Instructions to use selorahomes/Selora-AI with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use selorahomes/Selora-AI with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="selorahomes/Selora-AI")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("selorahomes/Selora-AI", dtype="auto")

llama-cpp-python

How to use selorahomes/Selora-AI with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="selorahomes/Selora-AI",
	filename="qwen3_17b_base.Q6_K.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use selorahomes/Selora-AI with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf selorahomes/Selora-AI:Q6_K
# Run inference directly in the terminal:
llama-cli -hf selorahomes/Selora-AI:Q6_K

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf selorahomes/Selora-AI:Q6_K
# Run inference directly in the terminal:
llama-cli -hf selorahomes/Selora-AI:Q6_K

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf selorahomes/Selora-AI:Q6_K
# Run inference directly in the terminal:
./llama-cli -hf selorahomes/Selora-AI:Q6_K

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf selorahomes/Selora-AI:Q6_K
# Run inference directly in the terminal:
./build/bin/llama-cli -hf selorahomes/Selora-AI:Q6_K

Use Docker

docker model run hf.co/selorahomes/Selora-AI:Q6_K

LM Studio
Jan

vLLM

How to use selorahomes/Selora-AI with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "selorahomes/Selora-AI"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "selorahomes/Selora-AI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/selorahomes/Selora-AI:Q6_K

SGLang

How to use selorahomes/Selora-AI with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "selorahomes/Selora-AI" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "selorahomes/Selora-AI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "selorahomes/Selora-AI" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "selorahomes/Selora-AI",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use selorahomes/Selora-AI with Ollama:
```
ollama run hf.co/selorahomes/Selora-AI:Q6_K
```

Unsloth Studio

How to use selorahomes/Selora-AI with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for selorahomes/Selora-AI to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for selorahomes/Selora-AI to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for selorahomes/Selora-AI to start chatting

How to use selorahomes/Selora-AI with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf selorahomes/Selora-AI:Q6_K

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "selorahomes/Selora-AI:Q6_K"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use selorahomes/Selora-AI with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf selorahomes/Selora-AI:Q6_K

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default selorahomes/Selora-AI:Q6_K

Run Hermes

hermes

Docker Model Runner
How to use selorahomes/Selora-AI with Docker Model Runner:
```
docker model run hf.co/selorahomes/Selora-AI:Q6_K
```

Lemonade

How to use selorahomes/Selora-AI with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull selorahomes/Selora-AI:Q6_K

Run and chat with the model

lemonade run user.Selora-AI-Q6_K

List all available models

lemonade list

lafoush commited on 18 days ago

Commit

7ace5ac

verified ·

1 Parent(s): bb30ff0

Publish selora-ai-local 0.4.2

Browse files

Files changed (18) hide show

.gitattributes +4 -0
Modelfile.answers +31 -11
Modelfile.automations +6 -6
Modelfile.clarifications +6 -6
Modelfile.commands +6 -6
README.md +28 -21
manifest.json +28 -32
prompts/answers.txt +23 -5
qwen25_15b_base.Q4_K_M.gguf +0 -3
qwen25_15b_clarification.lora.gguf +0 -3
qwen3_17b_automation.lora.gguf +0 -3
qwen3_17b_base.IQ4_XS.gguf +0 -3
qwen3_17b_clarification.lora.gguf +0 -3
qwen3_17b_command.lora.gguf +0 -3
qwen25_15b_answer.lora.gguf → selora-v042-answer.f16.gguf +2 -2
qwen25_15b_automation.lora.gguf → selora-v042-automation.f16.gguf +2 -2
qwen3_17b_answer.lora.gguf → selora-v042-clarification.f16.gguf +1 -1
qwen25_15b_command.lora.gguf → selora-v042-command.f16.gguf +2 -2

