Liquid AILFM2.5Lfm2ForCausalLM

LFM2.5 350M — Hardware Requirements & GPU Compatibility

ChatFunctions

LFM2.5 350M is a 354M-parameter open language model from Liquid AI in the LFM2.5 family. It supports a context window of up to 128,000 tokens. At Q4_K_M it needs about 0.58 GB of VRAM — see which GPUs and Macs can run it below.

354M params · dense128K contextOtherMar 202653.0K downloads437 likes82.8K quant downloads

Based on LFM2.5 350M Base

Your hardware

Verdicts below are for the reference RTX 4090. Set yours to personalise the page.

Detection reads your browser's GPU name with no data sent anywhere.

Runs great on a 24 GB GPU

Runs from 8 GB (Q4_K_M · 125K)Best on 24 GB (Q4_K_M · ~1130 tok/s · 125K)Macs from 8 GB (Q4_K_M · ~124 tok/s)

Other

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run LFM2.5 350M

One recommended way. Every other method, OS and quant is in the step-by-step guide.

Other ways

each opens step-by-step

Run in cloud

Fits on RTX 3060 12GB (11 GB headroom) · Q4_K_M

Generation speed: ~403 tok/s; generation speed
Cost per 1M output tokens: $0.057; per 1M output tokens

Deployon Vast.aiBest optionfrom $0.083/hr

Compare GPUs →

Run LFM2.5 350M

0.8 GB VRAM at 16K context · fits the RTX 4090 (23.2 GB free)

Works on macOS, Windows and Linux.

Install Ollama (Linux)
$curl -fsSL https://ollama.com/install.sh | sh
Install Ollama (macOS / Windows)
Download and run the app; it installs the ollama command.
Download Ollama ↗
Run Q4_K_M
$ollama run hf.co/LiquidAI/LFM2.5-350M-GGUF:Q4_K_M
Raise the context window
Type it inside the chat. For the API server instead, start it with OLLAMA_CONTEXT_LENGTH=16384 ollama serve.
$/set parameter num_ctx 16384
A larger context needs more memory than the weights alone.
Use the API
Ollama serves its API on http://localhost:11434.

Works on macOS, Windows and Linux.

Install LM Studio
Desktop app for macOS (Apple silicon), Windows and Linux.
lmstudio.ai ↗
Download the model
In the app press Ctrl/Cmd + Shift + M and paste https://huggingface.co/LiquidAI/LFM2.5-350M-GGUF, then pick Q4_K_M. Or use the CLI (run LM Studio once first):
$lms get https://huggingface.co/LiquidAI/LFM2.5-350M-GGUF@Q4_K_M
Open on Hugging Face ↗
Load and chat
Open the model loader, select the model and set Context Length to 16384. Leave GPU offload at maximum if it fits.
Start the local server (optional)
$lms server start

Works on macOS, Windows and Linux.

Install llama.cpp (macOS)
MacPorts, Nix and conda-forge also work.
$brew install llama.cpp
Install options ↗
Install llama.cpp (Windows)
$winget install llama.cpp
Install llama.cpp (Linux)
Download the prebuilt llama-*-bin-ubuntu-* archive for your GPU backend (CPU, Vulkan) from the releases page, unpack it and run the binaries from that folder. Homebrew (brew install llama.cpp) and conda-forge also work on Linux.
llama.cpp releases ↗
Start the server
Downloads the model on first run. The chat UI and an OpenAI-compatible API are at http://localhost:8080.
$llama-server -hf LiquidAI/LFM2.5-350M-GGUF:Q4_K_M -c 16384 -ngl 99 --port 8080

Works on macOS, Windows and Linux.

Option A: Docker Model Runner
Docker Desktop: Settings > AI > Enable Docker Model Runner. Docker Engine on Linux: install the docker-model-plugin package.
$docker model run hf.co/LiquidAI/LFM2.5-350M-GGUF:Q4_K_M
Model Runner docs ↗
The quant tag after the repo is how Hugging Face cards write it; if your Docker version rejects it, drop the tag and pick the quant by repo.
Option B: llama.cpp server image (NVIDIA)
Needs the NVIDIA Container Toolkit. Models are cached in the llama-cache volume. Open http://localhost:8080.
$docker run --gpus all -p 8080:8080 -v llama-cache:/root/.cache ghcr.io/ggml-org/llama.cpp:server-cuda -hf LiquidAI/LFM2.5-350M-GGUF:Q4_K_M -c 16384 -ngl 99 --host 0.0.0.0 --port 8080
Image tags (cuda13, vulkan, rocm) ↗
Cache path inside the image is not in the docs; remove the -v flag if the volume complains.

