github/home-llm
1
1
Fork 0
mirror of https://github.com/acon96/home-llm.git synced 2026-01-08 21:28:05 -05:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'develop' into feature/dataset-customization

Browse Source
This commit is contained in:
Alex O'Connell
2024-02-22 21:20:59 -05:00
parent d22ef571b8 a78e57031c
commit fc171ac04d
3 changed files with 221 additions and 97 deletions
Download Patch File Download Diff File Expand all files Collapse all files

116
README.md
View File

@@ -1,6 +1,20 @@
# Home LLM
This project provides the required "glue" components to control your Home Assistant installation with a completely local Large Language Model acting as a personal assistant. The goal is to provide a drop in solution to be used as a "conversation agent" component by Home Assistant.
## Quick Start
Please see the [Setup Guide](./docs/Setup.md) for more information on installation.
## Home Assistant Component
In order to integrate with Home Assistant, we provide a `custom_component` that exposes the locally running LLM as a "conversation agent".
This component can be interacted with in a few ways:
- using a chat interface so you can chat with it.
- integrating with Speech-to-Text and Text-to-Speech addons so you can just speak to it.
The component can either run the model directly as part of the Home Assistant software using llama-cpp-python, or you can run the [oobabooga/text-generation-webui](https://github.com/oobabooga/text-generation-webui) project to provide access to the LLM via an API interface.
When doing this, you can host the model yourself and point the add-on at machine where the model is hosted, or you can run the model using text-generation-webui using the provided [custom Home Assistant add-on](./addon).
## Model
The "Home" models are a fine tuning of the Phi model series from Microsoft and the StableLM model series from StabilityAI. The model is able to control devices in the user's house as well as perform basic question and answering. The fine tuning dataset is a [custom synthetic dataset](./data) designed to teach the model function calling based on the device information in the context.
@@ -18,8 +32,6 @@ The latest models can be found on HuggingFace:
</details>
Make sure you have `llama-cpp-python>=0.2.29` in order to run these models.
The model is quantized using Llama.cpp in order to enable running the model in super low resource environments that are common with Home Assistant installations such as Raspberry Pis.
The model can be used as an "instruct" type model using the [ChatML](https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/ai-services/openai/includes/chat-markup-language.md) or [Zephyr](https://huggingface.co/HuggingFaceH4/zephyr-7b-alpha/discussions/3) prompt format (depends on the model). The system prompt is used to provide information about the state of the Home Assistant installation including available devices and callable services.
@@ -41,7 +53,7 @@ Output from the model will consist of a response that should be relayed back to
`````
turning on the kitchen lights for you now
```homeassistant
```Home Assistant
{ "service": "light.turn_on", "target_device": "light.kitchen" }
`````
@@ -70,7 +82,7 @@ The 3B model was trained as a LoRA on an RTX 3090 (24GB) using the following set
<details>
<summary>Training Arguments</summary>
```
```console
python3 train.py \
--run_name home-3b \
--base_model microsoft/phi-2 \
@@ -93,7 +105,7 @@ The 1B model was trained as a full fine-tuning on on an RTX 3090 (24GB). Trainin
<details>
<summary>Training Arguments</summary>
```
```console
python3 train.py \
--run_name home-1b \
--base_model microsoft/phi-1_5 \
@@ -109,99 +121,9 @@ python3 train.py \
</details>
<br/>
## Home Assistant Component
In order to integrate with Home Assistant, we provide a `custom_component` that exposes the locally running LLM as a "conversation agent" that can be interacted with using a chat interface as well as integrate with Speech-to-Text and Text-to-Speech addons to enable interacting with the model by speaking.
## Home Assistant Addon
In order to facilitate running the project entirely on the system where Home Assistant is installed, there is an experimental Home Assistant Add-on that runs the oobabooga/text-generation-webui to connect to using the "remote" backend options. The addon can be found in the [addon/](./addon/README.md) directory.