.gitattributes CHANGED Viewed

@@ -44,3 +44,7 @@ qwen3_17b_base.IQ4_XS.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3_17b_base.f16.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3_17b_clarification.lora.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3_17b_command.lora.gguf filter=lfs diff=lfs merge=lfs -text

 qwen3_17b_base.f16.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3_17b_clarification.lora.gguf filter=lfs diff=lfs merge=lfs -text
 qwen3_17b_command.lora.gguf filter=lfs diff=lfs merge=lfs -text
+selora-v042-answer.f16.gguf filter=lfs diff=lfs merge=lfs -text
+selora-v042-automation.f16.gguf filter=lfs diff=lfs merge=lfs -text
+selora-v042-clarification.f16.gguf filter=lfs diff=lfs merge=lfs -text
+selora-v042-command.f16.gguf filter=lfs diff=lfs merge=lfs -text

Modelfile.answers CHANGED Viewed

@@ -1,29 +1,49 @@
-# Ollama Modelfile for SeloraAI-Local / answer specialist (Qwen 2.5 1.5B)
 # Build:  ollama create selora-qwen-answer -f Modelfile.answers
 # Run:    ollama run selora-qwen-answer
-FROM ../qwen25_15b_base.f16.gguf
-ADAPTER ../qwen25_15b_answer.lora.gguf
-# Qwen 2.5 chat template (ChatML)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
-{{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
-# Trained per-specialist system prompt (matches v2 training data)
 SYSTEM """You are Selora AI, a home automation assistant on Home Assistant. You CAN: control lights/climate/locks/switches, run scripts and scenes, set timers and reminders via timer/input_datetime entities, query device states, and create automations on request. Never say you are a "text-based AI" or that you cannot do something Home Assistant supports — describe how you would do it instead.
-Return ONE JSON object:
 {"intent":"answer","response":"<1-3 sentences>"}
 RULES:
-- Answer the user's question directly. No preamble ("Sure!", "Great question!").
-- 1-3 sentences. Add detail only if the user asked for it.
-- If the question is about home state, ground the answer in AVAILABLE ENTITIES.
-- If the user asks what you can do, list 2-4 concrete capabilities (control devices, set timers, build automations, summarize home state) — not generic phrases.
 - Output ONLY the JSON object."""
 # Generation params — matches what the integration sends + repeat_penalty for Qwen

+# Ollama Modelfile for SeloraAI-Local / answer specialist (Qwen3 1.7B)
 # Build:  ollama create selora-qwen-answer -f Modelfile.answers
 # Run:    ollama run selora-qwen-answer
+FROM ../qwen3_17b_base.IQ4_XS.gguf
+ADAPTER ../qwen3_17b_answer.lora.gguf
+# Qwen3 chat template (ChatML, /no_think to suppress reasoning blocks for
+# short structured JSON output)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
+/no_think {{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
+# Trained per-specialist system prompt (matches current training data,
+# includes the query_state tool envelope).
 SYSTEM """You are Selora AI, a home automation assistant on Home Assistant. You CAN: control lights/climate/locks/switches, run scripts and scenes, set timers and reminders via timer/input_datetime entities, query device states, and create automations on request. Never say you are a "text-based AI" or that you cannot do something Home Assistant supports — describe how you would do it instead.
+Return ONE JSON object using one of these envelope shapes:
+ANSWER — for conversational questions, recommendations, or when AVAILABLE ENTITIES already has the full answer:
 {"intent":"answer","response":"<1-3 sentences>"}
+QUERY_STATE — for live state queries that need filtering by state/attribute:
+{"intent":"query_state","calls":[{"tool":"query_state","args":{"domain":"<domain>","filter":{"state":"<value>"}}}]}
+TOOL SCHEMA:
+- tool: "query_state"
+- args:
+    domain (str, required): HA domain — light/switch/lock/cover/fan/media_player/climate/binary_sensor/sensor/person/device_tracker
+    filter (dict, optional):
+      state (str): match exact state ("on", "off", "locked", "open", "home", ...)
+      entity_id (str): match a specific entity_id
+      device_class (str): match HA device_class ("door", "window", "motion", ...)
+      attribute (dict): match attribute key/value (e.g. {"hvac_mode": "heat"})
+WHEN TO USE EACH:
+- query_state for "what's on?", "is X locked?", "how many windows are open?", "which thermostats are heating?".
+- answer for "what can you do?", "explain X", or when the catalog already gives a complete 1-3 sentence answer ("am I home?" → check person entity).
 RULES:
+- 1-3 sentences for answer. Add detail only if the user asked for it.
+- Ground answer responses in AVAILABLE ENTITIES — name actual friendly_names and current state values.
+- When naming a specific device in an answer, wrap its friendly_name in entity markers like [[entity:light.kitchen|Kitchen Lights]] so the panel renders it as a live tile.
 - Output ONLY the JSON object."""
 # Generation params — matches what the integration sends + repeat_penalty for Qwen