Linux only.

Install vLLM
Linux only; on Windows run it inside WSL2. NVIDIA GPU shown; AMD ROCm, Intel and CPU builds have their own install pages.
$uv venv --python 3.12 --seed
$source .venv/bin/activate
$uv pip install vllm --torch-backend=auto
vLLM installation ↗
Serve the model
OpenAI-compatible API at http://localhost:8000/v1.
$vllm serve LiquidAI/LFM2.5-350M --max-model-len 16384
Runs the original weights, not the Q4_K_M GGUF; needs ~1 GB of GPU memory.

macOS only.

MLX supports only macOS, so follow these steps for that setup.

Install mlx-lm
$pip install mlx-lm
Chat in the terminal
$mlx_lm.chat --model LiquidAI/LFM2.5-350M-MLX-8bit
Downloads the model on first run. The MLX repo has its own quantisation; it may not match the quant selected above.
Or start a local server
Listens on http://localhost:8080.
$mlx_lm.server --model LiquidAI/LFM2.5-350M-MLX-8bit

Works on macOS, Windows and Linux.

Install Jan
jan.ai ↗
Find the model
Open the Hub tab, paste this model ID in Search, and choose Q4_K_M if offered.
$LiquidAI/LFM2.5-350M-GGUF
Use the local API (optional)
Settings > Local API Server > Start Server. It listens on http://127.0.0.1:1337.

Rent GPUs hourly to run Q4_K_M right there. Pricing comes from partner marketplaces and varies.

Available GPU rentals for Q4_K_M, prioritized by value
GPU	Speed	$/1M tokens	Rent
RTX 3060 12GBCheapestCheapest11 GB free	~403 tok/s	$0.057	Deploy on Vast.aifrom $0.083/hr (opens in new tab)	11 GB free · ~403 tok/s · $0.057 / 1M tokens
RTX 3070 (8 GB)7 GB free	~502 tok/s	$0.072	Deploy on RunPodfrom $0.13/hr (opens in new tab)	8 GB · 7 GB free · ~502 tok/s · $0.072 / 1M tokens
RTX 3070 (8 GB)7 GB free	~502 tok/s	$0.075	Deploy on Vast.aifrom $0.14/hr (opens in new tab)	8 GB · 7 GB free · ~502 tok/s · $0.075 / 1M tokens
RTX 3060 Ti (8 GB)7 GB free	~502 tok/s	$0.083	Deploy on Vast.aifrom $0.15/hr (opens in new tab)	8 GB · 7 GB free · ~502 tok/s · $0.083 / 1M tokens
RTX 5060 Ti 16GB15 GB free	~502 tok/s	$0.091	Deploy on Vast.aifrom $0.16/hr (opens in new tab)	15 GB free · ~502 tok/s · $0.091 / 1M tokens
RTX 3080 Ti (12 GB)11 GB free	~1023 tok/s	$0.049	Deploy on RunPodfrom $0.18/hr (opens in new tab)	12 GB · 11 GB free · ~1023 tok/s · $0.049 / 1M tokens
Tesla V100 PCIe 16GB15 GB free	~1009 tok/s	$0.052	Deploy on RunPodfrom $0.19/hr (opens in new tab)	15 GB free · ~1009 tok/s · $0.052 / 1M tokens
RTX 4090 (24 GB)23 GB free	~1130 tok/s	$0.084	Deploy on RunPodfrom $0.34/hr (opens in new tab)	24 GB · 23 GB free · ~1130 tok/s · $0.084 / 1M tokens

Compare GPUs →

How Much VRAM Does LFM2.5 350M Need?

Select a quantization level; the listed hardware adheres to it.

