Try LFM • Docs • LEAP • Discord

LFM2.5-2.6B-ONNX

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 details in the original model card: https://huggingface.co/LiquidAI/LFM2.5-2.6B

Recommended Variants

Precision	Size	Platform	Use Case
Q4	~1.9 GB	WebGPU, Server	Recommended for most uses (quantized embedding)
Q4F16	~1.5 GB	WebGPU	Quantized embedding and q4 weights with FP16 runtime and caches
FP16	~2.1 GB	WebGPU, Server	Higher quality
Q8	~2.1 GB	Server only	Balance of quality and size

• WebGPU: Use Q4, Q4F16, or FP16 (Q8 is not supported on WebGPU).
• Server (CPU/GPU): All variants supported.

Q4 and Q4F16 use a quantized input embedding. Q4F16 uses FP16 runtime tensors and caches while quantizing the LM head and decoder linear weights to q4.

Model Files

onnx/
├── model.onnx              # FP32
├── model_fp16.onnx         # FP16
├── model_q4.onnx           # Q4, quantized embedding (WebGPU)
├── model_q4f16.onnx        # Q4 embedding/weights, FP16 runtime and caches (WebGPU)
└── model_q8.onnx           # Q8

Python (onnxruntime)

pip install onnxruntime transformers numpy huggingface_hub
# or, for GPU:
pip install onnxruntime-gpu transformers numpy huggingface_hub

from huggingface_hub import hf_hub_download

model_id = "LiquidAI/LFM2.5-2.6B-ONNX"
# Q8 recommended for server CPU/GPU; use model_q4.onnx for WebGPU.
hf_hub_download(model_id, "onnx/model_q8.onnx")
hf_hub_download(model_id, "onnx/model_q8.onnx_data")

WebGPU (Transformers.js)

import { pipeline } from "@huggingface/transformers";

const generator = await pipeline("text-generation", "LiquidAI/LFM2.5-2.6B-ONNX", {
  device: "webgpu",
  dtype: "q4", // or "q4f16" or "fp16"
});