Issues / #1286

#1286 PCIe Issue with Strata and RTX 3060

open · @clawdsrlb-hub · 0 评论 · 在 GitHub 查看

BenchmarksServer & APINVIDIA / CUDAModels & quantsWindowsLinux

描述

I have Strata running on Mint Linux, with an RTX 3060 12 GB card. After running for several minutes Strata will fail with something like the below. I have updated my bios. I have updated my nvidia drivers. I have reseated the card. This is on model swift-iq3_xxs. I have tried two of the models but it always ends up doing this. Is there anything that I can do to fix it?

strata serve: expert tiers: GPU 81167 hits this request; since the start RAM 0 blobs, files 0 blobs 1003790.3 MB read
strata serve: suffix drafts: 4 windows, 10 of 12 drafts accepted
strata serve: prompt 30394 tokens = 0 reused + 30394 read in 65177 ms (466.3 tok/s), 9 generated in 495 ms (18.2 tok/s), drafts accepted 5 of 5, 3 checkpoints
strata serve: decode expert cache hit rate: 28.2% (1489 hits / 5280 lookups)
strata serve: expert tiers: GPU 1489 hits this request; since the start RAM 0 blobs, files 0 blobs 1209481.9 MB read
prefill mmq: gather_native_group: unspecified launch failure
strata generate: PCIe probe failed -> pcie_frac default 0.55
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native embedding: cannot pin 260 MiB, nor place it in VRAM
strata generate: PCIe probe failed -> pcie_frac default 0.55
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native embedding: cannot pin 260 MiB, nor place it in VRAM
strata generate: PCIe probe failed -> pcie_frac default 0.55
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native embedding: cannot pin 260 MiB, nor place it in VRAM
strata generate: PCIe probe failed -> pcie_frac default 0.55
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native embedding: cannot pin 260 MiB, nor place it in VRAM
strata generate: PCIe probe failed -> pcie_frac default 0.55
strata generate: this CPU has no AVX-512: the expert kernels run on AVX-2 (multi-token for the i-quant gate/up rows)
strata generate: native embedding: cannot pin 260 MiB, nor place it in VRAM

站内延伸阅读

链到安装、模型与版本说明,便于 SEO/GEO,非官方 issue 正文。