Modelfile.automations CHANGED Viewed

@@ -1,19 +1,19 @@
-# Ollama Modelfile for SeloraAI-Local / automation specialist (Qwen 2.5 1.5B)
 # Build:  ollama create selora-qwen-automation -f Modelfile.automations
 # Run:    ollama run selora-qwen-automation
-FROM ../qwen25_15b_base.f16.gguf
-ADAPTER ../qwen25_15b_automation.lora.gguf
-# Qwen 2.5 chat template (ChatML)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
-{{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
-# Trained per-specialist system prompt (matches v2 training data)
 SYSTEM """You are Selora AI, an automation architect for Home Assistant. The user wants a recurring rule, schedule, or multi-step sequence saved as an automation.
 Return ONE JSON object with this shape and nothing else:

+# Ollama Modelfile for SeloraAI-Local / automation specialist (Qwen3 1.7B)
 # Build:  ollama create selora-qwen-automation -f Modelfile.automations
 # Run:    ollama run selora-qwen-automation
+FROM ../qwen3_17b_base.IQ4_XS.gguf
+ADAPTER ../qwen3_17b_automation.lora.gguf
+# Qwen3 chat template (ChatML, /no_think to suppress reasoning)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
+/no_think {{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
+# Trained per-specialist system prompt (matches current training data)
 SYSTEM """You are Selora AI, an automation architect for Home Assistant. The user wants a recurring rule, schedule, or multi-step sequence saved as an automation.
 Return ONE JSON object with this shape and nothing else:

Modelfile.clarifications CHANGED Viewed

@@ -1,19 +1,19 @@
-# Ollama Modelfile for SeloraAI-Local / clarification specialist (Qwen 2.5 1.5B)
 # Build:  ollama create selora-qwen-clarification -f Modelfile.clarifications
 # Run:    ollama run selora-qwen-clarification
-FROM ../qwen25_15b_base.f16.gguf
-ADAPTER ../qwen25_15b_clarification.lora.gguf
-# Qwen 2.5 chat template (ChatML)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
-{{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
-# Trained per-specialist system prompt (matches v2 training data)
 SYSTEM """You are Selora AI on Home Assistant. The user's request is ambiguous and you need ONE focused follow-up question to disambiguate.
 Return ONE JSON object:

+# Ollama Modelfile for SeloraAI-Local / clarification specialist (Qwen3 1.7B)
 # Build:  ollama create selora-qwen-clarification -f Modelfile.clarifications
 # Run:    ollama run selora-qwen-clarification
+FROM ../qwen3_17b_base.IQ4_XS.gguf
+ADAPTER ../qwen3_17b_clarification.lora.gguf
+# Qwen3 chat template (ChatML, /no_think to suppress reasoning)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
+/no_think {{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
+# Trained per-specialist system prompt (matches current training data)
 SYSTEM """You are Selora AI on Home Assistant. The user's request is ambiguous and you need ONE focused follow-up question to disambiguate.
 Return ONE JSON object:

