Yes — with 69.0 GB to spare
Ling 3.0 Flash 124B-A5B at Q4_K_M fits your H200 SXM entirely on the GPU at 8K context, at an estimated 390 tokens per second. There is room for its full 256K window.
A 124B hybrid (5 linear-attention layers per MLA layer) with 5.1B active: SWE-bench Pro 56.6 and AIME 93 claimed. Built for 96–128 GB machines.
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| Q8_0 | 122.7 GB | 123.4 GB | 256K | 221 | −0.1% ppl | Long context |
| Q6_K | 94.7 GB | 95.4 GB | 256K | 287 | −0.4% ppl | Long context |
| Q5_K_M | 81.8 GB | 82.5 GB | 256K | 332 | −0.8% ppl | Long context |
| Q4_K_M | 69.7 GB | 70.4 GB | 256K | 390 | −1.9% ppl | Recommended |
| Q3_K_M | 56.4 GB | 57.1 GB | 256K | 481 | −5.4% ppl | Long context |
| Q2_K | 48.4 GB | 49.0 GB | 256K | 562 | −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 7 of its 42 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
$ llama-server \
-hf inclusionAI/Ling-3.0-flash:Q4_K_M \
-c 8192 -ngl 99
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.