Merge branch 'main' into add-watsonx-inference-adapter

llama_stack/cli/stack/_build.py

@@ -89,6 +89,43 @@ def run_stack_build_command(args: argparse.Namespace) -> None:
                 color="red",
             )
             sys.exit(1)
+    elif args.providers:
+        providers = dict()
+        for api_provider in args.providers.split(","):
+            if "=" not in api_provider:
+                cprint(
+                    "Could not parse `--providers`. Please ensure the list is in the format api1=provider1,api2=provider2",
+                    color="red",
+                )
+                sys.exit(1)
+            api, provider = api_provider.split("=")
+            providers_for_api = get_provider_registry().get(Api(api), None)
+            if providers_for_api is None:
+                cprint(
+                    f"{api} is not a valid API.",
+                    color="red",
+                )
+                sys.exit(1)
+            if provider in providers_for_api:
+                providers.setdefault(api, []).append(provider)
+            else:
+                cprint(
+                    f"{provider} is not a valid provider for the {api} API.",
+                    color="red",
+                )
+                sys.exit(1)
+        distribution_spec = DistributionSpec(
+            providers=providers,
+            description=",".join(args.providers),
+        )
+        if not args.image_type:
+            cprint(
+                f"Please specify a image-type (container | conda | venv) for {args.template}",
+                color="red",
+            )
+            sys.exit(1)
+
+        build_config = BuildConfig(image_type=args.image_type, distribution_spec=distribution_spec)
     elif not args.config and not args.template:
         name = prompt(
             "> Enter a name for your Llama Stack (e.g. my-local-stack): ",

llama_stack/cli/stack/build.py

@@ -75,6 +75,12 @@ the build. If not specified, currently active environment will be used if found.
             default=False,
             help="Run the stack after building using the same image type, name, and other applicable arguments",
         )
+        self.parser.add_argument(
+            "--providers",
+            type=str,
+            default=None,
+            help="Build a config for a list of providers and only those providers. This list is formatted like: api1=provider1,api2=provider2. Where there can be multiple providers per API.",
+        )
 
     def _run_stack_build_command(self, args: argparse.Namespace) -> None:
         # always keep implementation completely silo-ed away from CLI so CLI

llama_stack/distribution/ui/page/playground/tools.py

@@ -56,6 +56,17 @@ def tool_chat_page():
         st.subheader(f"Active Tools: 🛠 {len(active_tool_list)}")
         st.json(active_tool_list)
 
+        st.subheader("Chat Configurations")
+        max_tokens = st.slider(
+            "Max Tokens",
+            min_value=0,
+            max_value=4096,
+            value=512,
+            step=1,
+            help="The maximum number of tokens to generate",
+            on_change=reset_agent,
+        )
+
     @st.cache_resource
     def create_agent():
         return Agent(
@@ -63,9 +74,7 @@ def tool_chat_page():
             model=model,
             instructions="You are a helpful assistant. When you use a tool always respond with a summary of the result.",
             tools=toolgroup_selection,
-            sampling_params={
-                "strategy": {"type": "greedy"},
-            },
+            sampling_params={"strategy": {"type": "greedy"}, "max_tokens": max_tokens},
         )
 
     agent = create_agent()

llama_stack/providers/remote/inference/vllm/vllm.py

@@ -374,7 +374,8 @@ class VLLMInferenceAdapter(Inference, ModelsProtocolPrivate):
             options["max_tokens"] = self.config.max_tokens
 
         input_dict: dict[str, Any] = {}
-        if isinstance(request, ChatCompletionRequest) and request.tools is not None:
+        # Only include the 'tools' param if there is any. It can break things if an empty list is sent to the vLLM.
+        if isinstance(request, ChatCompletionRequest) and request.tools:
             input_dict = {"tools": _convert_to_vllm_tools_in_request(request.tools)}
 
         if isinstance(request, ChatCompletionRequest):

llama_stack/providers/remote/post_training/nvidia/models.py

@@ -16,7 +16,11 @@ _MODEL_ENTRIES = [
     build_hf_repo_model_entry(
         "meta/llama-3.1-8b-instruct",
         CoreModelId.llama3_1_8b_instruct.value,
-    )
+    ),
+    build_hf_repo_model_entry(
+        "meta/llama-3.2-1b-instruct",
+        CoreModelId.llama3_2_1b_instruct.value,
+    ),
 ]
 