The component can either run the model directly as part of the Home Assistant software using llama-cpp-python, or you can run the [oobabooga/text-generation-webui](https://github.com/oobabooga/text-generation-webui) project to provide access to the LLM via an API interface. When doing this, you can host the model yourself and point the add-on at machine where the model is hosted, or you can run the model using text-generation-webui using the provided [custom Home Assistant add-on](./addon).
When running the model locally with Llama.cpp, the component also constrains the model output using a GBNF grammar. This forces the model to provide valid output no matter what because it is constrained to outputting valid JSON every time. This helps the model perform significantly better at lower quantization levels where it would generate syntax errors previously. See [output.gbnf](./custom_components/llama_conversation/output.gbnf) for the existing grammar.
### Installing with HACS
You can use this button to add the repository to HACS and open the download page
[![Open your Home Assistant instance and open a repository inside the Home Assistant Community Store.](https://my.home-assistant.io/badges/hacs_repository.svg)](https://my.home-assistant.io/redirect/hacs_repository/?category=Integration&repository=home-llm&owner=acon96)
### Installing Manually
1. Ensure you have either the Samba, SSH, FTP, or another add-on installed that gives you access to the `config` folder
2. If there is not already a `custom_components` folder, create one now.
3. Copy the `custom_components/llama_conversation` folder from this repo to `config/custom_components/llama_conversation` on your Home Assistant machine.
4. Restart Home Assistant using the "Developer Tools" tab -> Services -> Run `homeassistant.restart`
5. The "LLaMA Conversation" integration should show up in the "Devices" section now.
### Setting up
When setting up the component, there are 4 different "backend" options to choose from:
1. Llama.cpp with a model from HuggingFace
2. Llama.cpp with a locally provided model
3. A remote instance of text-generation-webui
4. A generic OpenAI API compatible interface; *should* be compatible with LocalAI, LM Studio, and all other OpenAI compatible backends
See [docs/Backend Configuration.md](/docs/Backend%20Configuration.md) for more info.
**Installing llama-cpp-python for local model usage**:
In order to run a model directly as part of your Home Assistant installation, you will need to install one of the pre-build wheels because there are no existing musllinux wheels for the package. Compatible wheels for x86_x64 and arm64 are provided in the [dist](./dist) folder. Copy the `*.whl` files to the `custom_components/llama_conversation/` folder. They will be installed while setting up the component.
NOTE: Home Assistant recently moved from Python 3.11 to Python 3.12. If you are using HA 2024.1.4 and prior, use the `cp311` wheels and if you are using HA 2024.2.1 or newer then use the `cp312` wheels.
**Setting up the Llama.cpp backend with a model from HuggingFace**:
You need the following settings to configure the local backend from HuggingFace:
1. Model Name: the name of the model in the form `repo/model-name`. The repo MUST contain a GGUF quantized model.
2. Model Quantization: The quantization level to download. Pick from the list. Higher quantizations use more RAM but have higher quality responses.
**Setting up the Llama.cpp backend with a locally downloaded model**:
You need the following settings to configure the local backend from HuggingFace:
1. Model File Name: the file name where Home Assistant can access the model to load. Most likely a sub-path of `/config` or `/media` or wherever you copied the model file to.
**Setting up the "remote" backends**:
You need the following settings in order to configure the "remote" backend:
1. Hostname: the host of the machine where text-generation-webui API is hosted. If you are using the provided add-on then the hostname is `local-text-generation-webui` or `f459db47-text-generation-webui` depending on how the addon was installed.
2. Port: the port for accessing the text-generation-webui API. NOTE: this is not the same as the UI port. (Usually 5000)
3. Name of the Model: This name must EXACTLY match the name as it appears in `text-generation-webui`
With the remote text-generation-webui backend, the component will validate that the selected model is available for use and will ensure it is loaded remotely. The Generic OpenAI compatible version does NOT do any validation or model loading.
