Issues

• [Feature] Could you add timestamps to the output logs?

  mw00/project-maya#783 · Filter by author SunYong0821 opened on Nov 2, 2026

• Speed report: 2x V100-SXM2 16GB, dual E5-2696 v4, 128GB DDR3-1866, DeepSeek V4.1 ISTA IQ2_XXS (24.6 tok/s)

  mw00/project-maya#782 · Filter by author c200762620-max opened on Nov 2, 2026

• Windows + HIP: the dashboard shows the wrong GPUs when HIP doesn't see every AMD adapter (DXGI order vs HIP numbering)

  mw00/project-maya#780 · Filter by author Meistertech-JP opened on Nov 2, 2026
  1

• Windows, one 20 GB GPU: --parallel 2 gives two real sessions (+14% aggregate); --parallel 3 collapses

  mw00/project-maya#779 · Filter by author pietroMastro92 opened on Nov 2, 2026

• ngram table keeps downloading for both Maya's

  mw00/project-maya#776 · Filter by author anbow1 opened on Nov 1, 2026
  1

• add high context for glm

  mw00/project-maya#769 · Filter by author TingTingin opened on Nov 1, 2026

• hardcoded logit_bias specified in server command line

  mw00/project-maya#768 · Filter by author vbooka1 opened on Nov 1, 2026
  1

• [Speed Report + Unfilled VRAM issue] DS4.1 1017 flash UD_Q8_K_XL (4x rtx 3090 + Threadripper + 504GB DDR4)

  mw00/project-maya#766 · Filter by author bandito1 opened on Nov 1, 2026
  1

• A Blender/Unreal assistant on one 48 GB GPU: DeepSeek Q8, 26/26 at 41 tok/s

  mw00/project-maya#765 · Filter by author talisp opened on Nov 1, 2026
  1

• Speed report: RX 7900 XTX 24 GB (gfx1100, ROCm 10.2), DeepSeek-V4.1-Flash Maya-XS 64K + DSpark - 18.4 tok/s decode, 395 tok/s prefill (v1.1.46)

  mw00/project-maya#757 · Filter by author Cayaf89 opened on Nov 1, 2026

• Speed report: RTX 5070 12 GB + EPYC 7302 (4-channel DDR4), Maya-S24 - 16.6 -> 24.2 tok/s decode, 323 -> 428 prefill; DFlash2 ~5 tok/s on one GPU + one socket

  mw00/project-maya#755 · Filter by author hugou74130 opened on Nov 1, 2026

• Speed report: AC922 (2x POWER9, 6x V100-16GB) - 69.6 tok/s decode, 1M context, Maya-S24 built from source on ppc64le

  mw00/project-maya#754 · Filter by author jefflaichunyin opened on Nov 1, 2026
  1

• [bug] Maya-XS: every request fails with "cudaEventRecord(d.hc_done, d.st2): an illegal memory access" on 4x RTX 2080 Ti (sm_75) — v1.1.39 and v1.1.46

  mw00/project-maya#750 · Filter by author Fox-B1ack opened on Nov 1, 2026
  1

• Windows: DeepSeek vision works on v1.1.45 with a pack rebuilt by tools/ds4/vision_pack.py (v1.1.44 crashed the CUDA context on every picture)

  mw00/project-maya#744 · Filter by author pietroMastro92 opened on Nov 1, 2026

• Request new Exl3 quant: DSV4.1-Flash-1017-SAGE-EXL3-1.62bpw

  mw00/project-maya#740 · Filter by author BahamutRU opened on Oct 31, 2026

• DeepSeek: make the 8 GiB system-RAM budgeting margin configurable

  mw00/project-maya#739 · Filter by author davedgd opened on Oct 31, 2026

• Windows 11 speed report: MiniMax-M3.1-Flash Maya-S on RTX PRO 6000 Blackwell 96 GB — 48.6 tok/s decode, 757 tok/s prefill (8K)

  mw00/project-maya#738 · Filter by author Andrey0710 opened on Oct 31, 2026

• Windows 11 speed report: DeepSeek-V4.1-Flash Maya-S on RTX PRO 6000 Blackwell 96 GB — 72.3 tok/s decode, 1757 tok/s prefill (8K)

  mw00/project-maya#737 · Filter by author Andrey0710 opened on Oct 31, 2026
  7

• [bug] DS4 HIP: reloading DeepSeek fails while copying CPU-lane experts from model files

  mw00/project-maya#735 · Filter by author Biotrioo opened on Oct 31, 2026
  2

• TP on two cards: the second card keeps ~1.5-2 GB free — is asymmetric expert residency possible, or is that expected?

  mw00/project-maya#734 · Filter by author tonykee opened on Oct 31, 2026

• DeepSeek: the automatic layer split balances streamed bytes, not time, so short prompts wait on the slow-link GPUs

  mw00/project-maya#733 · Filter by author mgabor3141 opened on Oct 31, 2026

• DeepSeek: v1.1.42's slow-link ring rule makes fresh prompt reads ~8% slower on 3090s on PCIe 3.0 x8 (#609)

  mw00/project-maya#731 · Filter by author mgabor3141 opened on Oct 31, 2026

• Windows 11 speed report: RTX PRO 6000 Blackwell 96 GB, Maya-L — 48.8 tok/s decode, 3810 tok/s prefill (8K)

  mw00/project-maya#730 · Filter by author Andrey0710 opened on Oct 31, 2026

• [bug] loop-guard retry overflows context by (note - slack) -> "reply was cut short" stubs

  mw00/project-maya#727 · Filter by author sexavier opened on Oct 31, 2026
  3

• [bug]Codex CLI 0.160.0 fails against Maya's Responses API

  mw00/project-maya#726 · Filter by author zhouxiaoxing91 opened on Oct 31, 2026
  2