Instructions to use espetro/htlm-lfm2.5-350m with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use espetro/htlm-lfm2.5-350m with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("espetro/htlm-lfm2.5-350m")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

llama-cpp-python

How to use espetro/htlm-lfm2.5-350m with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="espetro/htlm-lfm2.5-350m",
	filename="lfm2.5-350m-mlx-q8.gguf",
)

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

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use espetro/htlm-lfm2.5-350m with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf espetro/htlm-lfm2.5-350m
# Run inference directly in the terminal:
llama cli -hf espetro/htlm-lfm2.5-350m

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf espetro/htlm-lfm2.5-350m
# Run inference directly in the terminal:
llama cli -hf espetro/htlm-lfm2.5-350m

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 espetro/htlm-lfm2.5-350m
# Run inference directly in the terminal:
./llama-cli -hf espetro/htlm-lfm2.5-350m

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 espetro/htlm-lfm2.5-350m
# Run inference directly in the terminal:
./build/bin/llama-cli -hf espetro/htlm-lfm2.5-350m

Use Docker

docker model run hf.co/espetro/htlm-lfm2.5-350m

LM Studio
Jan

vLLM

How to use espetro/htlm-lfm2.5-350m with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "espetro/htlm-lfm2.5-350m"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "espetro/htlm-lfm2.5-350m",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/espetro/htlm-lfm2.5-350m

Ollama
How to use espetro/htlm-lfm2.5-350m with Ollama:
```
ollama run hf.co/espetro/htlm-lfm2.5-350m
```

Unsloth Studio

How to use espetro/htlm-lfm2.5-350m 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 espetro/htlm-lfm2.5-350m 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 espetro/htlm-lfm2.5-350m to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for espetro/htlm-lfm2.5-350m to start chatting

How to use espetro/htlm-lfm2.5-350m with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "espetro/htlm-lfm2.5-350m"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "espetro/htlm-lfm2.5-350m"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use espetro/htlm-lfm2.5-350m with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "espetro/htlm-lfm2.5-350m"

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 espetro/htlm-lfm2.5-350m

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use espetro/htlm-lfm2.5-350m with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "espetro/htlm-lfm2.5-350m"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "espetro/htlm-lfm2.5-350m" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

MLX LM

How to use espetro/htlm-lfm2.5-350m with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "espetro/htlm-lfm2.5-350m"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "espetro/htlm-lfm2.5-350m"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "espetro/htlm-lfm2.5-350m",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use espetro/htlm-lfm2.5-350m with Docker Model Runner:
```
docker model run hf.co/espetro/htlm-lfm2.5-350m
```

Lemonade

How to use espetro/htlm-lfm2.5-350m with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull espetro/htlm-lfm2.5-350m

Run and chat with the model

lemonade run user.htlm-lfm2.5-350m-{{QUANT_TAG}}

List all available models

lemonade list

HTLM — Browser-Agent Fine-Tune of LFM2.5-350M

HTLM (HyperText Language Model) is a fine-tuned LFM2.5-350M that predicts web UI actions — click, type, select — on an indexed element list, entirely in-browser via wllama WebAssembly.

Benchmark Results

Metric	Value
Strict action accuracy	91.2%
Action type accuracy	92.2%
Element index accuracy	99.8%
Parse failure rate	0.0%
p95 latency (browser WASM)	1245 ms
Model size (Q8 GGUF)	362 MB
Base model (no fine-tune)	0.2%

Evaluated on 408 held-out tasks from Mind2Web. Full evaluation details: docs/go-no-go-checklist.md.

How It Works

HTLM takes a structured page representation (element list with role/tag/text) and an instruction, and predicts {type, index, [value]}. The element index refers to the candidate list derived from the page HTML.

HTML → element list → HTLM → {type, index, value?}

Usage

Browser (wllama)

import { Wllama } from '@wllama/wllama';

const wllama = new Wllama({ default: './wllama.wasm' });
await wllama.loadModelFromHF({
  repo: 'espetro/htlm-lfm2.5-350m',
  file: 'htlm-350m-q8.gguf',
});

const result = await wllama.createCompletion({
  prompt: JSON.stringify({
    instruction: "Click the submit button",
    page: { elements: [{role:"button",tag:"button",text:"Submit"}] },
  }),
  max_tokens: 128,
});

llama.cpp CLI

llama-cli -m htlm-350m-q8.gguf -p "[INPUT JSON]" -n 128 --temp 0

mlx-lm (Apple Silicon)

from mlx_lm import load, generate
model, tokenizer = load('espetro/htlm-lfm2.5-350m')
# LoRA merge required first — see GitHub repo

Training

Fine-tuned via LoRA (rank 16) on Mind2Web using the mlx-lm / Unsloth-compatible API on Apple Silicon. Full pipeline, hyperparameters, and reproducibility steps: docs/pipeline.md.