zzhxxx
|
eac72f5f23
|
[Feat] Flashcomm2 use o_shared linear (#4188)
### What this PR does / why we need it?
It is mentioned in the [flashcomm2 technical
report](https://gitcode.com/ascend-tribe/ascend-inference-cluster/blob/main/FlashComm/FlashComm2%E5%A4%A7%E6%A8%A1%E5%9E%8B%E6%8E%A8%E7%90%86%E4%B8%AD%E4%BB%A5%E5%AD%98%E6%8D%A2%E4%BC%A0%E7%9A%84%E9%80%9A%E4%BF%A1%E4%BC%98%E5%8C%96%E6%8A%80%E6%9C%AF.pdf)
that FC2 will introduce full redundant storage of the o_proj matrix,
which will put pressure on the memory. Therefore, the technical report
proposed a compromise solution using otp2, but it will introduce
additional reduce-scatter communication.
We propose a shared linear feature (#2931 ) that supports distributing
weights layer by layer to each card, avoiding the need for TP splitting,
and can solve the memory issue.
This PR depends on #3232 and #2931
### Flashcomm2 flowchart
<img width="1142" height="878" alt="PixPin_2025-11-14_13-37-39"
src="https://github.com/user-attachments/assets/d45ea8db-d8ef-4d45-8e18-abd4d82ce3e0"
/>
### Does this PR introduce _any_ user-facing change?
Use environment variables
```bash
export VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE=1
export VLLM_ASCEND_ENABLE_FLASHCOMM2_OSHARED=1
```
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: zzhx1 <zzh_201018@outlook.com>
Signed-off-by: zzhxx <2783294813@qq.com>
Co-authored-by: zzh02232027 <zzh02232027@antgroup.com>
Co-authored-by: clrs97 <524936896@qq.com>
Co-authored-by: Levi-JQ <yujinqi2@huawei.com>
|
2025-12-11 12:43:04 +08:00 |
|