**Setting up with LocalAI**:
If you are an existing LocalAI user or would like to use LocalAI as your backend, please refer to [this](https://io.midori-ai.xyz/howtos/setup-with-ha/) website which has instructions on how to setup LocalAI to work with Home-LLM including automatic installation of the latest version of the the Home-LLM model. The auto-installer (LocalAI Manager) will automatically download and setup LocalAI and/or the model of your choice and automatically create the necessary template files for the model to work with this integration.
### Configuring the component as a Conversation Agent
**NOTE: ANY DEVICES THAT YOU SELECT TO BE EXPOSED TO THE MODEL WILL BE ADDED AS CONTEXT AND POTENTIALLY HAVE THEIR STATE CHANGED BY THE MODEL. ONLY EXPOSE DEVICES THAT YOU ARE OK WITH THE MODEL MODIFYING THE STATE OF, EVEN IF IT IS NOT WHAT YOU REQUESTED. THE MODEL MAY OCCASIONALLY HALLUCINATE AND ISSUE COMMANDS TO THE WRONG DEVICE! USE AT YOUR OWN RISK.**
In order to utilize the conversation agent in HomeAssistant:
1. Navigate to "Settings" -> "Voice Assistants"
2. Select "+ Add Assistant"
3. Name the assistant whatever you want.
4. Select the "Conversation Agent" that we created previously
5. If using STT or TTS configure these now
6. Return to the "Overview" dashboard and select chat icon in the top left.
From here you can submit queries to the AI agent.
In order for any entities be available to the agent, you must "expose" them first.
1. Navigate to "Settings" -> "Voice Assistants" -> "Expose" Tab
2. Select "+ Expose Entities" in the bottom right
3. Check any entities you would like to be exposed to the conversation agent.
### Running the text-generation-webui add-on
In order to facilitate running the project entirely on the system where Home Assistant is installed, there is an experimental Home Assistant Add-on that runs the oobabooga/text-generation-webui to connect to using the "remote" backend option.
You can use this button to automatically download and build the addon for `oobabooga/text-generation-webui`
[![Open your Home Assistant instance and show the dashboard of an add-on.](https://my.home-assistant.io/badges/supervisor_addon.svg)](https://my.home-assistant.io/redirect/supervisor_addon/?addon=f459db47_text-generation-webui&repository_url=https%3A%2F%2Fgithub.com%2Facon96%2Fhome-llm)
If the automatic installation fails then you can install the addon manually using the following steps:
1. Ensure you have either the Samba, SSH, FTP, or another add-on installed that gives you access to the `addons` folder
2. Copy the `addon` folder from this repo to `addons/text-generation-webui` on your Home Assistant machine.
3. Go to the "Add-ons" section in settings and then pick the "Add-on Store" from the bottom right corner.
4. Select the 3 dots in the top right and click "Check for Updates" and Refresh the webpage.
5. There should now be a "Local Add-ons" section at the top of the "Add-on Store"
6. Install the `oobabooga-text-generation-webui` add-on. It will take ~15-20 minutes to build the image on a Raspberry Pi.
7. Copy any models you want to use to the `addon_configs/local_text-generation-webui/models` folder or download them using the UI.
8. Load up a model to use. NOTE: The timeout for ingress pages is only 60 seconds so if the model takes longer than 60 seconds to load (very likely) then the UI will appear to time out and you will need to navigate to the add-on's logs to see when the model is fully loaded.
### Performance of running the model on a Raspberry Pi
The RPI4 4GB that I have was sitting right at 1.5 tokens/sec for prompt eval and 1.6 tokens/sec for token generation when running the `Q4_K_M` quant. I was reliably getting responses in 30-60 seconds after the initial prompt processing which took almost 5 minutes. It depends significantly on the number of devices that have been exposed as well as how many states have changed since the last invocation because llama.cpp caches KV values for identical prompt prefixes.
It is highly recommend to set up text-generation-webui on a separate machine that can take advantage of a GPU.
