Distributions updates (slight updates to ollama, add inline-vllm and remote-vllm) (#408)

* remote vllm distro

* add inline-vllm details, fix things

* Write some docs

distributions/inline-vllm/build.yaml

@@ -0,0 +1 @@
+../../llama_stack/templates/inline-vllm/build.yaml

distributions/inline-vllm/compose.yaml

@@ -0,0 +1,35 @@
+services:
+  llamastack:
+    image: llamastack/distribution-inline-vllm
+    network_mode: "host"
+    volumes:
+      - ~/.llama:/root/.llama
+      - ./run.yaml:/root/my-run.yaml
+    ports:
+      - "5000:5000"
+    devices:
+      - nvidia.com/gpu=all
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+    command: []
+    deploy:
+      resources:
+        reservations:
+          devices:
+          - driver: nvidia
+            # that's the closest analogue to --gpus; provide
+            # an integer amount of devices or 'all'
+            count: 1
+            # Devices are reserved using a list of capabilities, making
+            # capabilities the only required field. A device MUST
+            # satisfy all the requested capabilities for a successful
+            # reservation.
+            capabilities: [gpu]
+    runtime: nvidia
+    entrypoint: bash -c "python -m llama_stack.distribution.server.server --yaml_config /root/my-run.yaml"
+    deploy:
+      restart_policy:
+        condition: on-failure
+        delay: 3s
+        max_attempts: 5
+        window: 60s

distributions/inline-vllm/run.yaml

@@ -0,0 +1,66 @@
+version: '2'
+built_at: '2024-10-08T17:40:45.325529'
+image_name: local
+docker_image: null
+conda_env: local
+apis:
+- shields
+- agents
+- models
+- memory
+- memory_banks
+- inference
+- safety
+providers:
+  inference:
+  - provider_id: vllm-inference
+    provider_type: inline::vllm
+    config:
+      model: Llama3.2-3B-Instruct
+      tensor_parallel_size: 1
+      gpu_memory_utilization: 0.4
+      enforce_eager: true
+      max_tokens: 4096
+  - provider_id: vllm-safety
+    provider_type: inline::vllm
+    config:
+      model: Llama-Guard-3-1B
+      tensor_parallel_size: 1
+      gpu_memory_utilization: 0.2
+      enforce_eager: true
+      max_tokens: 4096
+  safety:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config:
+      llama_guard_shield:
+        model: Llama-Guard-3-1B
+        excluded_categories: []
+# Uncomment to use prompt guard
+#      prompt_guard_shield:
+#        model: Prompt-Guard-86M
+  memory:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config: {}
+  # Uncomment to use pgvector
+  # - provider_id: pgvector
+  #   provider_type: remote::pgvector
+  #   config:
+  #     host: 127.0.0.1
+  #     port: 5432
+  #     db: postgres
+  #     user: postgres
+  #     password: mysecretpassword
+  agents:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config:
+      persistence_store:
+        namespace: null
+        type: sqlite
+        db_path: ~/.llama/runtime/agents_store.db
+  telemetry:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config: {}

distributions/ollama-gpu/build.yaml

@@ -0,0 +1 @@
+../../llama_stack/templates/ollama/build.yaml

distributions/remote-vllm/build.yaml

@@ -0,0 +1 @@
+../../llama_stack/templates/remote-vllm/build.yaml

distributions/remote-vllm/compose.yaml

@@ -0,0 +1,48 @@
+services:
+  vllm:
+    image: vllm/vllm-openai:latest
+    network_mode: "host"
+    volumes:
+      - $HOME/.cache/huggingface:/root/.cache/huggingface
+    ports:
+      - "8000:8000"
+    devices:
+      - nvidia.com/gpu=all
+    environment:
+      - CUDA_VISIBLE_DEVICES=0
+    command: []
+    deploy:
+      resources:
+        reservations:
+          devices:
+          - driver: nvidia
+            # that's the closest analogue to --gpus; provide
+            # an integer amount of devices or 'all'
+            count: 1
+            # Devices are reserved using a list of capabilities, making
+            # capabilities the only required field. A device MUST
+            # satisfy all the requested capabilities for a successful
+            # reservation.
+            capabilities: [gpu]
+    runtime: nvidia
+  llamastack:
+    depends_on:
+    - vllm
+    image: llamastack/distribution-remote-vllm
+    network_mode: "host"
+    volumes:
+      - ~/.llama:/root/.llama
+      # Link to ollama run.yaml file
+      - ./run.yaml:/root/llamastack-run-remote-vllm.yaml
+    ports:
+      - "5000:5000"
+    # Hack: wait for vllm server to start before starting docker
+    entrypoint: bash -c "sleep 60; python -m llama_stack.distribution.server.server --yaml_config /root/llamastack-run-remote-vllm.yaml"
+    deploy:
+      restart_policy:
+        condition: on-failure
+        delay: 3s
+        max_attempts: 5
+        window: 60s
+volumes:
+  vllm:

