antirez 0accf8b395 Merge #1073: speed up DeepSeek V4.1 Flash on Metal ...

Integrate Tarjei Mandt (kernelpool) and contributors’ Metal decode, prefill and tensor-parallel work. Keep the existing four-row index scorer for short and non-FP4 inputs, and retain the Spark small-batch projection.

2026-11-02 11:09:09 +01:00

..

CLIENTS.md

Split README into platform setup and focused reference guides

2026-09-05 14:29:52 +02:00

CUDA_MULTI_GPU.md

Put common getting-started instructions in the README

2026-09-05 19:38:18 +02:00

DGX_SPARK.md

Document V4.1 DSpark conversion, usage and quality checks

2026-10-09 15:22:37 +02:00

DISTRIBUTED.md

Remove the DeepSeek V4.1 DSpark runtime before merging upstream's

2026-10-20 09:09:49 +11:00

METAL.md

DeepSeek v4.1 Flash support for Metal

2026-09-12 11:47:54 +02:00

MODELS.md

Add verified V4.1 Flash MXFP4 downloads and resumable assembly

2026-10-19 14:14:09 +02:00

PERFORMANCE.md

Split README into platform setup and focused reference guides

2026-09-05 14:29:52 +02:00

QWEN38_FLASH_NEXT.md

Document Qwen Metal batching and its quality checks

2026-09-16 15:17:26 +02:00

QWEN38_Q4_REGISTER_REUSE.md

metal: reuse Q4_K raw loads in Qwen NAX prefill

2026-10-30 17:56:53 +01:00

SERVER.md

Merge #1073: speed up DeepSeek V4.1 Flash on Metal

2026-11-02 11:09:09 +01:00

SPECULATIVE_DECODING.md

Remove the DeepSeek V4.1 DSpark runtime before merging upstream's

2026-10-20 09:09:49 +11:00

SSD_STREAMING.md

cuda: improve SSD-streamed prefill with ordered expert reads

2026-10-21 09:30:33 +02:00

STRIX_HALO.md

Document V4.1 DSpark conversion, usage and quality checks

2026-10-09 15:22:37 +02:00

TESTING.md

Split README into platform setup and focused reference guides

2026-09-05 14:29:52 +02:00