Issues

• CUDA: Qwen3.5/3.6 attention caches on the core KV pool (#781 follow-up)

  ashhart/TensorFold#1156 · Filter by author jschmied opened on Oct 21, 2026
  1

• Qwen3.5-MoE on CUDA: load ExLlamaV3 (EXL3) checkpoints

  ashhart/TensorFold#1149 · Filter by author MiaAI-Lab opened on Oct 21, 2026

• Proposal: Qwen3.8-27B ModelOpt NVFP4 checkpoints on the native CUDA engine (both precision modes)

  ashhart/TensorFold#1144 · Filter by author jschmied opened on Oct 21, 2026

• Proposal: GLM-5.3-Flash on CUDA takes its streams' KV pages from the core pool, and kept prompt states copy only their rows

  ashhart/TensorFold#1142 · Filter by author Bizuayeu opened on Oct 21, 2026

• Proposal: dense BF16 Qwen3.5 CUDA scoring

  ashhart/TensorFold#1136 · Filter by author akol1 opened on Oct 21, 2026

• Tool arguments: oneOf / anyOf object parameters come back as JSON strings (Nemotron 3.5 Lightning, native 1.1.1)

  ashhart/TensorFold#1121 · Filter by author aartiles opened on Oct 20, 2026

• Chat prompts: control-token spellings in request text (user, system, tool results) become control tokens

  ashhart/TensorFold#1117 · Filter by author Acermax opened on Oct 20, 2026
  2

• Illegal address (Xid 31 FAULT_PDE write) after a NoRoom growth refusal mid-prefill: sequence KV unmapped without a GPU fence

  ashhart/TensorFold#1116 · Filter by author Idoo689 opened on Oct 20, 2026

• Linux: video is refused with libtfvideo.so in the server's lib directory, where capabilities finds it

  ashhart/TensorFold#1107 · Filter by author Bizuayeu opened on Oct 20, 2026

• CUDA kernel contract on nvcc 13.0: 656 pending launches have no 13.0 records, so kernel-contract and kernel-contract-limits fail on plain v1.1.1

  ashhart/TensorFold#1105 · Filter by author Bizuayeu opened on Oct 20, 2026

• HIP/Nemotron contribution stack: review map and validation

  ashhart/TensorFold#1081 · Filter by author BlivionIaG opened on Oct 19, 2026

• Linux desktop support using external OpenAI-compatible inference

  ashhart/TensorFold#1077 · Filter by author eXsoR65 opened on Oct 19, 2026
  1

• Feature Request: Unload model after idle timeout

  ashhart/TensorFold#1074 · Filter by author mikog8tor opened on Oct 19, 2026

• GLM-5.3-Flash on CUDA at TP=3: the default glm_mm path refuses KDA's input projection, and the first request fails with BadShape

  ashhart/TensorFold#1067 · Filter by author Bizuayeu opened on Oct 19, 2026
  1

• GLM-5.3-Flash (Metal): ~5% prefill improvement — a tensor-tile kernel for the MTP draft head's eh_proj, on by default

  ashhart/TensorFold#1065 · Filter by author EugeneClaw opened on Oct 19, 2026

• Nemotron 3.5 Super VL on two GB10s: TP=2 serve deadlocks with NCCL 2.28.9 (proxy slot exhaustion); 2.32.3 works

  ashhart/TensorFold#1060 · Filter by author plotarmordev opened on Oct 19, 2026
  1

• Shared rounds keep draft depths at their priors, so a cost-table change moved 4-stream depth by one draft (8-11% at 4 streams)

  ashhart/TensorFold#1059 · Filter by author plotarmordev opened on Oct 19, 2026
  1

• GLM-5.3 Metal path: DSA indexer scan 100-300 ms per 1K context (85x the CUDA rate of #259)

  ashhart/TensorFold#1054 · Filter by author Chris9121 opened on Oct 19, 2026
  2

• Flash Next 4-bit doesn't fit natively on a 128 GB Mac at default limits. Any plans for --ple-on-ssd on Metal?

  ashhart/TensorFold#1039 · Filter by author hudsonwa opened on Oct 18, 2026
  2

• App beta.5 / beta.7: a TUN fake-IP proxy (Clash Verge, mihomo) fails every Hub check — the address guard reads 198.18.0.0/15 as private, and the UI reports it as a connection problem

  ashhart/TensorFold#1037 · Filter by author samwang0041-star opened on Oct 18, 2026
  2

• GLM on CUDA, GB10: the native engine spins two CPU cores per rank, and long runs of replies overheat the CPU

  ashhart/TensorFold#1033 · Filter by author Bizuayeu opened on Oct 18, 2026
  2

• Flash Next on Metal (1.1.0): a decoding stream stops for the whole of another stream's prefill

  ashhart/TensorFold#1032 · Filter by author tpischke opened on Oct 18, 2026
  1

• Proposal: several models on one Mac under one memory budget

  ashhart/TensorFold#1027 · Filter by author cwschroeder opened on Oct 18, 2026
  1

• NVIDIA's GLM-5.3-Flash-FP4-Local on the native CUDA engine (tracker)

  ashhart/TensorFold#1015 · Filter by author Bizuayeu opened on Oct 17, 2026
  7

• Proposal: TensorFold Notes, memory inside the server (for 1.1.0)

  ashhart/TensorFold#1012 · Filter by author ThinkOffApp opened on Oct 17, 2026
  4