VRAM	+32K	+128K	File	Quality	On RTX 4090 (reference)
0.5 GB	0.9 GB	—	0.1 GB	2-bit · i-quant extreme compression — Importance-weighted 2-bit, extreme compression — significant quality loss	Fits: 125K context · ~1424 tok/s
0.5 GB	0.9 GB	—	0.1 GB	2-bit · i-quant extra small — Importance-weighted 2-bit, extra small	Fits: 125K context · ~1394 tok/s
0.5 GB	0.9 GB	—	0.1 GB	2-bit · i-quant small — Importance-weighted 2-bit, small	Fits: 125K context · ~1365 tok/s
0.5 GB	0.9 GB	—	0.1 GB	2-bit · i-quant medium — Importance-weighted 2-bit, medium	Fits: 125K context · ~1337 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit · i-quant — Importance-weighted 3-bit	Fits: 125K context · ~1310 tok/s
0.5 GB	0.9 GB	—	0.2 GB	2-bit · small K-quant — 2-bit small K-quant	Fits: 125K context · ~1285 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit · i-quant extra small — Importance-weighted 3-bit, extra small	Fits: 125K context · ~1285 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit · i-quant small — Importance-weighted 3-bit, small	Fits: 125K context · ~1260 tok/s
0.5 GB	0.9 GB	—	0.2 GB	2-bit · K-quant — 2-bit quantization with K-quant improvements	Fits: 125K context · ~1260 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit small — 3-bit small quantization	Fits: 125K context · ~1260 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit · i-quant medium — Importance-weighted 3-bit, medium	Fits: 125K context · ~1236 tok/s
0.5 GB	0.9 GB	—	0.2 GB	3-bit medium — 3-bit medium quantization	Fits: 125K context · ~1213 tok/s
0.5 GB	0.9 GB	—	0.2 GB	4-bit · legacy — 4-bit legacy quantization	Fits: 125K context · ~1213 tok/s
0.6 GB	1 GB	—	0.2 GB	3-bit large — 3-bit large quantization	Fits: 125K context · ~1191 tok/s
0.6 GB	1 GB	—	0.2 GB	4-bit · i-quant compact — Importance-weighted 4-bit, compact	Fits: 125K context · ~1170 tok/s
0.6 GB	1 GB	—	0.2 GB	4-bit · legacy+ — 4-bit legacy quantization with offset	Fits: 125K context · ~1150 tok/s
0.6 GB	1 GB	—	0.2 GB	4-bit small — 4-bit small quantization	Fits: 125K context · ~1150 tok/s
0.6 GB	1 GB	—	0.2 GB	4-bit · i-quant non-linear — Importance-weighted 4-bit, non-linear	Fits: 125K context · ~1150 tok/s
0.6 GB	1 GB	—	0.2 GB	4-bit medium — 4-bit medium quantization — most popular sweet spot	Fits: 125K context · ~1130 tok/s
0.6 GB	1 GB	—	0.3 GB	5-bit small — 5-bit small quantization	Fits: 125K context · ~1074 tok/s
0.6 GB	1 GB	—	0.3 GB	5-bit medium — 5-bit medium quantization — good quality/size tradeoff	Fits: 125K context · ~1057 tok/s
0.7 GB	1.1 GB	—	0.3 GB	6-bit, very good — 6-bit quantization, very good quality	Fits: 125K context · ~993 tok/s
0.7 GB	1.1 GB	—	0.4 GB	8-bit, near-lossless — 8-bit quantization, near-lossless	Fits: 125K context · ~910 tok/s
1.1 GB	1.5 GB	—	0.7 GB	16-bit · bfloat — Brain floating point 16 — preferred for training	Fits: 125K context · ~607 tok/s

Selected Q4_K_M

4-bit medium quantization

0.8GBat 16K context

Weights: 0.2 GB
KV cache: 0.2 GB
Overhead: 0.3 GB

Fits RTX 4090 (reference) · 23.3 GB free · ~1130 tok/s

≥50 tok/s needs ≥45 GB/s · RTX 5090 ~2008, RTX 3090 Ti ~1130

Which Hardware Can Run LFM2.5 350M?

