LFM2-350M-Extract

This model is deprecated. Use LFM2.5-350M for compact structured extraction with the latest 350M checkpoint.

LFM2-350M-Extract is the fastest extraction model, optimized for edge deployment with strict memory and compute constraints. It delivers structured data extraction with minimal latency.

HF GGUF ONNX

Specifications

Property	Value
Parameters	350M
Context Length	32K tokens
Task	Structured Extraction
Output Formats	JSON, XML, YAML

Edge Deployment

Runs on mobile devices

Low Latency

Fastest extraction model

Batch Processing

High-volume extraction

Prompting Recipe

Use temperature=0 (greedy decoding) for best results. This model is intended for single-turn conversations only.

System Prompt Format:

Identify and extract information matching the following schema.
Return data as a JSON object. Missing data should be omitted.

field_name: Description of what to extract
nested_object:
  nested_field: Description

The system prompt is recommended but optional: without one, the model defaults to JSON output. Specify the format (JSON, XML, or YAML) and the schema to improve accuracy. Supported languages: English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish.

Quick Start

Transformers
llama.cpp

Install:

pip install "transformers>=5.2.0" torch accelerate

Run:

from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = "LiquidAI/LFM2-350M-Extract"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

system_prompt = """Identify and extract information matching the following schema.
Return data as a JSON object. Missing data should be omitted.

product: Product name
price: Price in dollars
quantity: Number of items
"""

user_input = "Order: 5 units of Widget Pro at $29.99 each"

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": user_input}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt", return_dict=True).to(model.device)
outputs = model.generate(**inputs, max_new_tokens=256, temperature=0, do_sample=False)
response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:], skip_special_tokens=True)
print(response)

Download GGUF:

hf download LiquidAI/LFM2-350M-Extract-GGUF \
  --local-dir ./LFM2-350M-Extract-GGUF

Run:

llama-cli -m ./LFM2-350M-Extract-GGUF/LFM2-350M-Extract-Q4_K_M.gguf \
  -sys $'Identify and extract information matching the following schema.\nReturn data as a JSON object. Missing data should be omitted.\n\nproduct: Product name\nprice: Price in dollars\nquantity: Number of items' \
  -p 'Order: 5 units of Widget Pro at $29.99 each' \
  --temp 0

Models

Fine-tuning

Edge Inference

GPU Inference

Cloud Inference

Guides

Help

Specifications

Edge Deployment

Low Latency

Batch Processing

Prompting Recipe

Quick Start

​Specifications

Edge Deployment

Low Latency

Batch Processing

​Prompting Recipe

​Quick Start

Specifications

Prompting Recipe

Quick Start