Modelfile.commands CHANGED Viewed

@@ -1,19 +1,19 @@
-# Ollama Modelfile for SeloraAI-Local / command specialist (Qwen 2.5 1.5B)
 # Build:  ollama create selora-qwen-command -f Modelfile.commands
 # Run:    ollama run selora-qwen-command
-FROM ../qwen25_15b_base.f16.gguf
-ADAPTER ../qwen25_15b_command.lora.gguf
-# Qwen 2.5 chat template (ChatML)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
-{{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
-# Trained per-specialist system prompt (matches v2 training data)
 SYSTEM """You are Selora AI, controlling devices on a Home Assistant instance. The user wants an immediate action.
 Return ONE JSON object with this shape and nothing else:

+# Ollama Modelfile for SeloraAI-Local / command specialist (Qwen3 1.7B)
 # Build:  ollama create selora-qwen-command -f Modelfile.commands
 # Run:    ollama run selora-qwen-command
+FROM ../qwen3_17b_base.IQ4_XS.gguf
+ADAPTER ../qwen3_17b_command.lora.gguf
+# Qwen3 chat template (ChatML, /no_think to suppress reasoning)
 TEMPLATE """{{ if .System }}<|im_start|>system
 {{ .System }}<|im_end|>
 {{ end }}{{ if .Prompt }}<|im_start|>user
+/no_think {{ .Prompt }}<|im_end|>
 {{ end }}<|im_start|>assistant
 """
+# Trained per-specialist system prompt (matches current training data)
 SYSTEM """You are Selora AI, controlling devices on a Home Assistant instance. The user wants an immediate action.
 Return ONE JSON object with this shape and nothing else:

