A LocalLLaMA user reports negative hands-on experiences with the quantized LFM2.5-24B-A2B (Q4_K_M GGUF) model, citing looping behavior, broken code generation on agentic tasks, and poor plugin development results despite using a corrected chat template and Hermes integration.
EDIT: MODEL IS LFM2.5-24B-A2B-Q4_K_M.gguf So, yesterday I made an apparently too simplistic post about this shitty model that everyone is cheering (probably bots, I would like to see what they achieved with it). Link https://www.reddit.com/r/LocalLLaMA/comments/1xb9hrb/tried_lfm2524a2b_its_crap/ The main critic is that I didn't provided any configuration or anything. Well here it is. I tried to run it on llamaampere a llama.cpp version optimized for certain architectures which allowed me to run correctly Qwen 3.8 27b Q4 Small with very high (for me) 40-60 tk/s and satisfying results. But it wasn't the right choice as it keep looping itself over and over and tripping over the easiest things. But I got a 130tk/s on a RTX A4500 card, so it was kinda of an achievement lol Than I moved directly to LM studio, I said ok, let's sacrifice speed and see how it goes. It stabilized the erratic behavious. I tried a bit of chatting in it and it was responding correct-ish. CONFIGURATION, for you control freaks that otherwise would botch my post: LMSTUDIO-> All Engine&Frameworks updated to today 22/10/2026, last version all green checks Model: https://preview.redd.it/sw6dnz6agzwh1.png?width=608&format=png&auto=webp&s=43154be5df8ce228fe6ee9e9a3471f74e0348576 Then I read that this thing is supah-dupah trained on agentic harnesses like openclaw, hermes and so on and they made a pride point in it. So I installed hermes desktop and linked it to hermes. Someone will come to yell at template fix. Got you covered. https://huggingface.co/peculiar-ragdoll/Sharp-LFM2.5-Chat-Templates/raw/main/chat_template.jinja Using this one which is the fix for the template. Asked it in hermes desktop to make a simple plugin to use garnet from microsoft which is a redis clone. Result analyzed by chatgpt which was leading the prompt engineering and checking the result with GPT Sol: "LFM2.5-24B-A2B performed poorly on the Hermes memory-plugin task: it produced broken code, omitted required functionality and claimed verification without end-to-end tests." I tried with the fix (sharp template fix) and it didn't change much. Suggestions? EDIT: MODEL NAME: LFM2.5-24B-A2B-Q4_K_M.gguf
A user shares their positive experience switching from Qwen 3.5 9B to Liquid AI's new LFM2.5-8B-A1B model, praising its speed and reliability for agentic tasks while noting coding remains a weakness. The model is an 8B MoE with 1.5B active parameters and 128K context, optimized for devices and server-side use.
Introducing LFM2.5 8b A1b, a new AI model with performance on par with Nemotron 3 Nano but at higher speed. Support is being added to SmallCode for non-standard tool calls.
The author shares hands-on experience with LFM 2.6B, a small model designed for phones, praising its speed and usefulness for quick tasks like summarization and autocomplete, though it has a 128k context limit.
This Hugging Face model card presents LiquidAI's LFM2.5-2.6B model in GGUF quantized format, with instructions for running it locally via llama.cpp, vLLM, Ollama, and other tools.