Issues

• Qwen3.8-27B on pre-M5 Macs: 32 rows a pass for one stream

  ashhart/TensorFold#1621 · Filter by author ashhart opened on Oct 29, 2026
  2

• Flash Next on HIP: which tensors an oQ checkpoint can narrow for bandwidth-bound decode

  ashhart/TensorFold#1617 · Filter by author gilby opened on Oct 29, 2026

• Flash Next CUDA: a shared round's sampled streams draft one at a time, so sampled chat at 5 streams decodes at half the greedy rate on a GB10

  ashhart/TensorFold#1608 · Filter by author jrb-ai opened on Oct 29, 2026
  2

• HIP Flash Next: the prompt's n-gram gather is a single-threaded host copy

  ashhart/TensorFold#1606 · Filter by author gilby opened on Oct 29, 2026

• HIP: verify windows past 4 rows and past 32 rows fall off the stream and K-parallel tiles

  ashhart/TensorFold#1605 · Filter by author gilby opened on Oct 29, 2026
  1

• HIP decode: the 4-bit stream tile for MTP verify rows (scale/bias layout, lanes at K=640, row reuse)

  ashhart/TensorFold#1604 · Filter by author gilby opened on Oct 29, 2026
  1

• HIP Flash Next: about 850 small kernels per verify round cost a fifth of decode

  ashhart/TensorFold#1603 · Filter by author gilby opened on Oct 29, 2026

• HIP Flash Next: the first request after load decodes at half speed

  ashhart/TensorFold#1602 · Filter by author gilby opened on Oct 29, 2026

• HIP: a 16/32-row routed-expert prompt tile for many-expert MoE (Flash Next)

  ashhart/TensorFold#1601 · Filter by author gilby opened on Oct 29, 2026

• HIP: the prompt GEMM pays a group-32 scale and bias fold every K stage

  ashhart/TensorFold#1600 · Filter by author gilby opened on Oct 29, 2026

• HIP Flash Next: a shared expert wider than the routed experts cannot load

  ashhart/TensorFold#1599 · Filter by author gilby opened on Oct 29, 2026

• HIP Flash Next: the loader requires the MTP head under language_model.mtp, so oMLX checkpoints fail

  ashhart/TensorFold#1598 · Filter by author gilby opened on Oct 29, 2026

• HIP Flash Next: WMMA sparse attention for prompt chunks on gfx11

  ashhart/TensorFold#1597 · Filter by author gilby opened on Oct 29, 2026
  2

• MCDMA over a ConnectX link, and serve --cluster on real hosts (transport only)

  ashhart/TensorFold#1594 · Filter by author ThinkOffApp opened on Oct 29, 2026

• Living Weights: lessons are refused when the server runs with --no-thinking

  ashhart/TensorFold#1587 · Filter by author ashhart opened on Oct 29, 2026

• glm cuda: load an Apache-2.0 DFlash2 drafter (canada-quant DFlash2-G: 8 full-attention layers, 9 taps)?

  ashhart/TensorFold#1583 · Filter by author gilby opened on Oct 29, 2026
  2

• glm cuda: FP4-Local's routed experts in their declared NVFP4 (W4A4) math, opt-in?

  ashhart/TensorFold#1582 · Filter by author gilby opened on Oct 29, 2026
  2

• d1-2B (LFM3-VL decisions) on native CUDA: bf16 text decoder, LFM3-VL tower, set attention

  ashhart/TensorFold#1581 · Filter by author drowzeys opened on Oct 29, 2026
  1

• Backlog: land open pull requests and fix open issues

  ashhart/TensorFold#1578 · Filter by author ashhart opened on Oct 29, 2026
  2

• moss_image: serve pictures over POST /v1/images/generations

  ashhart/TensorFold#1573 · Filter by author BobClawblaw opened on Oct 29, 2026

• GLM Metal 1.1.8: high idle memory use and repeated prefills, restart releases 44.5 GiB

  ashhart/TensorFold#1572 · Filter by author wSzki opened on Oct 29, 2026
  2

• GLM-5.3-Flash on CUDA: prompt chunks of 4,096 rows cut cold first-token time 13% at 32K at TP=3 (and stack with #1423)

  ashhart/TensorFold#1570 · Filter by author gilby opened on Oct 28, 2026
  1

• Living Weights on DeepSeek V4.1 Flash (CUDA and Metal)

  ashhart/TensorFold#1569 · Filter by author ashhart opened on Oct 28, 2026
  2

• Living Weights on Qwen3.8-27B on Metal

  ashhart/TensorFold#1568 · Filter by author ashhart opened on Oct 28, 2026
  3

• Living Weights on GLM-5.3-Flash (CUDA and Metal)

  ashhart/TensorFold#1567 · Filter by author ashhart opened on Oct 28, 2026
  3