A purpose-trained smart home automation model fine-tuned from LiquidAI/LFM2.5-350M. This model controls lights, doors, thermostats, TVs, fans, speakers, and home scenes through structured tool calls, with full awareness of current device states.

👉 Which file should I download? For the most stable experience right now, download one of the following GGUF files:

LFM2.5-350M-home-assistant-sft-exp.F16.gguf
LFM2.5-350M-home-assistant-sft-exp.Q4_K_M.gguf (Recommended for most users, best balance of speed and size)
LFM2.5-350M-home-assistant-sft-exp.Q8_0.gguf (Highest quality quantization)

🛠️ Get the Code (Implementation) Want to run this model in a real smart home environment? You can clone the fully integrated, modified, and open-sourced Home Assistant setup (adapted from the original Liquid cookbook) here: 👉 [GitHub: (https://github.com/OrdenWills/home-assistant) ]

What It Does

Given a natural language command and the current state of all connected devices, the model outputs the correct tool call — or explains in plain text why no action is needed. It handles:

Advanced Device Disambiguation — If you say "Turn off the TV", the model will intelligently resolve which TV you mean by checking if only one TV is connected, checking if you are in a room with a TV, or inferring intent from the state (e.g., if only one TV is currently ON, it turns that one off via a <think> trace).
Media & Music Playback — "Play Truth In The World By Lucky Dube" → control_speaker(room='living_room', action='play', media='Truth In The World By Lucky Dube')
Already-Satisfied & Bulk State Awareness — "Turn off what's on" → reads STATE, emits one call per lit room using <think> logic. If all lights are already off, gracefully outputs plain text: "All lights are already off" (zero tool calls).
Universal Scope Commands — "Off everything" or "Close everywhere" → dynamically checks lights, doors, TVs, speakers, and fans, emitting combined calls only for devices that need action.
Pronoun Resolution & Log-based Undo — "Undo that" or "Turn them off" + [RECENT ACTIONS: toggle_lights(bedroom, on)] → toggle_lights(room='bedroom', state='off'). Recognizes correct tool types for logged devices (e.g., locking doors vs toggling lights).
Relative State Clauses — "Close the door that is open" → directly filters [STATE:] to find the unlocked door, ignoring local room context.
Multi-device Compound Commands — "Lock the front door, turn off the living room light, and play music" → uses a rigid reasoning format (Total: N tool calls required. Emitting all N.) to emit parallel tool calls without truncation.
Topology-aware Rejection — "Turn on the garage fan" when the garage is not in connected fans → intent_unclear(reason='unsupported_device').
Rigid Syntactic Action Triggers — Internal reasoning strictly concludes with ACTION REQUIRED. or ACTION NOT REQUIRED. Text reply only. to reliably signal structural intent before opening a JSON block, eliminating JSON bleed.

Evaluation Results (V14)

The model achieves an overall accuracy of 98.5% across 59 highly specific edge-case and logic-stress categories.