README.md CHANGED Viewed

@@ -1,14 +1,15 @@
 ---
 license: apache-2.0
-base_model: Qwen/Qwen2.5-1.5B-Instruct
 tags:
   - text-generation
   - qwen
-  - qwen2.5
   - lora
   - home-assistant
   - home-automation
   - smart-home
 language:
   - en
 library_name: transformers
@@ -17,8 +18,10 @@ pipeline_tag: text-generation
 # Selora AI
-Qwen 2.5 1.5B fine-tuned for Home Assistant with four specialist LoRA
-adapters. Used by the [Selora AI Home Assistant
 integration](https://gitlab.com/selorahomes/products/selora-ai/ha-integration);
 also runnable directly via Ollama, llama.cpp, or vLLM.
@@ -54,12 +57,12 @@ are also published as separate Ollama models.
 ```bash
 llama-server \
-  --model qwen25_15b_base.Q4_K_M.gguf \
   --lora-init-without-apply \
-  --lora qwen25_15b_command.lora.gguf \
-  --lora qwen25_15b_automation.lora.gguf \
-  --lora qwen25_15b_answer.lora.gguf \
-  --lora qwen25_15b_clarification.lora.gguf \
   --ctx-size 8192
 ```
@@ -70,7 +73,7 @@ POST to `/lora-adapters` to switch the active LoRA before each
 ```bash
 python -m vllm.entrypoints.openai.api_server \
-  --model ./qwen25_15b_hf \
   --enable-lora --max-loras 4 --max-lora-rank 32 \
   --lora-modules \
     selora-v1-commands=/path/to/peft/command \
@@ -98,12 +101,16 @@ Bump `max_tokens` to 1536 for automation requests (longer JSON output).
 ## Training
-Base: [Qwen 2.5 1.5B Instruct](https://huggingface.co/Qwen/Qwen2.5-1.5B-Instruct)
-fine-tuned with [Apple mlx-lm](https://github.com/ml-explore/mlx-examples).
-Each specialist has its own LoRA (rank 8, scale 20) trained on a curated
-HA-domain corpus (forum threads, HA docs, synthetic command/automation
-pairs). System prompts trained per-specialist; see
-[`prompts/`](prompts/).
 ## Evaluation
@@ -115,8 +122,8 @@ scenarios live in [`parity/`](parity/).
 | Artifact | Purpose | Distribution |
 | --- | --- | --- |
-| `qwen25_15b_base.Q4_K_M.gguf` | Quantized base for Ollama / llama.cpp | Hugging Face, ollama.com |
-| `qwen25_15b_{intent}.lora.gguf` (×4) | Specialist LoRA adapters | Hugging Face, ollama.com |
 | `Modelfile.{intent}` (×4) | Ollama recipes (base + LoRA + system prompt) | this repo, ollama.com |
 | `prompts/{intent}.txt` (×4) | Plain-text trained prompts (reference / testing) | this repo |
@@ -128,15 +135,15 @@ mirrored to Hugging Face.
 ```bibtex
 @misc{selora-ai-2026,
-  title  = {Selora AI: Qwen 2.5 1.5B + LoRA Specialists for Home Assistant},
   author = {{Selora Homes}},
   year   = {2026},
   url    = {https://huggingface.co/selora-homes/selora-ai}
 }
 ```
-Base model citation: Qwen Team, *Qwen2.5: A Party of Foundation Models* (2024).
 ## License
-Apache-2.0 (matches the Qwen 2.5 base license).

 ---
 license: apache-2.0
+base_model: Qwen/Qwen3-1.7B
 tags:
   - text-generation
   - qwen
+  - qwen3
   - lora
   - home-assistant
   - home-automation
   - smart-home
+  - tool-use
 language:
   - en
 library_name: transformers
 # Selora AI
+Qwen3 1.7B fine-tuned for Home Assistant with four specialist LoRA
+adapters. The `answer` adapter additionally emits a `query_state` tool
+envelope for live device-state queries against the Home Assistant REST
+API. Used by the [Selora AI Home Assistant
 integration](https://gitlab.com/selorahomes/products/selora-ai/ha-integration);
 also runnable directly via Ollama, llama.cpp, or vLLM.
 ```bash
 llama-server \
+  --model qwen3_17b_base.Q4_K_M.gguf \
   --lora-init-without-apply \
+  --lora qwen3_17b_command.lora.gguf \
+  --lora qwen3_17b_automation.lora.gguf \
+  --lora qwen3_17b_answer.lora.gguf \
+  --lora qwen3_17b_clarification.lora.gguf \
   --ctx-size 8192
 ```
 ```bash
 python -m vllm.entrypoints.openai.api_server \
+  --model ./qwen3_17b_hf \
   --enable-lora --max-loras 4 --max-lora-rank 32 \
   --lora-modules \
     selora-v1-commands=/path/to/peft/command \
 ## Training
+Base: [Qwen3 1.7B](https://huggingface.co/Qwen/Qwen3-1.7B) fine-tuned
+with [Apple mlx-lm](https://github.com/ml-explore/mlx-examples). Each
+specialist has its own LoRA (rank 8–28, scale 20) trained on a curated
+HA-domain corpus (forum threads, HA docs, synthetic command /
+automation pairs). System prompts trained per-specialist; see
+[`prompts/`](prompts/). The `answer` adapter went through a sequential
+continuation pass that added a `query_state` tool envelope on top of
+the original answer-only training distribution; that's preserved in
+the augmented `prompts/answers.txt` and the `Modelfile.answers` SYSTEM
+block.
 ## Evaluation
 | Artifact | Purpose | Distribution |
 | --- | --- | --- |
+| `qwen3_17b_base.IQ4_XS.gguf` | Quantized base for Ollama / llama.cpp | Hugging Face, ollama.com |
+| `qwen3_17b_{intent}.lora.gguf` (×4) | Specialist LoRA adapters | Hugging Face, ollama.com |
 | `Modelfile.{intent}` (×4) | Ollama recipes (base + LoRA + system prompt) | this repo, ollama.com |
 | `prompts/{intent}.txt` (×4) | Plain-text trained prompts (reference / testing) | this repo |
 ```bibtex
 @misc{selora-ai-2026,
+  title  = {Selora AI: Qwen3 1.7B + LoRA Specialists for Home Assistant},
   author = {{Selora Homes}},
   year   = {2026},
   url    = {https://huggingface.co/selora-homes/selora-ai}
 }
 ```
+Base model citation: Qwen Team, *Qwen3 Technical Report* (2025).
 ## License
+Apache-2.0 (matches the Qwen3 base license).

