Skip to content

GitLab

Commit 83e4e0fe authored by zhuwenwen's avatar zhuwenwen
Browse files

update triton fa

parent 37d73995
Show whitespace changes
Inline Side-by-side
Showing with 10 additions and 10 deletions
README.md
View file @ 83e4e0fe
...@@ -74,7 +74,7 @@ VLLM_INSTALL_PUNICA_KERNELS=1 python3 setup.py install ...@@ -74,7 +74,7 @@ VLLM_INSTALL_PUNICA_KERNELS=1 python3 setup.py install
+ 若使用 pip install 下载安装过慢，可添加源：-i https://pypi.tuna.tsinghua.edu.cn/simple/ + 若使用 pip install 下载安装过慢，可添加源：-i https://pypi.tuna.tsinghua.edu.cn/simple/
## 验证 ## 验证
- python -c "import vllm; print(vllm.\_\_version__)"，版本号与官方版本同步，查询该软件的版本号，例如0.4.3； - python -c "import vllm; print(vllm.\_\_version__)"，版本号与官方版本同步，查询该软件的版本号，例如0.5.0.post1；
## Known Issue ## Known Issue
- 无 - 无
......
vllm/attention/ops/triton_flash_attention.py
View file @ 83e4e0fe
...@@ -213,7 +213,7 @@ def _attn_fwd_inner( ...@@ -213,7 +213,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 256, "BLOCK_M": 256,
"BLOCK_N": 64, "BLOCK_N": 64,
"waves_per_eu": 2, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -223,7 +223,7 @@ def _attn_fwd_inner( ...@@ -223,7 +223,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 128, "BLOCK_M": 128,
"BLOCK_N": 128, "BLOCK_N": 128,
"waves_per_eu": 2, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -233,7 +233,7 @@ def _attn_fwd_inner( ...@@ -233,7 +233,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 256, "BLOCK_M": 256,
"BLOCK_N": 128, "BLOCK_N": 128,
"waves_per_eu": 2, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -243,7 +243,7 @@ def _attn_fwd_inner( ...@@ -243,7 +243,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 128, "BLOCK_M": 128,
"BLOCK_N": 64, "BLOCK_N": 64,
"waves_per_eu": 1, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -253,7 +253,7 @@ def _attn_fwd_inner( ...@@ -253,7 +253,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 128, "BLOCK_M": 128,
"BLOCK_N": 64, "BLOCK_N": 64,
"waves_per_eu": 3, "waves_per_eu": 0,
"PRE_LOAD_V": True, "PRE_LOAD_V": True,
}, },
num_stages=1, num_stages=1,
...@@ -263,7 +263,7 @@ def _attn_fwd_inner( ...@@ -263,7 +263,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 128, "BLOCK_M": 128,
"BLOCK_N": 64, "BLOCK_N": 64,
"waves_per_eu": 3, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -273,7 +273,7 @@ def _attn_fwd_inner( ...@@ -273,7 +273,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 64, "BLOCK_M": 64,
"BLOCK_N": 64, "BLOCK_N": 64,
"waves_per_eu": 4, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -283,7 +283,7 @@ def _attn_fwd_inner( ...@@ -283,7 +283,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 32, "BLOCK_M": 32,
"BLOCK_N": 32, "BLOCK_N": 32,
"waves_per_eu": 4, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
...@@ -296,7 +296,7 @@ def _attn_fwd_inner( ...@@ -296,7 +296,7 @@ def _attn_fwd_inner(
{ {
"BLOCK_M": 16, "BLOCK_M": 16,
"BLOCK_N": 16, "BLOCK_N": 16,
"waves_per_eu": 1, "waves_per_eu": 0,
"PRE_LOAD_V": False, "PRE_LOAD_V": False,
}, },
num_stages=1, num_stages=1,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment