LFM2.5-350M-F32-GGUF

LiquidAI/LFM2.5-350M is an ultra-compact 350M-parameter model from Liquid AI's LFM2.5 series, leveraging a hybrid architecture with 10 double-gated Linear Input-Varying (LIV) convolution blocks for efficient sequence processing and 6 Grouped Query Attention (GQA) blocks for precise long-range context handling, trained on 28T tokens (80K:1 token-to-parameter ratio) with extensive reinforcement learning to excel at agentic tasks like tool calling, data extraction, structured JSON outputs, and multi-step reasoning—outperforming models twice its size on GPQA Diamond, MMLU-Pro, IFEval, BFCLv3/4, and CaseReportBench while achieving blazing-fast inference (313 tok/s on AMD CPUs, 188 tok/s on Snapdragon Gen4). Optimized for edge deployment under 1GB memory with native llama.cpp/MLX/vLLM support, it represents peak "intelligence density" for running reliable agent loops on mobiles, IoT devices, and low-power servers where traditional Transformers fail, making high-quality structured data processing and function calling viable at consumer-grade hardware scales.

Model Files

File Name	Quant Type	File Size	File Link
LFM2.5-350M.BF16.gguf	BF16	711 MB	Download
LFM2.5-350M.F16.gguf	F16	711 MB	Download
LFM2.5-350M.F32.gguf	F32	1.42 GB	Download
LFM2.5-350M.Q8_0.gguf	Q8_0	379 MB	Download

Quants Usage

(sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants)

Here is a handy graph by ikawrakow comparing some lower-quality quant types (lower is better):