hf_mirrors/LFM2.5-1.2B-Base
Model Introduction
Files & versions
Analysis
Liquid AI
Try LFM • Documentation • LEAP

LFM2.5-1.2B-Base

LFM2.5 is a new family of hybrid models designed for on-device deployment. It builds on the LFM2 architecture with extended pre-training and reinforcement learning.

Find more information about LFM2.5 in our blog post.

🗒️ Model Details

ModelParametersDescription
LFM2.5-1.2B-Base1.2BPre-trained base model for fine-tuning
LFM2.5-1.2B-Instruct1.2BGeneral-purpose instruction-tuned model
LFM2.5-1.2B-JP1.2BJapanese-optimized chat model
LFM2.5-VL-1.6B1.6BVision-language model with fast inference
LFM2.5-Audio-1.5B1.5BAudio-language model for speech and text I/O

LFM2.5-1.2B-Base is the pre-trained text-only checkpoint, used to create all the LFM2.5-1.2B variants. It has the following features:

  • Number of parameters: 1.17B
  • Number of layers: 16 (10 double-gated LIV convolution blocks + 6 GQA blocks)
  • Training budget: 28T tokens
  • Context length: 32,768 tokens
  • Vocabulary size: 65,536
  • Languages: English, Arabic, Chinese, French, German, Japanese, Korean, Spanish
ModelDescription
LFM2.5-1.2B-BaseOriginal model checkpoint in native format. Best for fine-tuning or inference with Transformers and vLLM.
LFM2.5-1.2B-Base-GGUFQuantized format for llama.cpp and compatible tools. Optimized for CPU inference and local deployment with reduced memory usage.
LFM2.5-1.2B-Base-ONNXONNX Runtime format for cross-platform deployment. Enables hardware-accelerated inference across diverse environments (cloud, edge, mobile).

This pre-trained checkpoint is only recommended for tasks that require heavy fine-tuning, like language-specific (e.g., Japanese) or domain-specific (e.g., medical) assistants, training on proprietary data, or experimenting with novel post-training approaches.

🏃 Inference

LFM2.5 is supported by many inference frameworks. See the Inference documentation for the full list.

NameDescriptionDocsNotebook
TransformersSimple inference with direct access to model internals.LinkColab link
vLLMHigh-throughput production deployments with GPU.LinkColab link
llama.cppCross-platform inference with CPU offloading.LinkColab link

Here's a quick start example with transformers:

from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer

model_id = "LiquidAI/LFM2.5-1.2B-Base"
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    dtype="bfloat16",
#   attn_implementation="flash_attention_2" <- uncomment on compatible GPU
)
tokenizer = AutoTokenizer.from_pretrained(model_id)
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)

prompt = "What is C. elegans?"

input_ids = tokenizer.apply_chat_template(
    [{"role": "user", "content": prompt}],
    add_generation_prompt=True,
    return_tensors="pt",
    tokenize=True,
).to(model.device)

output = model.generate(
    input_ids,
    do_sample=True,
    temperature=0.3,
    min_p=0.15,
    repetition_penalty=1.05,
    max_new_tokens=512,
    streamer=streamer,
)

🔧 Fine-tuning

We recommend fine-tuning LFM2.5 for your specific use case to achieve the best results.

NameDescriptionDocsNotebook
SFT (Unsloth)Supervised Fine-Tuning with LoRA using Unsloth.LinkColab link
SFT (TRL)Supervised Fine-Tuning with LoRA using TRL.LinkColab link
DPO (TRL)Direct Preference Optimization with LoRA using TRL.LinkColab link

Contact

For enterprise solutions and edge deployment, contact sales@liquid.ai.

Citation

@article{liquidai2025lfm2,
  title={LFM2 Technical Report},
  author={Liquid AI},
  journal={arXiv preprint arXiv:2511.23404},
  year={2025}
}