Q4_K_M · 0.6 GB

LFM2.5 350M (Q4_K_M) requires 0.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ GB is recommended. Using the full 128K context window can add up to 1.8 GB, bringing total usage to 2.4 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run LFM2.5 350M, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M

Typical hardwareQuantVRAMtok/sContext that fits

8 GB · RTX 4060Q4_K_M · 0.8 GB · ~305 tok/s · 125K0.8~305125K8 GB tier →

12 GB · RTX 3060 12GBQ4_K_M · 0.8 GB · ~403 tok/s · 125K0.8~403125K12 GB tier →

16 GB · RTX 5060 Ti 16GBQ4_K_M · 0.8 GB · ~502 tok/s · 125K0.8~502125K16 GB tier →

24 GB · RTX 4090referenceQ4_K_M · 0.8 GB · ~1130 tok/s · 125K0.8~1130125K24 GB tier →

Mac 32 GB · Mac Mini M4Q4_K_M · 0.8 GB · ~145 tok/s · 125K0.8~145125KMacs →

Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 0.8 GB · ~659 tok/s · 125K0.8~659125KMacs →

Alternatives to LFM2.5 350M

LFM2.5 family · 14 models →

Similar models

6tok/s on RTX 4090

Model	Score	VRAM	tok/s	Why it is similar
Granite 4.0 350MIBM	—	1 GB	~1150	same size
SmolLM 360MHugging Face	—	0.6 GB	~1092	same size
ERNIE 4.5 0.3B PTBaidu	—	0.7 GB	~1213	same size
Opt 350MMeta	—	0.8 GB	~851	same size
Codegen 350M MonoSalesforce	—	0.8 GB	~851	same size
Gpt2 MediumOpenAI	—	0.3 GB	~2621	same size

Fine-tunes & variants(3)community builds of LFM2.5 350M

Specifications

Publisher: Liquid AI
Family: LFM2.5
Parameters: 354M
Architecture: Lfm2ForCausalLM
Context Length: 128,000 tokens
Vocabulary Size: 65,536
Release Date: 2026-03-31
License: Other

Resources

Frequently Asked Questions

How much VRAM does LFM2.5 350M need?

LFM2.5 350M requires 0.6 GB of VRAM at Q4_K_M, or 1.1 GB at BF16. Full 128K context adds up to 1.8 GB (2.4 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 354M × 4.8 bits ÷ 8 = 0.2 GB

KV Cache + Overhead ≈ 0.4 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

KV Cache + Overhead ≈ 2.2 GB (at full 128K context)

VRAM usage by quantization

Q4_K_M

0.6 GB

Q4_K_M + full context

2.4 GB

Learn more about VRAM estimation →

What's the best quantization for LFM2.5 350M?

For LFM2.5 350M, Q4_K_M (0.6 GB) offers the best balance of quality and VRAM usage. Q5_K_S (0.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 0.5 GB.

VRAM requirement by quantization

IQ2_XXS

0.5 GB~53%

IQ3_XS

0.5 GB~73%

Q4_0

0.5 GB~85%

IQ4_NL

0.6 GB~88%

Q4_K_M ★

0.6 GB~89%

BF16

1.1 GB~100%

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run LFM2.5 350M on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run LFM2.5 350M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.6 GB of usable unified memory (RAM minus macOS overhead).

Can I run LFM2.5 350M locally?

Yes — LFM2.5 350M can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is LFM2.5 350M?

At Q4_K_M, LFM2.5 350M can reach ~8276 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~1130 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 0.58 × 0.65 = ~8966 tok/s

Estimated speed at Q4_K_M (0.6 GB)

NVIDIA B200

~8966 tok/s

NVIDIA GeForce RTX 4090

~1130 tok/s

NVIDIA B300

~8966 tok/s

AMD Instinct MI350X

~8276 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of LFM2.5 350M?

At Q4_K_M, the download is about 0.21 GB. The full-precision BF16 version is 0.71 GB. The smallest option (IQ2_XXS) is 0.10 GB.

Which GPUs can run LFM2.5 350M?

52 consumer GPUs can run LFM2.5 350M at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.

Which devices can run LFM2.5 350M?

163 devices with unified memory can run LFM2.5 350M at Q4_K_M (0.6 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.