NeverRaR 807686dec9 perf : optimize memory for deepseek mtp (#2713) ...

### What this PR does / why we need it?
delete the temp tensor to optimize memory for deepseek mtp for torchair
case

- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0

Signed-off-by: boying <897013703@qq.com>

2025-10-23 15:52:17 +08:00

..

attention

[Fix] Fixes attribute error in MLA implementation (#3618)

2025-10-23 09:12:50 +08:00

compilation

[Feat]Make full graph mode compalible with MTP (#3276)

2025-10-17 20:19:56 +08:00

core

[Feat] Dynamic Batch Feature (#3490)

2025-10-22 14:13:32 +08:00

device_allocator

[Misc]Clean up useless import from vllm (#2049)

2025-07-28 16:01:59 +08:00

distributed

Mooncake store use adxl inferface (#3350)

2025-10-21 20:18:17 +08:00

eplb

[CI]Add EPLB CI. (#3568)

2025-10-21 22:58:02 +08:00

kv_offload

[Feat] add native kvcache offload (#3433)

2025-10-22 14:15:49 +08:00

lora

[Bugfix][LoRA] Fix forward error and shape mismatch when using LoRA (#3153)

2025-09-28 17:30:50 +08:00

models

Revert "[Feat] Shared expert dp for deepseek and deepseek_mtp (#3495)" (#3586)

2025-10-21 22:24:30 +08:00

multistream

[Quickfix] update CachedRequestState as NewRequestData changed (#2367)

2025-08-15 07:35:27 +08:00

ops

[main][refactor] refactor SequenceRowParallelOp forward (#3616)

2025-10-23 14:41:15 +08:00

patch

[BugFix][main] Fix quantization related mtp bug with patch (#3620)

2025-10-23 09:54:31 +08:00

quantization

Reapply "[MoE] [Refactor] Remove manual memory cleanup (#3365)" (#3483) (#3512)

2025-10-22 11:41:30 +08:00

sample

Drop 0.10.2 (#3284)

2025-10-09 10:28:38 +08:00

spec_decode

unify logic between aclgraph and torchair (#3560)

2025-10-22 21:52:57 +08:00

torchair

perf : optimize memory for deepseek mtp (#2713)

2025-10-23 15:52:17 +08:00

worker

[Fix] Fix attention metadata handling for profiling and MLA (#3636)

2025-10-23 09:35:18 +08:00

__init__.py

[Refactor] Adapt deepseek-v3.2 to vllm 0.11.0 (#3432)

2025-10-15 17:48:58 +08:00

ascend_config.py

[Feat] Dynamic Batch Feature (#3490)

2025-10-22 14:13:32 +08:00

ascend_forward_context.py

[Bugfix] fix logging and d2h bug for flash comm1 (#3505)

2025-10-17 21:13:41 +08:00

cpu_binding.py

[main] support cpu binding (#3546)

2025-10-21 09:17:03 +08:00

envs.py

[Feat] Flash comm allgher ep (#3334)

2025-10-15 19:36:32 +08:00

meta_registration.py

Fix the bugs about operator registration by PyTorch Dispatcher (#2786)

2025-09-13 11:58:52 +08:00

platform.py

[Feat] Dynamic Batch Feature (#3490)

2025-10-22 14:13:32 +08:00

utils.py

[Bugfix][MTP] Fix performance degradation when mtp>1 (#3597)

2025-10-22 22:04:43 +08:00