manifest.json CHANGED Viewed

@@ -1,28 +1,28 @@
 {
   "artifacts": {
     "Modelfile.answers": {
-      "sha256": "171f4ed4f4523e683a23c5db1fb28c74853d9d8f56e6923e46fb645c264bb01c",
-      "size": 1651
     },
     "Modelfile.automations": {
-      "sha256": "25f7ebae897190aacaf04e0c82485acbdbe599b2d107fedcbb4143723e2a9c3f",
-      "size": 1757
     },
     "Modelfile.clarifications": {
-      "sha256": "cd012269c14a8f7f923fed7737ca13c4141489152a958fa3138f593fe409b3e5",
-      "size": 1260
     },
     "Modelfile.commands": {
-      "sha256": "a3aaaa8043c6506457fd589c2c6c86da30008e773948b1b6ad5dda4432c6a5cb",
-      "size": 1448
     },
     "README.md": {
-      "sha256": "4c56a84da36ac8bdead77563c6f802293f7e8bb9d9d123c7fb3e2b9846fee313",
-      "size": 4207
     },
     "prompts/answers.txt": {
-      "sha256": "644b59993e6304e2a62b0ce3966005474e9c2b5bf1a76cfa13a9ab9fc7161b2f",
-      "size": 884
     },
     "prompts/automations.txt": {
       "sha256": "91a2e51752acb7b477b5b296710cff1de226deabbe49622c2be374e201422562",
@@ -36,32 +36,28 @@
       "sha256": "b8aea3ac5448921e333285862846b2b47ed70ee95e0fa9527832ff139fc094b5",
       "size": 676
     },
-    "qwen3_17b_answer.lora.gguf": {
-      "sha256": "78326df5b6f5e87dba23213757eb9831e1e7ba8c3c9675ad17fe1b862ce543e8",
-      "size": 9977056
-    },
-    "qwen3_17b_automation.lora.gguf": {
-      "sha256": "35ebf02a91efebee22ed39527779b3ff737172317af939f51ad6fe4a536867a7",
-      "size": 17459488
-    },
-    "qwen3_17b_base.IQ4_XS.gguf": {
-      "sha256": "db25eadd961385299483baec0db07fd29d5963d1faf025a7a9468f60789df292",
-      "size": 1181587232
-    },
     "qwen3_17b_base.f16.gguf": {
       "sha256": "3e4009f0d96955a45f29aa77bded839d376d7832823c6909f76c84ace81dc445",
       "size": 4069678880
     },
-    "qwen3_17b_clarification.lora.gguf": {
-      "sha256": "88efbca12c75c8a6c92b4843455186e8723e19a904af2b701c496846370e9a01",
-      "size": 4988672
     },
-    "qwen3_17b_command.lora.gguf": {
-      "sha256": "df7b60e7c70e4650e4d353d76c875dc74471cc8445bb395af662fad654a5765a",
       "size": 9977056
     }
   },
-  "base_model": "Qwen/Qwen2.5-1.5B-Instruct",
-  "released_at": "2026-05-12T15:59:10Z",
-  "version": "0.4.0"
 }

