Yes — with 3.1 GB to spare
Fara 7B at Q4_K_M fits your GeForce RTX 3080 entirely on the GPU at 8K context, at an estimated 99 tokens per second. Past 64K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
Fully on GPU
8K context
Q4_K_M · 4.7 GB
MIT
Released 24 Nov 2025
Vision
Not in the Ollama library
A web computer-use agent on a Qwen2.5-VL base — it clicks, fills forms and stops for permission. Not a chat model; size it like an 8B with vision.
The VRAM budget
weights 4.7 GB
Weights 4.7 GB
KV cache @ 8K 0.44 GB
Runtime overhead 0.6 GB
Free 3.1 GB of 8.8 GB
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 15.4 GB | 16.5 GB | — | ~4.4 | Reference | 7.7 GB over |
| Q8_0 | 8.2 GB | 9.2 GB | — | ~34 | −0.1% ppl | 0.4 GB over |
| Q6_K | 6.3 GB | 7.4 GB | 33K | 73 | −0.4% ppl | Long context |
| Q5_K_M | 5.5 GB | 6.5 GB | 49K | 84 | −0.8% ppl | Long context |
| Q4_K_M | 4.7 GB | 5.7 GB | 64K | 99 | −1.9% ppl | Recommended |
| Q3_K_M | 3.8 GB | 4.8 GB | 80K | 122 | −5.4% ppl | Long context |
| Q2_K | 3.2 GB | 4.3 GB | 90K | 142 | −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.
How to run it
$ llama-server \
-hf microsoft/Fara-7B:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
01Download is 4.7 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 64K context on this card.