Why this quant
- 🗜️ 23.91 GB against 70.24 GB in BF16. The routed experts are 91.8% of this model's weight, so
quantizing them alone gets almost the whole reduction.
- 🎯 Nothing measurable is lost. 90.5 on the 1,170-item knowledge suite and 77.0 on the 200-item
tool-calling suite, against the BF16 original's 89.6 and 75.3 on the same box. The original is at
the bottom of the band, not the top.
- ⚡ 1,019 tok/s at concurrency 32, 1.53× the BF16 original's 664. Single-stream it runs 145 tok/s
against BF16's 141.
- 🧠 A 2.74M-token KV pool at 32K context, three times what BF16 leaves room for on the same card,
which is what lets one GPU hold 83 concurrent 32K sessions.
- 🧩 Stock vLLM, no patches.
compressed-tensors, nvfp4-pack-quantized, served by the released
0.28.0 image.
Serve it
hf download primitive-ai/Nex-N2.5-mini-NVFP4 --local-dir ./Nex-N2.5-mini-NVFP4
docker run --gpus all --ipc=host -p 8000:8000 -v $PWD:/models \
vllm/vllm-openai:v0.28.0 \
--model /models/Nex-N2.5-mini-NVFP4 \
--max-model-len 32768 --gpu-memory-utilization 0.92 \
--enable-auto-tool-choice --tool-call-parser qwen3_coder \
--reasoning-parser qwen3
Qwen3_5MoeForConditionalGeneration is in the 0.28.0 release, so no nightly is needed. The numbers
on this page were measured on 0.28.1rc1.dev472, and 0.28.0 serves the same weights with the same
kernels (FlashInfer CUTLASS for the NVFP4 experts). Boot takes about 210 s on an idle card.
Two things about this model's chat template are worth knowing before you wire it up. It emits an
empty thinking block by default, so replies arrive with two leading newlines and the reasoning
field holds only whitespace; a short correct answer looks like "\n\n4". And reasoning_effort
(low / medium / xhigh) is a per-request field here, unlike some other reasoning models where
it is fixed at server start.
Measured
One RTX PRO 6000 Blackwell, 96 GB, one card. The 1,170-item knowledge suite and the 200-item
tool-calling suite, temperature 0.6 / top_p 0.95 / top_k 20, thinking on, a 16,384-token
budget, concurrency 32, auto-scored with no LLM judge. Throughput is 8K in / 512 out, prefix-cache
free, two seeds per cell. Every row below was run on the same box in the same sitting.
Table with columns: build, size, knowledge, tool-calling, call, abstain, finished, tok/s @1, tok/s @32| build | size | knowledge | tool-calling | call | abstain | finished | tok/s @1 | tok/s @32 |
|---|
| BF16 original | 70.24 GB | 89.6 | 75.3 | 79.7 | 58.7 | 99.7% | 141.0 | 664 |
| FP8 experts |
Tool-calling is the mean of two runs per build (three for BF16). Pooled within-build spread is
0.60 standard deviations over those nine runs, and the four build means span 1.7, so the column is
one band. Read the ordering with that in mind: it runs opposite to precision, which no mechanism
would predict, and is the clearest evidence that this suite cannot resolve the difference between
4-bit and 16-bit experts on this model.
Speed does separate, and it crosses over. FP8 experts decode 13% faster single-stream because the
4-bit path pays for activation quantization and group scales on every token, a cost that does not
amortize at batch 1. At concurrency 32 the ranking inverts and NVFP4 wins by 13%, because there
the bottleneck is weight bandwidth and 4-bit moves half the bytes. Pick by the concurrency you
actually run.
Comparable with our other models
Accuracy numbers move for reasons that have nothing to do with the model: a shorter token budget, a
different temperature, or whether the model was allowed to reason at all. So every number in this
table, on this card and on our other cards, comes from the one fixed protocol described above, the
same 1,370 items, auto-scored, no LLM judge.
Table with columns: model, shape, size, overall, knowledge, call, abstain, finished, out/answer| model | shape | size | overall | knowledge | call | abstain | finished | out/answer |
|---|
| Laguna-XS-2.1 | 31 B MoE | 19.3 GiB | 81.7 | 83.8 | 68.4 | 73.5 | 98.9% | 1097 |
|
overall pools the two suites as 1,370 items, weighted 85.4% knowledge and 14.6% tool calling by
item count. Read it with finished: overall scores an answer that overran the token budget as
wrong, and cannot say whether the model needed the room or failed to stop. A gap under 1.0 in
overall is a tie. Sizes are as each card reports them, which mixes GB and GiB.
What's quantized to what
Nex-N2.5-mini is 35.11 B parameters, and they are distributed very unevenly:
Table with columns: params, share | params | share |
|---|
| routed experts, 40 layers × 256 × (gate+up+down) | 32.21 B | 91.8% |
| linear attention (Gated DeltaNet), 30 layers | 1.01 B | 2.9% |
embed_tokens and lm_head, untied, vocab 248320 | 1.02 B | 2.9% |
| vision tower, 27 blocks | 0.45 B | 1.3% |
| full attention, 10 layers | 0.27 B | 0.8% |
Table with columns: tensors, count, format| tensors | count | format |
|---|
| routed experts on all 40 layers | 30,720 modules | NVFP4, group 16 |
| everything else | 946 tensors | BF16, byte-identical to the source |
compressed-tensors, format nvfp4-pack-quantized, one config group, W4A4 with group-16 weight
scales and a per-module global scale. The source ships experts as two batched 3-D tensors per layer;
they are unfolded into per-expert modules, which is the layout vLLM's loader consumes. gate and up
share one weight_global_scale per expert. vLLM fuses those two halves and keeps a single scale,
warning and taking the maximum when they disagree, which silently shrinks half the weights, so the
build is audited for that before release: 20,480 fused groups checked, no mismatches.
Weights-only round-to-nearest, no calibration. Nothing in the quantization is fitted to data.