Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
docs(qwen36 tier): Ollama reference row for the placement calibration
kreuzzelg181d722790cb
diff --git a/docs/qwen36-cuda-tier.md b/docs/qwen36-cuda-tier.mdindex 300d90cf..5cf6dceb 100644--- a/docs/qwen36-cuda-tier.md+++ b/docs/qwen36-cuda-tier.md@@ -82,6 +82,7 @@ decode, same prompt, output bit-identical in all four runs:| same card, budget capped at 5 GB (a 6 GB card's share), warm | 88.9 % / 9.50 | 81.2 % / **13.15** || Quadro RTX 4000 (8 GB) alone, warm | 93.6 % / 11.09 | 88.4 % / **16.55** || both cards, experts on both, warm | 100 % / 10.79 | 100 % / **14.80** |+| reference: Ollama 0.32.5, same model Q4_K_M, same prompt, both cards (57 % CPU / 43 % GPU, 11.8 GB VRAM) | 20.3 warm (21.3 cold) | |The warm row is the one that matters: at a 95 % hit rate the marginal expertis as valuable as it gets on this card, and the trunk still wins by a third.@@ -89,7 +90,11 @@ The hit rate drops only 4.4 points for 796 fewer residents because thedisplaced experts are the coldest of the heat order -- exactly the ones theplacer priced as cheap. The smaller or slower the card, the larger therelative win: +34 % on the 3070, +38 % at a 5 GB budget, +49 % on the Quadro.-All sixteen runs of this calibration produced bit-identical text.+All sixteen runs of this calibration produced bit-identical text. Against+Ollama on the same box the gap closes from 1.5× (14.97 vs 22.4 in August, two+cards, hand-placed) to **1.23× on a single 8 GB card** (16.55 vs 20.3) --+with Ollama holding its dense weights at ~0.56 bytes per weight (Q4_K_M)+against this engine's 1.0 (int8), and using both cards.**Two cards are the open case.** With experts on both cards the second cardpaces every layer (the slower `take()` gates the chain), so `off` on two cards