Yes — with 32.6 GB to spare
Llama 4 Scout 109B-A17B at Q4_K_M fits your Ryzen AI Max+ 395 · 128 GB entirely in unified memory at 8K context, at an estimated 7.0 tokens per second. Past 703K the KV cache pushes it over — quantise the cache to q8_0, or step down a quantisation, to go longer.
A 10M-token window on paper, chunked attention in practice (8K chunks on 3 of 4 layers). Needs 64 GB+ at Q4.
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 107.9 GB | 110.0 GB | — | ~4.0 | −0.1% ppl | 14.0 GB over |
| Q6_K | 83.2 GB | 85.3 GB | 235K | 5.1 | −0.4% ppl | Long context |
| Q5_K_M | 71.9 GB | 74.0 GB | 476K | 5.9 | −0.8% ppl | Long context |
| Q4_K_M | 61.3 GB | 63.4 GB | 703K | 7.0 | −1.9% ppl | Recommended |
| Q3_K_M | 49.6 GB | 51.7 GB | 952K | 8.6 | −5.4% ppl | Long context |
| Q2_K | 42.5 GB | 44.6 GB | 1024K | 10 | −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. This model interleaves sliding-window layers (8192 tokens, 12 of 48 layers global), which is why its cache barely grows with context.
How to run it
$ ollama pull llama4:scout $ OLLAMA_CONTEXT_LENGTH=8192 \ ollama run llama4:scout
The default. One binary, a model registry, an OpenAI-compatible port. More on Ollama.