 

llama_stack/templates/remote-vllm/doc_template.md

@@ -28,7 +28,7 @@ The following environment variables can be configured:
 
 ## Setting up vLLM server
 
-In the following sections, we'll use either AMD and NVIDIA GPUs to serve as hardware accelerators for the vLLM
+In the following sections, we'll use AMD, NVIDIA or Intel GPUs to serve as hardware accelerators for the vLLM
 server, which acts as both the LLM inference provider and the safety provider. Note that vLLM also
 [supports many other hardware accelerators](https://docs.vllm.ai/en/latest/getting_started/installation.html) and
 that we only use GPUs here for demonstration purposes.
@@ -149,6 +149,55 @@ docker run \
     --port $SAFETY_PORT
 ```
 
+### Setting up vLLM server on Intel GPU
+
+Refer to [vLLM Documentation for XPU](https://docs.vllm.ai/en/v0.8.2/getting_started/installation/gpu.html?device=xpu) to get a vLLM endpoint. In addition to vLLM side setup which guides towards installing vLLM from sources orself-building vLLM Docker container, Intel provides prebuilt vLLM container to use on systems with Intel GPUs supported by PyTorch XPU backend:
+- [intel/vllm](https://hub.docker.com/r/intel/vllm)
+
+Here is a sample script to start a vLLM server locally via Docker using Intel provided container:
+
+```bash
+export INFERENCE_PORT=8000
+export INFERENCE_MODEL=meta-llama/Llama-3.2-1B-Instruct
+export ZE_AFFINITY_MASK=0
+
+docker run \
+    --pull always \
+    --device /dev/dri \
+    -v /dev/dri/by-path:/dev/dri/by-path \
+    -v ~/.cache/huggingface:/root/.cache/huggingface \
+    --env "HUGGING_FACE_HUB_TOKEN=$HF_TOKEN" \
+    --env ZE_AFFINITY_MASK=$ZE_AFFINITY_MASK \
+    -p $INFERENCE_PORT:$INFERENCE_PORT \
+    --ipc=host \
+    intel/vllm:xpu \
+    --gpu-memory-utilization 0.7 \
+    --model $INFERENCE_MODEL \
+    --port $INFERENCE_PORT
+```
+
+If you are using Llama Stack Safety / Shield APIs, then you will need to also run another instance of a vLLM with a corresponding safety model like `meta-llama/Llama-Guard-3-1B` using a script like:
+
+```bash
+export SAFETY_PORT=8081
+export SAFETY_MODEL=meta-llama/Llama-Guard-3-1B
+export ZE_AFFINITY_MASK=1
+
+docker run \
+    --pull always \
+    --device /dev/dri \
+    -v /dev/dri/by-path:/dev/dri/by-path \
+    -v ~/.cache/huggingface:/root/.cache/huggingface \
+    --env "HUGGING_FACE_HUB_TOKEN=$HF_TOKEN" \
+    --env ZE_AFFINITY_MASK=$ZE_AFFINITY_MASK \
+    -p $SAFETY_PORT:$SAFETY_PORT \
+    --ipc=host \
+    intel/vllm:xpu \
+    --gpu-memory-utilization 0.7 \
+    --model $SAFETY_MODEL \
+    --port $SAFETY_PORT
+```
+
 ## Running Llama Stack
 
 Now you are ready to run Llama Stack with vLLM as the inference provider. You can do this via Conda (build code) or Docker which has a pre-built image.