsonx API Key",
+            ),
+            "WATSONX_PROJECT_ID": (
+                "",
+                "watsonx Project ID",
+            ),
+        },
+    )

pyproject.toml

@@ -274,6 +274,7 @@ exclude = [
     "^llama_stack/providers/remote/inference/sample/",
     "^llama_stack/providers/remote/inference/tgi/",
     "^llama_stack/providers/remote/inference/together/",
+    "^llama_stack/providers/remote/inference/watsonx/",
     "^llama_stack/providers/remote/safety/bedrock/",
     "^llama_stack/providers/remote/safety/nvidia/",
     "^llama_stack/providers/remote/safety/sample/",