Mengqing Cao 367edff5af [HybridKV] Fix prefill disaggregation kvcache addr alignment & use hybrid kv cache only when running qwen3_next (#3007) ...

### What this PR does / why we need it?
This pr fixes a few issues on prefill disaggregation:
1. Fix prefill disaggregation kvcache addr alignment issue, llmdatadist
needs the addr of tensors to be aligned with 2M
2. Fix prefill disaggregation kvcache shape error, llmdatadist requires
k/v tensors with shape [num_blocks, ...], however the implentment before
this pr is [2, num_blocks, ...], which will break prefill disaggregation
3. Use hybrid kv cache only when running qwen3_next to fix accuracy
issue on prefill disaggregation.

### Does this PR introduce _any_ user-facing change?
N/A

### How was this patch tested?
Tested locally by @liziyu179 

- vLLM version: v0.10.2
- vLLM main:
4f02b77de4

---------

Signed-off-by: MengqingCao <cmq0113@163.com>

2025-09-18 21:43:22 +08:00

..

attention

Remove chunked_prefill_for_mla and fix ring_mla bug (#2781)

2025-09-18 19:43:26 +08:00

compilation

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

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

core

main add ascend scheduler support multimodal (#2844)

2025-09-14 09:38:51 +08:00

device_allocator

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

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

distributed

[Feat] A Connector that supports Mooncake store (#2913)

2025-09-18 14:04:45 +08:00

eplb

Dynamic Expert Load Balance with Zero-like-overhead (#2956)

2025-09-17 10:36:43 +08:00

lora

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

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

models

[main] addrmsnorm + quant fusion optim in Dense Models (#2772)

2025-09-16 22:31:38 +08:00

multistream

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

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

ops

refactor linear (#2867)

2025-09-18 14:09:19 +08:00

patch

[refactor] refactor deepseek-related files (#2849)

2025-09-16 14:13:07 +08:00

quantization

Dynamic Expert Load Balance with Zero-like-overhead (#2956)

2025-09-17 10:36:43 +08:00

sample

[main] add pd transfer for ascend scheduler (#2753)

2025-09-10 08:46:39 +08:00

spec_decode

[Feat][Graph] Support MTP for ACL Graph (#2932)

2025-09-18 14:05:33 +08:00

torchair

Remove chunked_prefill_for_mla and fix ring_mla bug (#2781)

2025-09-18 19:43:26 +08:00

worker

[HybridKV] Fix prefill disaggregation kvcache addr alignment & use hybrid kv cache only when running qwen3_next (#3007)

2025-09-18 21:43:22 +08:00

__init__.py

Bump torch version to 2.7.1 (#1562)

2025-08-05 08:43:24 +08:00

ascend_config.py

Add an option of enable frozen parameter (#2869)

2025-09-17 12:00:44 +08:00

ascend_forward_context.py

[main] addrmsnorm + quant fusion optim in Dense Models (#2772)

2025-09-16 22:31:38 +08:00

envs.py

[Ops] Fix bug in register_custom_ops without forward_context (#2883)

2025-09-12 16:58:08 +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

[HybridKV] Fix prefill disaggregation kvcache addr alignment & use hybrid kv cache only when running qwen3_next (#3007)

2025-09-18 21:43:22 +08:00

utils.py

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

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