Yes — with 188.2 GB to spare

LFM2.5 2.6B at Q4_K_M fits your Instinct MI300X entirely on the GPU at 8K context, at an estimated 2113 tokens per second. There is room for its full 128K window.

Fully on GPU 8K context Q4_K_M · 1.5 GB LFM Open License v1.0 Released 28 Jul 2026 Not in the Ollama library

Convolution-heavy hybrid for CPUs and NPUs: 22 of 30 blocks keep no KV cache at all.

Hardware requirements What hardware do I need to run LFM2.5 2.6B? →

At Q4_K_M with an 8K context: a 6 GB GPU (RTX 3060 Laptop) or a 8 GB Mac (M3). Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.

The VRAM budget

weights 1.5 GB
Weights 1.5 GB KV cache @ 8K 0.13 GB Runtime overhead 0.6 GB Free 188.2 GB of 190.4 GB

Quantisation ladder

QuantWeightsTotal @ 8KMax contextTok/sQualityFit
F16 5.0 GB 5.8 GB 128K 638 Reference Long context
Q8_0 2.7 GB 3.4 GB 128K 1201 −0.1% ppl Long context
Q6_K 2.1 GB 2.8 GB 128K 1556 −0.4% ppl Long context
Q5_K_M 1.8 GB 2.5 GB 128K 1800 −0.8% ppl Long context
Q4_K_M 1.5 GB 2.2 GB 128K 2113 −1.9% ppl Recommended
Q3_K_M 1.2 GB 2.0 GB 128K 2611 −5.4% ppl Long context
Q2_K 1.1 GB 1.8 GB 128K 3047 −15% ppl Long context

Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. Only 8 of its 30 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state.

How to run it

terminal
$ llama-server \
    -hf LiquidAI/LFM2.5-2.6B:Q4_K_M \
    -c 8192 -ngl 99

The engine underneath most of the others. Every knob is exposed. More on llama.cpp.

01Download is 1.5 GB. Keep it on an SSD — a first load off a spinning disk takes minutes.
02Close anything else holding VRAM. A browser with hardware acceleration can sit on 1–2 GB.
03There is room to go to the model's full 128K context on this card.
See all models for this rig Compare with another model

Related models on an Instinct MI300X

ModelTotalTok/s est.
LFM2.5 8B-A1B
8.47B MoE · same family
Q4_K_M · 5.5 GB 1463 Runs great
Llama 3.2 3B Instruct
3.21B · similar size
Q4_K_M · 3.3 GB 1778 Runs great
Qwen3.5 2B
2.27B · similar size
Q4_K_M · 2.0 GB 2514 Runs great
Granite 4.1 3B
3.4B · similar size
Q4_K_M · 3.1 GB 1678 Runs great
Phi-4-mini 3.8B
3.84B · similar size
Q4_K_M · 3.8 GB 1486 Runs great
Ministral 3 3B
3.85B · similar size
Q4_K_M · 3.6 GB 1482 Runs great
Qwen3 4B
4.02B · similar size
Q4_K_M · 4.0 GB 1419 Runs great

Every model against the Instinct MI300X →

LFM2.5 2.6B on other hardware

DeviceTotalTok/s est.
GeForce RTX 3060 Laptop
5.3 GB usable · 336 GB/s
2.2 GB 134 Runs great
GeForce RTX 4060 Laptop
7.0 GB usable · 256 GB/s
2.2 GB 102 Runs great
M3 · 8 GB
5.4 GB usable · 100 GB/s
2.2 GB 37 Runs great
GeForce RTX 3060 12 GB
10.6 GB usable · 360 GB/s
2.2 GB 144 Runs great
M4 · 16 GB
10.7 GB usable · 120 GB/s
2.2 GB 44 Runs great
GeForce RTX 4090
22.4 GB usable · 1008 GB/s
2.2 GB 402 Runs great
Radeon RX 7900 XTX
22.4 GB usable · 960 GB/s
2.2 GB 383 Runs great
M3 Max · 64 GB
48.0 GB usable · 400 GB/s
2.2 GB 147 Runs great

All 114 devices, by tier →