 {
   "artifacts": {
     "Modelfile.answers": {
+      "sha256": "fd6351414258a679a3b285f1a4882ef6f93b2355d555aaf153c776a3720ba758",
+      "size": 2871
     },
     "Modelfile.automations": {
+      "sha256": "0112f8d5e2bd2dbc839a90a0b9edd1b039af478cc2dbd589160d3b2fdc0f06a5",
+      "size": 1800
     },
     "Modelfile.clarifications": {
+      "sha256": "7fc51ef60f143b8341b3ad53c942b98df555059a0029c117f428861370111f09",
+      "size": 1303
     },
     "Modelfile.commands": {
+      "sha256": "2d46b2ce315d1fc30ac791f71c24f644cd4b13bf7caa8f0d088290ccddeccdde",
+      "size": 1491
     },
     "README.md": {
+      "sha256": "d65dca8d4af1936d1c4f22423417345bfc61a1c550f7bbd6d95a1f86a21ee2f3",
+      "size": 4558
     },
     "prompts/answers.txt": {
+      "sha256": "71d8badea043b2d7c3bb076040a1f0f4c66511ab2785e1e04fa392e3d82c22d2",
+      "size": 1976
     },
     "prompts/automations.txt": {
       "sha256": "91a2e51752acb7b477b5b296710cff1de226deabbe49622c2be374e201422562",
       "sha256": "b8aea3ac5448921e333285862846b2b47ed70ee95e0fa9527832ff139fc094b5",
       "size": 676
     },
     "qwen3_17b_base.f16.gguf": {
       "sha256": "3e4009f0d96955a45f29aa77bded839d376d7832823c6909f76c84ace81dc445",
       "size": 4069678880
     },
+    "selora-v042-answer.f16.gguf": {
+      "sha256": "7d5a7dea12fac72aebc1fb361d18e97e46172a8c2f5cf0c7968322167ef272f9",
+      "size": 14957792
+    },
+    "selora-v042-automation.f16.gguf": {
+      "sha256": "6ceb555a2a809b54294bf311474286612b18a1870789575f3ed9d49396adff3d",
+      "size": 37374880
     },
+    "selora-v042-clarification.f16.gguf": {
+      "sha256": "69d547c8175412a9dc52fdce22dc45cc3cb448e4eb05f11d73e9a27c8939bbe0",
       "size": 9977056
+    },
+    "selora-v042-command.f16.gguf": {
+      "sha256": "7555c5ee00ed46a1eef829d8e82d22ecc251502650c9e866e78c5d0b65e2e1d4",
+      "size": 19938528
     }
   },
+  "base_model": "Qwen/Qwen3-1.7B",
+  "released_at": "2026-05-14T15:16:49Z",
+  "version": "0.4.2"
 }

prompts/answers.txt CHANGED Viewed

@@ -1,11 +1,29 @@
 You are Selora AI, a home automation assistant on Home Assistant. You CAN: control lights/climate/locks/switches, run scripts and scenes, set timers and reminders via timer/input_datetime entities, query device states, and create automations on request. Never say you are a "text-based AI" or that you cannot do something Home Assistant supports — describe how you would do it instead.
-Return ONE JSON object:
 {"intent":"answer","response":"<1-3 sentences>"}
 RULES:
-- Answer the user's question directly. No preamble ("Sure!", "Great question!").
-- 1-3 sentences. Add detail only if the user asked for it.
-- If the question is about home state, ground the answer in AVAILABLE ENTITIES.
-- If the user asks what you can do, list 2-4 concrete capabilities (control devices, set timers, build automations, summarize home state) — not generic phrases.
 - Output ONLY the JSON object.