## Version History
| Version | Description |

36
docs/Performance.md Normal file
View File

@@ -0,0 +1,36 @@
### Performance of running the model on a Raspberry Pi
The RPI4 4GB that I have was sitting right at 1.5 tokens/sec for prompt eval and 1.6 tokens/sec for token generation when running the `Q4_K_M` quant. I was reliably getting responses in 30-60 seconds after the initial prompt processing which took almost 5 minutes. It depends significantly on the number of devices that have been exposed as well as how many states have changed since the last invocation because llama.cpp caches KV values for identical prompt prefixes.
It is highly recommend to set up text-generation-webui on a separate machine that can take advantage of a GPU.
# Home 1B V2 GGUF Q4_K_M RPI5
christmas.txt
llama_print_timings: load time = 678.37 ms
llama_print_timings: sample time = 16.38 ms / 45 runs ( 0.36 ms per token, 2747.09 tokens per second)
llama_print_timings: prompt eval time = 31356.56 ms / 487 tokens ( 64.39 ms per token, 15.53 tokens per second)
llama_print_timings: eval time = 4868.37 ms / 44 runs ( 110.64 ms per token, 9.04 tokens per second)
llama_print_timings: total time = 36265.33 ms / 531 tokens
climate.txt
llama_print_timings: load time = 613.87 ms
llama_print_timings: sample time = 20.62 ms / 55 runs ( 0.37 ms per token, 2667.96 tokens per second)
llama_print_timings: prompt eval time = 27324.34 ms / 431 tokens ( 63.40 ms per token, 15.77 tokens per second)
llama_print_timings: eval time = 5780.72 ms / 54 runs ( 107.05 ms per token, 9.34 tokens per second)
llama_print_timings: total time = 33152.48 ms / 485 tokens
# Home 3B V2 GGUF Q4_K_M RPI5
climate.txt
llama_print_timings: load time = 1179.64 ms
llama_print_timings: sample time = 19.25 ms / 52 runs ( 0.37 ms per token, 2702.00 tokens per second)
llama_print_timings: prompt eval time = 52688.82 ms / 431 tokens ( 122.25 ms per token, 8.18 tokens per second)
llama_print_timings: eval time = 10206.12 ms / 51 runs ( 200.12 ms per token, 5.00 tokens per second)
llama_print_timings: total time = 62942.85 ms / 482 tokens
sonnet.txt
llama_print_timings: load time = 1076.44 ms
llama_print_timings: sample time = 1225.34 ms / 236 runs ( 5.19 ms per token, 192.60 tokens per second)
llama_print_timings: prompt eval time = 60754.40 ms / 490 tokens ( 123.99 ms per token, 8.07 tokens per second)
llama_print_timings: eval time = 44885.82 ms / 213 runs ( 210.73 ms per token, 4.75 tokens per second)
llama_print_timings: total time = 107127.16 ms / 703 tokens

166
docs/Setup.md Normal file
View File

@@ -0,0 +1,166 @@
# Setup Instructions
1. [Home Assistant Component](#home-assistant-component)
2. [Configuring the LLM as a Conversation Agent](#configuring-as-a-conversation-agent)
3. [Setting up the text-generation-webui Addon](#text-generation-webui-add-on)
## Home Assistant Component
### Requirements
- A supported version of Home Assistant; `2023.10.0` or newer
- SSH or Samba access to your Home Assistant instance
**Optional:**
- [HACs](https://hacs.xyz/docs/setup/download/) (if you want to install it that way)
### 💾 🚕 Install the Home Assistant Component with HACs
> 🛑 ✋🏻 Requires HACs
>
> First make sure you have [HACs installed](https://hacs.xyz/docs/setup/download/)
Once you have HACs installed, this button will help you add the repository to HACS and open the download page
[![Open your Home Assistant instance and open a repository inside the Home Assistant Community Store.](https://my.home-assistant.io/badges/hacs_repository.svg)](https://my.home-assistant.io/redirect/hacs_repository/?category=Integration&repository=home-llm&owner=acon96)
**Remember to restart Home Assistant after installing the component!**
A "LLaMA Conversation" device should show up in the `Settings > Devices and Services > [Devices]` tab now:
![image](https://github.com/acon96/home-llm/assets/61225/4427e362-e443-4796-bee8-5bdda18305d0)
### 💾 🔨 Install the Home Assistant Component Manually
1. Ensure you have either the Samba, SSH, FTP, or another add-on installed that gives you access to the `config` folder
2. If there is not already a `custom_components` folder, create one now.
3. Copy the `custom_components/llama_conversation` folder from this repo to `config/custom_components/llama_conversation` on your Home Assistant machine.