distributions/remote-vllm/run.yaml

@@ -0,0 +1,46 @@
+version: '2'
+built_at: '2024-10-08T17:40:45.325529'
+image_name: local
+docker_image: null
+conda_env: local
+apis:
+- shields
+- agents
+- models
+- memory
+- memory_banks
+- inference
+- safety
+providers:
+  inference:
+  - provider_id: vllm0
+    provider_type: remote::vllm
+    config:
+      url: http://127.0.0.1:8000
+  safety:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config:
+      llama_guard_shield:
+        model: Llama-Guard-3-1B
+        excluded_categories: []
+        disable_input_check: false
+        disable_output_check: false
+      prompt_guard_shield:
+        model: Prompt-Guard-86M
+  memory:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config: {}
+  agents:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config:
+      persistence_store:
+        namespace: null
+        type: sqlite
+        db_path: ~/.llama/runtime/kvstore.db
+  telemetry:
+  - provider_id: meta0
+    provider_type: meta-reference
+    config: {}

distributions/vllm/build.yaml

@@ -1 +0,0 @@
-../../llama_stack/templates/vllm/build.yaml

docs/source/getting_started/distributions/self_hosted_distro/ollama.md

@@ -2,25 +2,35 @@
 
 The `llamastack/distribution-ollama` distribution consists of the following provider configurations.
 
-| **API**         	| **Inference**  	| **Agents**     	| **Memory**                       	| **Safety**     	| **Telemetry**  	|
-|-----------------	|----------------	|----------------	|----------------------------------	|----------------	|----------------	|
-| **Provider(s)** 	| remote::ollama 	| meta-reference 	| remote::pgvector, remote::chroma 	| remote::ollama 	| meta-reference 	|
+| **API**         	| **Inference**  	| **Agents**     	| **Memory**                       	  | **Safety**     	| **Telemetry**  	|
+|-----------------	|----------------	|----------------	|------------------------------------	|----------------	|----------------	|
+| **Provider(s)** 	| remote::ollama 	| meta-reference 	| remote::pgvector, remote::chromadb 	| meta-reference 	| meta-reference 	|
 
 
-### Docker: Start a Distribution (Single Node GPU)
+## Using Docker Compose
+
+You can use `docker compose` to start a Ollama server and connect with Llama Stack server in a single command.
+
+### Docker: Start the Distribution (Single Node regular Desktop machine)
+
+> [!NOTE]
+> This will start an ollama server with CPU only, please see [Ollama Documentations](https://github.com/ollama/ollama) for serving models on CPU only.
+
+```bash
+$ cd distributions/ollama; docker compose up
+```
+
+### Docker: Start a Distribution (Single Node with nvidia GPUs)
 
 > [!NOTE]
 > This assumes you have access to GPU to start a Ollama server with access to your GPU.
 
-```
-$ cd distributions/ollama/gpu
-$ ls
-compose.yaml  run.yaml
-$ docker compose up
+```bash
+$ cd distributions/ollama-gpu; docker compose up
 ```
 
 You will see outputs similar to following ---
-```
+```bash
 [ollama]               | [GIN] 2024/10/18 - 21:19:41 | 200 |     226.841µs |             ::1 | GET      "/api/ps"
 [ollama]               | [GIN] 2024/10/18 - 21:19:42 | 200 |      60.908µs |             ::1 | GET      "/api/ps"
 INFO:     Started server process [1]
@@ -34,36 +44,24 @@ INFO:     Uvicorn running on http://[::]:5000 (Press CTRL+C to quit)
 ```
 