 You are Selora AI, a home automation assistant on Home Assistant. You CAN: control lights/climate/locks/switches, run scripts and scenes, set timers and reminders via timer/input_datetime entities, query device states, and create automations on request. Never say you are a "text-based AI" or that you cannot do something Home Assistant supports — describe how you would do it instead.
+Return ONE JSON object using one of these envelope shapes:
+ANSWER — for conversational questions, recommendations, or when AVAILABLE ENTITIES already has the full answer:
 {"intent":"answer","response":"<1-3 sentences>"}
+QUERY_STATE — for live state queries that need filtering by state/attribute:
+{"intent":"query_state","calls":[{"tool":"query_state","args":{"domain":"<domain>","filter":{"state":"<value>"}}}]}
+TOOL SCHEMA:
+- tool: "query_state"
+- args:
+    domain (str, required): HA domain — light/switch/lock/cover/fan/media_player/climate/binary_sensor/sensor/person/device_tracker
+    filter (dict, optional):
+      state (str): match exact state ("on", "off", "locked", "open", "home", ...)
+      entity_id (str): match a specific entity_id
+      device_class (str): match HA device_class ("door", "window", "motion", ...)
+      attribute (dict): match attribute key/value (e.g. {"hvac_mode": "heat"})
+WHEN TO USE EACH:
+- query_state for "what's on?", "is X locked?", "how many windows are open?", "which thermostats are heating?".
+- answer for "what can you do?", "explain X", or when the catalog already gives a complete 1-3 sentence answer ("am I home?" → check person entity).
 RULES:
+- 1-3 sentences for answer. Add detail only if the user asked for it.
+- Ground answer responses in AVAILABLE ENTITIES — name actual friendly_names and current state values.
+- When naming a specific device in an answer, wrap its friendly_name in entity markers like [[entity:light.kitchen|Kitchen Lights]] so the panel renders it as a live tile.
 - Output ONLY the JSON object.

qwen25_15b_base.Q4_K_M.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:676f7cda1b9382c83d29c763e947416fe5db1abb4bc25fa7db5aa293164bf5ad
-size 986048000

qwen25_15b_clarification.lora.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:bb3980d049889f29aec831c4aab688983b374868bd218e0f9431d2dce4450e34
-size 10566880

qwen3_17b_automation.lora.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:35ebf02a91efebee22ed39527779b3ff737172317af939f51ad6fe4a536867a7
-size 17459488

qwen3_17b_base.IQ4_XS.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:db25eadd961385299483baec0db07fd29d5963d1faf025a7a9468f60789df292
-size 1181587232

qwen3_17b_clarification.lora.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:88efbca12c75c8a6c92b4843455186e8723e19a904af2b701c496846370e9a01
-size 4988672

qwen3_17b_command.lora.gguf DELETED Viewed

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:df7b60e7c70e4650e4d353d76c875dc74471cc8445bb395af662fad654a5765a
-size 9977056

qwen25_15b_answer.lora.gguf → selora-v042-answer.f16.gguf RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ba2f8c22ace9d8b3e0ff8152a356ab6aa689a2d4d71aa86ee8e2f782f4e2c35
-size 21118176

 version https://git-lfs.github.com/spec/v1
+oid sha256:7d5a7dea12fac72aebc1fb361d18e97e46172a8c2f5cf0c7968322167ef272f9
+size 14957792

qwen25_15b_automation.lora.gguf → selora-v042-automation.f16.gguf RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d49e5207e74a934d3d8730b5e3a7e2beb48e1339aed66d8b1e0d77bd702eeb4e
-size 42220768

 version https://git-lfs.github.com/spec/v1
+oid sha256:6ceb555a2a809b54294bf311474286612b18a1870789575f3ed9d49396adff3d
+size 37374880

qwen3_17b_answer.lora.gguf → selora-v042-clarification.f16.gguf RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:78326df5b6f5e87dba23213757eb9831e1e7ba8c3c9675ad17fe1b862ce543e8
 size 9977056

 version https://git-lfs.github.com/spec/v1
+oid sha256:69d547c8175412a9dc52fdce22dc45cc3cb448e4eb05f11d73e9a27c8939bbe0
 size 9977056

qwen25_15b_command.lora.gguf → selora-v042-command.f16.gguf RENAMED Viewed

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b341c6fe7bf1fef133567f48ae7122567a8b0654b42dafdf70c541adca5d91e4
-size 21118176

 version https://git-lfs.github.com/spec/v1
+oid sha256:7555c5ee00ed46a1eef829d8e82d22ecc251502650c9e866e78c5d0b65e2e1d4
+size 19938528