Yes — with 18.9 GB to spare
Ling 3.0 Tiny 7.9B-A1.3B at Q4_K_M fits your M4 · 32 GB entirely in unified memory at 8K context, at an estimated 43 tokens per second. There is room for its full 128K window.
An 8B MoE with 1.3B active and a latent KV cache on only 6 of 24 layers — reasoning and tool use sized for Apple Silicon and edge boxes.
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 14.7 GB | 15.4 GB | 128K | 13 | Reference | Long context |
| Q8_0 | 7.8 GB | 8.5 GB | 128K | 24 | −0.1% ppl | Long context |
| Q6_K | 6.0 GB | 6.7 GB | 128K | 32 | −0.4% ppl | Long context |
| Q5_K_M | 5.2 GB | 5.9 GB | 128K | 36 | −0.8% ppl | Long context |
| Q4_K_M | 4.4 GB | 5.1 GB | 128K | 43 | −1.9% ppl | Recommended |
| Q3_K_M | 3.6 GB | 4.2 GB | 128K | 53 | −5.4% ppl | Long context |
| Q2_K | 3.1 GB | 3.7 GB | 128K | 62 | −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 6 of its 24 blocks keep a per-token KV cache; the rest are linear-attention, Mamba or convolution blocks with a fixed-size state. This model uses multi-head latent attention, so its cache is a compressed latent rather than full K and V.
How to run it
$ pip install mlx-lm $ mlx_lm.generate --model mlx-community/Ling-3.0-tiny-4bit \ --max-tokens 512 --prompt "Hello"
Apple's own array framework. The fastest path on Apple Silicon. More on MLX.