 To kill the server
-```
+```bash
 docker compose down
 ```
 
-### Docker: Start the Distribution (Single Node CPU)
+## Starting Ollama and Llama Stack separately
 
-> [!NOTE]
-> This will start an ollama server with CPU only, please see [Ollama Documentations](https://github.com/ollama/ollama) for serving models on CPU only.
+If you wish to separately spin up a Ollama server, and connect with Llama Stack, you should use the following commands.
 
-```
-$ cd distributions/ollama/cpu
-$ ls
-compose.yaml  run.yaml
-$ docker compose up
-```
-
-### Conda: ollama run + llama stack run
-
-If you wish to separately spin up a Ollama server, and connect with Llama Stack, you may use the following commands.
-
-#### Start Ollama server.
-- Please check the [Ollama Documentations](https://github.com/ollama/ollama) for more details.
+#### Start Ollama server
+- Please check the [Ollama Documentation](https://github.com/ollama/ollama) for more details.
 
 **Via Docker**
-```
+```bash
 docker run -d -v ollama:/root/.ollama -p 11434:11434 --name ollama ollama/ollama
 ```
 
 **Via CLI**
-```
+```bash
 ollama run <model_id>
 ```
 
@@ -71,7 +69,7 @@ ollama run <model_id>
 
 **Via Conda**
 
-```
+```bash
 llama stack build --template ollama --image-type conda
 llama stack run ./gpu/run.yaml
 ```
@@ -82,7 +80,7 @@ docker run --network host -it -p 5000:5000 -v ~/.llama:/root/.llama -v ./gpu/run
 ```
 
 Make sure in your `run.yaml` file, your inference provider is pointing to the correct Ollama endpoint. E.g.
-```
+```yaml
 inference:
   - provider_id: ollama0
     provider_type: remote::ollama
@@ -96,7 +94,7 @@ inference:
 
 You can use ollama for managing model downloads.
 
-```
+```bash
 ollama pull llama3.1:8b-instruct-fp16
 ollama pull llama3.1:70b-instruct-fp16
 ```
@@ -106,7 +104,7 @@ ollama pull llama3.1:70b-instruct-fp16
 
 
 To serve a new model with `ollama`
-```
+```bash
 ollama run <model_name>
 ```
 
@@ -119,7 +117,7 @@ llama3.1:8b-instruct-fp16    4aacac419454    17 GB    100% GPU     4 minutes fro
 ```
 
 To verify that the model served by ollama is correctly connected to Llama Stack server
-```
+```bash
 $ llama-stack-client models list
 +----------------------+----------------------+---------------+-----------------------------------------------+
 | identifier           | llama_model          | provider_id   | metadata                                      |

docs/source/getting_started/distributions/self_hosted_distro/remote_vllm.md

@@ -0,0 +1,83 @@
+# Remote vLLM Distribution
+
+The `llamastack/distribution-remote-vllm` distribution consists of the following provider configurations.
+
+| **API**           | **Inference**   | **Agents**      | **Memory**                          | **Safety**     	| **Telemetry**  	|
+|-----------------  |---------------- |---------------- |------------------------------------	|----------------	|----------------	|
+| **Provider(s)**   | remote::vllm  	| meta-reference 	| remote::pgvector, remote::chromadb 	| meta-reference 	| meta-reference 	|
+
+You can use this distribution if you have GPUs and want to run an independent vLLM server container for running inference.
+
+## Using Docker Compose
+
+You can use `docker compose` to start a vLLM container and Llama Stack server container together.
+
+> [!NOTE]
+> This assumes you have access to GPU to start a vLLM server with access to your GPU.
+
+```bash
+$ cd distributions/remote-vllm; docker compose up
+```
+
+You will see outputs similar to following ---
+```
+<TO BE FILLED>
+```
+
+To kill the server
+```bash
+docker compose down
+```
+
+## Starting vLLM and Llama Stack separately
+
+You may want to start a vLLM server and connect with Llama Stack manually. There are two ways to start a vLLM server and connect with Llama Stack.
+
+
+#### Start vLLM server.
+
+```bash
+docker run --runtime nvidia --gpus all \
+    -v ~/.cache/huggingface:/root/.cache/huggingface \
+    --env "HUGGING_FACE_HUB_TOKEN=<secret>" \
+    -p 8000:8000 \
+    --ipc=host \
+    vllm/vllm-openai:latest \
+    --model meta-llama/Llama-3.1-8B-Instruct
+```
+
+Please check the [vLLM Documentation](https://docs.vllm.ai/en/v0.5.5/serving/deploying_with_docker.html) for more details.
+
+
+#### Start Llama Stack server pointing to your vLLM server
+
+
+We have provided a template `run.yaml` file in the `distributions/remote-vllm` directory. Please make sure to modify the `inference.provider_id` to point to your vLLM server endpoint. As an example, if your vLLM server is running on `http://127.0.0.1:8000`, your `run.yaml` file should look like the following:
+```yaml
+inference:
+  - provider_id: vllm0
+    provider_type: remote::vllm
+    config:
+      url: http://127.0.0.1:8000
+```
+
+**Via Conda**
+
+If you are using Conda, you can build and run the Llama Stack server with the following commands:
+```bash
+cd distributions/remote-vllm
+llama stack build --template remote_vllm --image-type conda
+llama stack run run.yaml
+```
+
+**Via Docker**
+
+You can use the Llama Stack Docker image to start the server with the following command:
+```bash
+docker run --network host -it -p 5000:5000 \
+  -v ~/.llama:/root/.llama \
+  -v ./gpu/run.yaml:/root/llamastack-run-remote-vllm.yaml \
+  --gpus=all \
+  llamastack/distribution-remote-vllm \
+  --yaml_config /root/llamastack-run-remote-vllm.yaml
+```

