Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
C
test(vulkan): Kimi K3's device-only upload failure keeps one host copy and every layer
JustVuggaa6fe739c8d8
diff --git a/c/tests/vulkan_partial_dsk.sh b/c/tests/vulkan_partial_dsk.shindex e8472224..dd363f61 100644--- a/c/tests/vulkan_partial_dsk.sh+++ b/c/tests/vulkan_partial_dsk.sh@@ -438,8 +438,19 @@ ptl_k3_cases() {ptl_k3_dho "partial kimi_k3 device only, staged" long 3 $O COLI_VK_STAGED=1CHAINMODE=2 RELOAD_MLA=1 ptl_k3_dho "partial kimi_k3 device only, prompts only" long 3 $OFAULT_BACK=3 REBUILD=1 ptl_k3_dho "partial kimi_k3 device only, device lost mid-decode" long 3 $O- ptl_k3_fault "partial kimi_k3 device only, an upload failing in layer 3" long 3 $O COLI_VK_DENSE_HOST=0- [ "$(dho_reloaded vk.log)" = 0 ] || { grep '^\[VK\]' vk.log; fail "partial kimi_k3 device only, an upload failing: read back from disk"; }+ # device only: the dense weights go up through the per-matrix placement before the chain+ # (k3_dho_place), not with each chain layer as deepseek_v41's; an upload failing there+ # keeps that matrix's host copy and the chain puts it up itself, so the run keeps every+ # layer: N = L, one host copy fewer dropped, nothing read back, the CPU's tokens+ local all fn L+ ptl_k3_probe ptl-probe.log long COLI_VK_STAGED=1 COLI_VK_STAGED_FAULT=submit:1000000 COLI_VK_CHAIN_LAYERS=99 $O COLI_VK_DENSE_HOST=0+ all=$(dho_dropped ptl-probe.log); L=$(ptl_L kimi_k3 ptl-probe.log)+ fn=$(ptl_calc fault kimi_k3 ptl-probe.log 3) || { grep '^\[VK\] kimi_k3 chain' ptl-probe.log; fail "partial kimi_k3 device only: no fault count from the probe"; }+ PTL_WANT=$L ptl_k3 "partial kimi_k3 device only, an upload failing in layer 3's placement (submit #$fn)" long "$L" \+ COLI_VK_STAGED=1 COLI_VK_STAGED_FAULT=submit:"$fn" $O COLI_VK_DENSE_HOST=0+ grep -q "staged upload failed (submit: .*): a matrix stays on the CPU" vk.log && [ "$(dho_dropped vk.log)" = $((all - 1)) ] &&+ [ "$(dho_reloaded vk.log)" = 0 ] ||+ { grep '^\[VK\]' vk.log; fail "partial kimi_k3 device only, an upload failing: not one host copy kept with every layer on the device"; }ptl_k3_cap "partial kimi_k3 device only, a device for 3 layers" long 3 $O COLI_VK_DENSE_HOST=0[ "$(dho_dropped vk.log)" = "$(ptl_k3_droppable 3)" ] && [ "$(dho_reloaded vk.log)" = 0 ] ||{ grep -E '^\[VK\]|^\[K3\]' vk.log; fail "partial kimi_k3 device only, a device for 3 layers: the host copies dropped are not the 3 layers'"; }