text

════════════════════════════════════════════════════════════
  CATEGORY ACCURACY SUMMARY (V14 HARDENED)
════════════════════════════════════════════════════════════
  Category                                  Pass  Total     Acc
────────────────────────────────────────────────────────────
  current_track_queries                       25     25  100.0%  ████████████████████
  fan_relative_adjustments                    25     25  100.0%  ████████████████████
  contextual_log_followups                    21     25   84.0%  ████████████████
  thermostat_min_max                          13     25   52.0%  ██████████
  mixed_device_undo                           25     25  100.0%  ████████████████████
  speaker_volume_controls                     25     25  100.0%  ████████████████████
  compound_volume_plus_devices                25     25  100.0%  ████████████████████
  single_gadget_ignore_room                   25     25  100.0%  ████████████████████
  compound_local_dedup                        25     25  100.0%  ████████████████████
  bulk_action_strict_counting                 25     25  100.0%  ████████████████████
  plurality_strict_log_reference              25     25  100.0%  ████████████████████
  living_room_door_supported                  25     25  100.0%  ████████████████████
  thermostat_relative                         25     25  100.0%  ████████████████████
  direct_command_ignore_log                   25     25  100.0%  ████████████████████
  relative_state_clause_stress                25     25  100.0%  ████████████████████
  off_everything                              25     25  100.0%  ████████████████████
  bulk_already_satisfied                      25     25  100.0%  ████████████████████
  scope_isolation                             25     25  100.0%  ████████████████████
  pronoun_correct_tool_type                   25     25  100.0%  ████████████████████
  list_and_local                              25     25  100.0%  ████████████████████
  compound_log_and_local                      25     25  100.0%  ████████████████████
  compound_three_action                       25     25  100.0%  ████████████████████
  gadget_explicit_room_rejection              25     25  100.0%  ████████████████████
  action_log_queries                          25     25  100.0%  ████████████████████
  social_off_topic                            25     25  100.0%  ████████████████████
  action_log_gadgets                          25     25  100.0%  ████████████████████
  compound_multi_gadget_explicit              25     25  100.0%  ████████████████████
  self_contradictory                          25     25  100.0%  ████████████████████
  log_plus_rel_clause                         25     25  100.0%  ████████████████████
  bulk_state_counting_hallucination           25     25  100.0%  ████████████████████
  zero_state_bulk                             25     25  100.0%  ████████████████████
  state_report_queries                        25     25  100.0%  ████████████████████
  double_bulk_stress                          25     25  100.0%  ████████████████████
  them_plurality                              25     25  100.0%  ████████████████████
  partial_exec_ambiguous                      25     25  100.0%  ████████████████████
  compound_media_devices                      25     25  100.0%  ████████████████████
  door_incomplete_vs_unsupported              24     24  100.0%  ████████████████████
  room_priority_over_log                      25     25  100.0%  ████████████████████
  compound_local                              25     25  100.0%  ████████████████████
  typo_domain_hallucination                   25     25  100.0%  ████████████████████
  bulk_undo_wrong_tool                        25     25  100.0%  ████████████████████
  current_room_door_hallucination             25     25  100.0%  ████████████████████
  already_satisfied                           25     25  100.0%  ████████████████████
  action_required                             25     25  100.0%  ████████████████████
  user_room_lights                            23     25   92.0%  ██████████████████
  user_room_doors                             23     25   92.0%  ██████████████████
  bulk_state_aware                            25     25  100.0%  ████████████████████
  action_log_lights                           25     25  100.0%  ████████████████████
  action_log_doors                            25     25  100.0%  ████████████████████
  action_log_scenes                           25     25  100.0%  ████████████████████
  scenes                                      25     25  100.0%  ████████████████████
  thermostat                                  25     25  100.0%  ████████████████████
  rejections                                  25     25  100.0%  ████████████████████
  incomplete_no_room                          21     25   84.0%  ████████████████
  missing_device                              25     25  100.0%  ████████████████████
  mixed_compound                              25     25  100.0%  ████████████████████
  status_queries                              25     25  100.0%  ████████████████████
  tv_commands                                 25     25  100.0%  ████████████████████
  speaker_commands                            25     25  100.0%  ████████████████████
  fan_commands                                25     25  100.0%  ████████████████████
  massive_media                               25     25  100.0%  ████████████████████
  relative_clause                             25     25  100.0%  ████████████████████
  disambiguate_back                           25     25  100.0%  ████████████████████
  rule3_inference                             25     25  100.0%  ████████████████████
  cur_room_unsupported                        24     25   96.0%  ███████████████████
────────────────────────────────────────────────────────────
  OVERALL                                   1599   1624   98.5%
════════════════════════════════════════════════════════════

Tool Schema

The model outputs calls from the following 10-tool schema. All tools use exact parameter names.

python

toggle_lights(room: str, state: 'on'|'off')
# room: living_room | bedroom | kitchen | bathroom | office | hallway

lock_door(door: str, state: 'lock'|'unlock')
# door: front | back | garage | side | bedroom | bathroom | office | kitchen | living_room

set_thermostat(temperature: int, mode: 'heat'|'cool'|'auto')
# temperature range: 60–80°F

set_scene(scene: 'movie_night'|'bedtime'|'morning'|'away'|'party')

control_tv(room: str, state: 'on'|'off')
# room: living_room | bedroom | office

control_fan(room: str, state: 'on'|'off', speed: 'low'|'medium'|'high' = optional)
# room: living_room | bedroom | kitchen | office

control_speaker(room: str, action: 'play'|'pause'|'stop'|'next'|'previous'|'volume', media: str = optional, volume: int = optional)
# room: living_room | bedroom | kitchen | office | hallway. volume: 0-100 (Required when action='volume')


intent_unclear(reason: 'off_topic'|'incomplete'|'unsupported_device'|'unsupported_feature')

State Format

Every user message must be prefixed with a [STATE:] block.

