feat: DFlash support for tensor-parallel Qwen3.6 targets - #533

Merged

davide221 merged 12 commits into

Luce-Org:mainfrom

jkyamog:feat-qwen35-dflash-tensor-parallel

Aug 3, 2026

Commits on Jul 30, 2026

vendor(ggml): add selected-device tensor parallel support

jkyamog
authored and
davide221
committed
2200b33
feat(qwen35): add target tensor parallelism

jkyamog
authored and
davide221
committed
971962e
fix(meta): preserve repeated split layouts

jkyamog
authored and
davide221
committed
35b7dca
fix(qwen35): validate tensor parallel placement

jkyamog
authored and
davide221
committed
50e0b65
fix(cuda): validate communicator devices and preserve API

jkyamog
authored and
davide221
committed
458274c
fix(tp): preserve validation and add NCCL smoke coverage
davide221
committed
4fdfb84
docs(vendor): record tensor parallel GGML source
davide221
committed
d538e6c
feat(dflash): support tensor-parallel targets

jkyamog
authored and
davide221
committed
af5f63e
perf(dflash): keep TP speculative state on device

jkyamog
authored and
davide221
committed
60100a1
perf(dflash): batch TP feature mirror transfers

jkyamog
authored and
davide221
committed
9793094
fix(dflash): address tensor parallel review feedback

jkyamog
authored and
davide221
committed
03f28ad
fix(tp): bound feature sync memory and propagate failures
davide221
committed
06b9600