docs/source/getting_started/index.md

@@ -80,6 +80,11 @@ Llama3.1-8B-Instruct  Llama3.2-1B                   Llama3.2-3B-Instruct  Llama-
 
 :::
 
+:::{tab-item} vLLM
+##### System Requirements
+Access to Single-Node GPU to start a vLLM server.
+:::
+
 :::{tab-item} tgi
 ##### System Requirements
 Access to Single-Node GPU to start a TGI server.
@@ -119,6 +124,22 @@ docker run -it -p 5000:5000 -v ~/.llama:/root/.llama -v ./run.yaml:/root/my-run.
 ```
 :::
 
+:::{tab-item} vLLM
+```
+$ cd llama-stack/distributions/remote-vllm && docker compose up
+```
+
+The script will first start up vLLM server on port 8000, then start up Llama Stack distribution server hooking up to it for inference. You should see the following outputs --
+```
+<TO BE FILLED>
+```
+
+To kill the server
+```
+docker compose down
+```
+:::
+
 :::{tab-item} tgi
 ```
 $ cd llama-stack/distributions/tgi && docker compose up
@@ -144,7 +165,11 @@ docker compose down
 
 :::{tab-item} ollama
 ```
-$ cd llama-stack/distributions/ollama/cpu && docker compose up
+$ cd llama-stack/distributions/ollama && docker compose up
+
+# OR
+
+$ cd llama-stack/distributions/ollama-gpu && docker compose up
 ```
 
 You will see outputs similar to following ---

llama_stack/providers/registry/inference.py

@@ -45,7 +45,7 @@ def available_providers() -> List[ProviderSpec]:
         ),
         InlineProviderSpec(
             api=Api.inference,
-            provider_type="vllm",
+            provider_type="inline::vllm",
             pip_packages=[
                 "vllm",
             ],

llama_stack/templates/inline-vllm/build.yaml

@@ -0,0 +1,13 @@
+name: meta-reference-gpu
+distribution_spec:
+  docker_image: pytorch/pytorch:2.5.0-cuda12.4-cudnn9-runtime
+  description: Use code from `llama_stack` itself to serve all llama stack APIs
+  providers:
+    inference: meta-reference
+    memory:
+    - meta-reference
+    - remote::chromadb
+    - remote::pgvector
+    safety: meta-reference
+    agents: meta-reference
+    telemetry: meta-reference

llama_stack/templates/remote-vllm/build.yaml

@@ -0,0 +1,12 @@
+name: remote-vllm
+distribution_spec:
+  description: Use (an external) vLLM server for running LLM inference
+  providers:
+    inference: remote::vllm
+    memory:
+    - meta-reference
+    - remote::chromadb
+    - remote::pgvector
+    safety: meta-reference
+    agents: meta-reference
+    telemetry: meta-reference

llama_stack/templates/vllm/build.yaml

@@ -1,9 +0,0 @@
-name: vllm
-distribution_spec:
-  description: Like local, but use vLLM for running LLM inference
-  providers:
-    inference: vllm
-    memory: meta-reference
-    safety: meta-reference
-    agents: meta-reference
-    telemetry: meta-reference