forked from kkokosa/dotLLM
-
Notifications
You must be signed in to change notification settings - Fork 0
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
- Status: Open.#432 In jamesburton/dotLLM;
- Status: Open.#431 In jamesburton/dotLLM;
- Status: Open.#430 In jamesburton/dotLLM;
eval: sub-0.1% perplexity deltas are sequence artifacts — two deterministic harnesses disagree 5.7x on identical inputs
bugSomething isn't workingSomething isn't workingtestingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#429 In jamesburton/dotLLM;test(quant): IQ4_NL has no llama.cpp-transcription anchor — codebook interior is unpinned, and Nemotron GGUFs are ~91% IQ4_NL
testingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#428 In jamesburton/dotLLM;bug(vulkan): routed-bank F32 fallback reads a null f32Experts pointer for non-MLA quant-expert models (Q4_0/Q4_1/MXFP4) — silent corruption
bugSomething isn't workingSomething isn't workingvulkanVulkan GPU backendVulkan GPU backendStatus: Open.#427 In jamesburton/dotLLM;chore: three branches with unlanded work, open upstream issues and no PR (kvcache reserve-slot, tiled GEMM, skippable hardware tests)
enhancementNew feature or requestNew feature or requestStatus: Open.#425 In jamesburton/dotLLM;build: Microsoft.Build.Tasks.Git cannot read this repo's reftable ref storage — every local build errors
bugSomething isn't workingSomething isn't workingStatus: Open.#424 In jamesburton/dotLLM;test(quant): executable negative controls exist for 1 of 24 types; four K-quant anchors assert a single element
testingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#423 In jamesburton/dotLLM;test(models): generate a llama.cpp logits sidecar for qwen35/qwen35moe — the CPU oracle has no external anchor
architectureModel architecture supportModel architecture supporttestingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#422 In jamesburton/dotLLM;test(cuda/vulkan): three *ParityTests classes cannot detect the bug they exist for (no type assertions; corpus-length-only; gate-region-only)
bugSomething isn't workingSomething isn't workingtestingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#421 In jamesburton/dotLLM;test(cpu/cuda): GDN K-head broadcast is degenerate in every model-level test — the failure CLAUDE.md says already shipped
bugSomething isn't workingSomething isn't workingtestingTest infrastructure and coverageTest infrastructure and coverageStatus: Open.#420 In jamesburton/dotLLM;