4. Restart Home Assistant: `Developer Tools -> Services -> Run` : `HomeAssistant.restart`
A "LLaMA Conversation" device should show up in the `Settings > Devices and Services > [Devices]` tab now:
![image](https://github.com/acon96/home-llm/assets/61225/4427e362-e443-4796-bee8-5bdda18305d0)
### ⚙️ Configuration and Setup
You must configure at least one model by configuring the integration.
1. `Settings > Devices and Services`.
2. Click the `Add Integration` button in the bottom right of the screen.
3. Filter the list of "brand names" for llama, and "LLaMa Conversation" should remain.
4. Choose the backend you will be using to host the model:
1. Using builtin llama.cpp with hugging face
2. Using builtin llama.cpp with existing model file
3. using text-generation-webui api
4. using generic openapi compatiable api
5. using ollama api
### llama-cpp-python Wheel Installation
If you plan on running the model locally on the same hardware as your Home Assistant server, then the recommended way to run the model is to use Llama.cpp. Unfortunately there are not pre-build wheels for this package for the musllinux runtime that Home Assistant Docker images use. To get around this, we provide compatible wheels for x86_x64 and arm64 in the [dist](./dist) folder.
Download the `*.whl` file that matches your hardware and then copy the `*.whl` file to the `custom_components/llama_conversation/` folder. It will be installed as a configuration step while setting up the Home Assistant component.
| wheel | platform | home assistant version |
| --- | --- | --- |
| llama_cpp_python-{version}-cp311-cp311-musllinux_1_2_aarch64.whl | aarch64 (RPi 4 and 5) | `2024.1.4` and older |
| llama_cpp_python-{version}-cp311-cp311-musllinux_1_2_x86_64.whl | x86_64 (Intel + AMD) | `2024.1.4` and older |
| llama_cpp_python-{version}-cp312-cp312-musllinux_1_2_aarch64.whl | aarch64 (RPi 4 and 5) | `2024.2.0` and newer |
| llama_cpp_python-{version}-cp312-cp312-musllinux_1_2_x86_64.whl | x86_64 (Intel + AMD) | `2024.2.0` and newer |
### Constrained Grammar
When running the model locally with [Llama.cpp], the component also constrains the model output using a GBNF grammar.
This forces the model to provide valid output no matter what since its outputs are constrained to valid JSON every time.
This helps the model perform significantly better at lower quantization levels where it would previously generate syntax errors. It is recommended to turn this on when using the component as it will reduce the incorrect output from the model.
For more information See [output.gbnf](./custom_components/llama_conversation/output.gbnf) for the existing grammar.
### Backend Configuration
![image](https://github.com/airtonix/home-llm/assets/61225/6f5d9748-5bfc-47ce-8abc-4f07d389a73f)
When setting up the component, there are 5 different "backend" options to choose from:
a. Llama.cpp with a model from HuggingFace
b. Llama.cpp with a locally provided model
c. A remote instance of text-generation-webui
d. A generic OpenAI API compatible interface; *should* be compatible with LocalAI, LM Studio, and all other OpenAI compatible backends
e. Ollama api
See [docs/Backend Configuration.md](/docs/Backend%20Configuration.md) for more info.
#### Llama.cpp Backend with a model from HuggingFace
This is option A
You need the following settings to configure the local backend from HuggingFace:
1. **Model Name**: the name of the model in the form `repo/model-name`. The repo MUST contain a GGUF quantized model.
2. **Model Quantization**: The quantization level to download. Pick from the list. Higher quantizations use more RAM but have higher quality responses.
