onwx' (onwx-dash) 1.1.0: 2-PC runs Nemotron 3.5 Bolt 74B-A9B, the model on the main PC only (streamed to the helper), faster 2-PC prompts and generation, 2-PC Linux only (part 4)
onwx' (onwx-dash) 1.1.0: 2-PC runs Nemotron 3.5 Bolt 74B-A9B, the model on the main PC only (streamed to the helper), faster 2-PC prompts and generation, 2-PC Linux only (part 3)
onwx' (onwx-dash) 1.1.0: 2-PC runs Nemotron 3.5 Bolt 74B-A9B, the model on the main PC only (streamed to the helper), faster 2-PC prompts and generation, 2-PC Linux only (part 2)
onwx' (onwx-dash) 1.1.0: 2-PC runs Nemotron 3.5 Bolt 74B-A9B, the model on the main PC only (streamed to the helper), faster 2-PC prompts and generation, 2-PC Linux only
onwx' (onwx-dash) 0.8.9: Linux GPU+NPU prompt reading faster than the GPU alone again (spilling GEMM tile fixed), NPU buffers shared as dma-buf (part 3)
onwx' (onwx-dash) 0.8.9: Linux GPU+NPU prompt reading faster than the GPU alone again (spilling GEMM tile fixed), NPU buffers shared as dma-buf (part 2)
onwx' (onwx-dash) 0.8.8: Linux NPU's help falls back to GPU-owned shared buffers when the GPU cannot register the NPU's buffers (hipHostRegister failed) (part 3)
onwx' (onwx-dash) 0.8.8: Linux NPU's help falls back to GPU-owned shared buffers when the GPU cannot register the NPU's buffers (hipHostRegister failed) (part 2)
onwx' (onwx-dash) 0.8.8: Linux NPU's help falls back to GPU-owned shared buffers when the GPU cannot register the NPU's buffers (hipHostRegister failed)
onwx' (onwx-dash) 0.8.7: Linux NPU's help loads the PC's own XRT (was 'No such library libxrt_core.so.2'); README / setup.sh steps for XRT and the memlock limit (part 2)
onwx' (onwx-dash) 0.8.7: Linux NPU's help loads the PC's own XRT (was 'No such library libxrt_core.so.2'); README / setup.sh steps for XRT and the memlock limit
onwx' (onwx-dash) 0.8.6: faster GPU decode (bit-identical; Qwen3.6 +3% on Ryzen AI Max), the GPU alias of NPU-shared buffers reserved on Windows (Strata #3431) (part 4)
onwx' (onwx-dash) 0.8.6: faster GPU decode (bit-identical; Qwen3.6 +3% on Ryzen AI Max), the GPU alias of NPU-shared buffers reserved on Windows (Strata #3431) (part 3)
onwx' (onwx-dash) 0.8.6: faster GPU decode (bit-identical; Qwen3.6 +3% on Ryzen AI Max), the GPU alias of NPU-shared buffers reserved on Windows (Strata #3431) (part 2)
onwx' (onwx-dash) 0.8.6: faster GPU decode (bit-identical; Qwen3.6 +3% on Ryzen AI Max), the GPU alias of NPU-shared buffers reserved on Windows (Strata #3431)
onwx' (onwx-dash) 0.8.5: updates with a model loaded no longer stop at the engine DLL in use (renamed aside); a half-done update from 0.8.4 or older is finished on the next start (part 2)
onwx' (onwx-dash) 0.8.5: updates with a model loaded no longer stop at the engine DLL in use (renamed aside); a half-done update from 0.8.4 or older is finished on the next start
onwx' (onwx-dash) 0.8.2: Unsloth's IQ4_XS / IQ3_XXS / IQ2_M Qwen3.6-35B-A3B GGUFs in the model list; model card: a port of gufo, Windows, GPU + NPU fixes; English card linked (part 3)
onwx' (onwx-dash) 0.8.2: Unsloth's IQ4_XS / IQ3_XXS / IQ2_M Qwen3.6-35B-A3B GGUFs in the model list; model card: a port of gufo, Windows, GPU + NPU fixes; English card linked (part 2)
onwx' (onwx-dash) 0.8.2: Unsloth's IQ4_XS / IQ3_XXS / IQ2_M Qwen3.6-35B-A3B GGUFs in the model list; model card: a port of gufo, Windows, GPU + NPU fixes; English card linked
