Repository navigation

All pull requests

feat(qwen3_5): add dense vision load balancing
#43VenusZLR opened Oct 30, 2026Contributor
1
feat(moe): periodic expert re-placement across EP ranks
#42lifenghit opened Oct 30, 2026
perf(moe): reorder dispatch and combine chunks in place
#41lifenghit opened Oct 30, 2026
feat(lora): export standalone LoRA adapters
#40Glucose-180 opened Oct 30, 2026Contributor
1
feat: support freezing arbitrary modules via FQN wildcard patterns
#39casparcwang opened Oct 29, 2026Collaborator
feat: add glm5.3 flash model support
#38saherby opened Oct 28, 2026Contributor
feat: add Ling-3.1 training optimizations
#37lifenghit opened Oct 28, 2026
feat(kcp): add FLA context parallelism for Bailing KDA
#34Cruise20 opened Oct 27, 2026Collaborator
feat: force expert balance
#33zongshuai1 opened Oct 27, 2026Contributor
feat: support tp and etp for fsdp2 backend
#28dazhuangzhuang1024 opened Oct 23, 2026Collaborator
feat(ops): fused AscendC kda on fla-npu main @ d77c3738
#21ChenxiQ opened Oct 21, 2026Contributor
9
Perf/packed workload balance
#20zongshuai1 opened Oct 21, 2026Contributor
1
perf(moe): add an optional standalone MegaMoE backend for Ascend
#19zongshuai1 opened Oct 21, 2026Contributor
2
feat(engine): add --tt_profiling_config profiling detail knobs
#18ChenxiQ opened Oct 21, 2026Contributor
feat(deepseek-v4): port model and NPU operator updates
#17i-zhen opened Oct 21, 2026Collaborator
2
feat(dcp): route SavePlan metadata exchange over a gloo control-plane…
#10junejanejin opened Oct 19, 2026
reorder tokens by torch_npu.npu_moe_re_routing
#8zongshuai1 opened Oct 18, 2026Contributor
4
fix(kimi_k25): seed all_tied_weights_keys for transformers
#7junejanejin opened Oct 17, 2026
4
feat: add Kimi K3 (Kimi Linear) model support
#6junejanejin opened Oct 17, 2026
7

Pull requests