[STATE: lights={bedroom:on, kitchen:off, living_room:on}, doors={back:locked, front:unlocked}, thermostat=70F/heat, scene=none, tv={bedroom:off, living_room:on}, speaker={kitchen:stopped}, fan={bedroom:on(low)},current_track_name=<name of song>, current_user_room=kitchen]

Field breakdown:

Field	Values	Notes
`lights`	`room:on\|off`	Only include connected rooms
`doors`	`door:locked\|unlocked`	Only include connected doors
`thermostat`	`<temp>F/<mode>`	e.g. `72F/heat`
`scene`	scene name or `none`	Active scene or none
`tv`	`{room:on\|off, ...}`	Dictionary of connected TVs
`speaker`	`{room:playing\|paused\|stopped, ...}`	Dictionary of connected speakers
`fan`	`{room:on\|off(speed), ...}`	Dictionary of connected fans
`current_user_room`	room name or empty	Drives pronoun ("this room") resolution

Usage

Minimal inference example

python

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "OrdenWills/LFM2.5-350M-home-assistant-sft"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)

SYSTEM_PROMPT = """You are a smart home assistant AI. Use tools to control the home.

Output function calls as JSON.

TOOLS:
  toggle_lights(room, state='on'|'off')
  lock_door(door, state='lock'|'unlock')
  set_thermostat(temperature=<int>, mode='heat'|'cool'|'auto')
  set_scene(scene='movie_night'|'bedtime'|'morning'|'away'|'party')
  control_tv(room, state='on'|'off')
  control_fan(room, state='on'|'off'[, speed='low'|'medium'|'high'])
  control_speaker(room, action='play'|'pause'|'stop'|'next'|'previous'|'volume'[, media='<str>', volume=<int 0-100>])
  intent_unclear(reason='off_topic'|'incomplete'|'unsupported_device'|'unsupported_feature')

CONNECTED ROOMS (lights): living_room, bedroom, kitchen, bathroom, office, hallway
CONNECTED DOORS: front, back, garage
CONNECTED TVs: living_room, bedroom
CONNECTED SPEAKERS: living_room
CONNECTED FANS: bedroom

STATE RULES:
  [STATE:] shows all current device states.
  State already matches request → plain text reply, NO tool call.
  Only rooms listed under CONNECTED TVs/SPEAKERS/FANS have those devices.
  Requesting a device in an unlisted room → intent_unclear(unsupported_device).

TV / SPEAKER / FAN RESOLUTION when user says 'the TV'/'the fan'/'the speaker':
  1. Exactly one connected → use that room automatically.
  2. Multiple connected + current_user_room has device → use current_user_room.
  3. Multiple connected + exactly ONE is in the eligible state for the action
     (e.g. only one TV is on and user says 'turn off the TV') → infer that room.
  4. Multiple connected + ambiguous (rule 2 & 3 don't apply) → intent_unclear(incomplete).

LIGHT / DOOR RESOLUTION:
  current_user_room set + connected → use current_user_room.
  current_user_room set + NOT connected → intent_unclear(unsupported_device).
  current_user_room empty → intent_unclear(incomplete).

  [RECENT ACTIONS:] → transaction log, newest entry first. Format:
    (X mins ago) [call1, call2, ...] -> summary.
  Each [...] bracket is ONE command the user previously issued.
  For 'undo'/'reverse'/'back': invert ONLY the most recent transaction
    (the FIRST [...] block). Older transactions are always ignored.
  For pronouns ('it'/'them'): refer to the device(s) in the first [...] block.
  Do NOT use recent actions to infer which room 'the/this light' or 'the/this door'
  refers to when current_user_room is explicitly set — current_user_room wins.
  For 'all lights' / 'all doors': check STATE for each device — act only
  on those whose state contradicts the request (issue individual tool calls).
  SYNONYMS: 'open'='unlock'; 'close'/'shut'='lock'; 'skip'='next';
  'back'='previous' (for speaker track navigation), but can also mean 'undo' for
  reverting device states based on [RECENT ACTIONS].
  'continue'/'resume'/'on the music'='play'; 'play <song/artist>' = action='play' + media='<str>'.
  Relative state clauses ('the light that is on', 'the door that is locked')
  override current_user_room — check STATE and act on the matching device."""