#### Llama.cpp Backend with a locally downloaded model
This is option B
You need the following settings to configure the local backend from HuggingFace:
1. **Model File Name**: the file name where Home Assistant can access the model to load. Most likely a sub-path of `/config` or `/media` or wherever you copied the model file to.
#### Remote Backends
This is options C, D and E
You need the following settings in order to configure the "remote" backend:
1. **Hostname**: the host of the machine where text-generation-webui API is hosted. If you are using the provided add-on then the hostname is `local-text-generation-webui` or `f459db47-text-generation-webui` depending on how the addon was installed.
2. **Port**: the port for accessing the text-generation-webui API. NOTE: this is not the same as the UI port. (Usually 5000)
3. **Name of the Model**: This name must EXACTLY match the name as it appears in `text-generation-webui`
With the remote text-generation-webui backend, the component will validate that the selected model is available for use and will ensure it is loaded remotely. The Generic OpenAI compatible version does NOT do any validation or model loading.
**Setting up with LocalAI**:
If you are an existing LocalAI user or would like to use LocalAI as your backend, please refer to [this](https://io.midori-ai.xyz/howtos/setup-with-ha/) website which has instructions on how to setup LocalAI to work with Home-LLM including automatic installation of the latest version of the the Home-LLM model. The auto-installer (LocalAI Manager) will automatically download and setup LocalAI and/or the model of your choice and automatically create the necessary template files for the model to work with this integration.
## Configuring as a Conversation Agent
> 🛑 ✋🏻 Security Warning
>
> Any devices that you select to be exposed to the model will be added as
> context and potentially have their state changed by the model.
>
> Only expose devices that you want the model modifying the state of.
>
> The model may occasionally hallucinate and issue commands to the wrong device!
>
> Use At Your Own Risk
1. Navigate to `Settings` -> `Voice Assistants`
2. Select `+ Add Assistant`
3. Name the assistant whatever you want.
4. Select the conversation agent that we created previously.
5. If using STT or TTS configure these now
6. Return to the "Overview" dashboard and select chat icon in the top left.
7. From here you can submit queries to the AI agent.
In order for any entities be available to the agent, you must "expose" them first.
1. Navigate to "Settings" -> "Voice Assistants" -> "Expose" Tab
2. Select "+ Expose Entities" in the bottom right
3. Check any entities you would like to be exposed to the conversation agent.
> Note:
> When exposing entities to the model, you are adding tokens to the model's context. If you exceed the context length of the model, then your interactions with the model will fail due to the instructions being dropped out of the context's sliding window.
> It is recommended to only expose a maximum of 32 entities to this conversation agent at this time.
## text-generation-webui add-on
You can use this button to automatically download and build the addon for `oobabooga/text-generation-webui`
[![Open your Home Assistant instance and show the dashboard of an add-on.](https://my.home-assistant.io/badges/supervisor_addon.svg)](https://my.home-assistant.io/redirect/supervisor_addon/?addon=f459db47_text-generation-webui&repository_url=https%3A%2F%2Fgithub.com%2Facon96%2Fhome-llm)
If the automatic installation fails then you can install the addon manually using the following steps:
1. Ensure you have either the Samba, SSH, FTP, or another add-on installed that gives you access to the `addons` folder
2. Copy the `addon` folder from this repo to `addons/text-generation-webui` on your Home Assistant machine.
3. Go to the "Add-ons" section in settings and then pick the "Add-on Store" from the bottom right corner.
4. Select the 3 dots in the top right and click "Check for Updates" and Refresh the webpage.
5. There should now be a "Local Add-ons" section at the top of the "Add-on Store"
6. Install the `oobabooga-text-generation-webui` add-on. It will take ~15-20 minutes to build the image on a Raspberry Pi.
7. Copy any models you want to use to the `addon_configs/local_text-generation-webui/models` folder or download them using the UI.
8. Load up a model to use. NOTE: The timeout for ingress pages is only 60 seconds so if the model takes longer than 60 seconds to load (very likely) then the UI will appear to time out and you will need to navigate to the add-on's logs to see when the model is fully loaded.