state = "[STATE: lights={bathroom:off, bedroom:off, hallway:off, kitchen:off, living_room:off, office:off}, doors={back:locked, bathroom:locked, bedroom:locked, front:locked, garage:locked, kitchen:locked, living_room:locked, office:locked, side:locked}, thermostat=65F/cool, scene=bedtime, tv={living_room:off}, speaker={living_room:paused}, fan={bedroom:on(medium), kitchen:off(medium), living_room:on(medium), office:off(low)}, current_user_room=bedroom]"

messages = [
    {"role": "system",    "content": SYSTEM_PROMPT},
    {"role": "user",      "content": f"{state}\nTurn off the TV."},
]

input_ids = tokenizer.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_tensors="pt"
).to(model.device)

with torch.no_grad():
    output = model.generate(
        input_ids,
        max_new_tokens=256,
        temperature=0.1,        # low temp for deterministic tool calls
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id,
    )

response = tokenizer.decode(
    output[0][input_ids.shape[-1]:], skip_special_tokens=True
)
print(response)
# Expected behavior: The model will generate a <think> trace noting that only the living_room TV is currently ON, infer the user wants to turn off the living_room TV, conclude with ACTION REQUIRED, and emit the tool call.

GGUF / Ollama

bash

# Pull the recommended q4_k_m quantization
ollama run hf.co/OrdenWills/LFM2.5-350M-home-assistant-sft:Q4_K_M

# Or use the higher precision q8_0 version
ollama run hf.co/OrdenWills/LFM2.5-350M-home-assistant-sft:Q8_0

Training Details

This release was fine-tuned directly on a highly augmented 160,000-example state-aware synthetic dataset ("V14").

Parameter	Value
Base model	LiquidAI/LFM2.5-350M
Dataset Size	160,000 examples
Categories	59 granular instruction schemas
Think Traces	Extensive (Positive-only reasoning patterns)
Hardware	Kaggle T4 (16 GB)

Key Training Features (V14)

Positive-Only Think Traces: Models reason entirely in the affirmative (learning what to do rather than what not to do), heavily reducing hallucination.
Syntactic Triggers: Internal logic traces now rigorously conclude with ACTION REQUIRED. or ACTION NOT REQUIRED. Text reply only. ensuring flawless tool call initiation boundaries.
Compound Count Enforcement: The model is trained to actively enumerate and count sub-actions within its reasoning (Total: N tool calls required. Emitting all N.) to prevent truncation on massive requests (e.g. locking 8 doors at once).
Extended Gadget Handling: intent_unclear(unsupported_device) now intelligently catches missing Doors, TVs, and Fans even when rooms are explicitly named.
"Off Everything" & Bulk Isolation: The model gracefully handles mixed-domain global commands ("lock everywhere and shut everything off") by iterating through device arrays independently.

Known Limitations

Temperature range is fixed at 60–80°F. Requests outside this range produce a plain-text explanation, not a tool call.
No brightness or colour control. Dimming and colour-change requests correctly trigger intent_unclear(reason='unsupported_feature'). This is by design — the connected lights only support on/off.
Local music library only. The speaker control media parameter maps to specific tracks from a bounded internal list of artists and songs. Out-of-domain conversational queries will likely trigger intent_unclear(reason='off_topic').
English only. All training data is English. Performance in other languages is untested.
State must be accurate. The model trusts [STATE:] completely. If your app sends stale state, the model may incorrectly say a device is already in the requested state or infer the wrong device during disambiguation.

Citation

bibtex

@misc{lfm2-home-assistant-2026,
  author    = {OrdenWills},
  title     = {LFM2.5-350M Home Assistant: A Purpose-Trained Smart Home Automation Model},
  year      = {2026},
  publisher = {HuggingFace},
  url       = {https://huggingface.co/OrdenWills/LFM2.5-350M-home-assistant-sft}
}

Acknowledgements

LiquidAI for the incredible and highly capable LFM2.5-350M base model.