Skip to content

Disable Flash Attention with USE_FLASH_ATTENTION #692

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 2 commits into from
Aug 15, 2025
Merged

Disable Flash Attention with USE_FLASH_ATTENTION #692

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7ac2214
Disable Flash Attention with `USE_FLASH_ATTENTION`
alvarobartt Aug 6, 2025
cfcec32
Merge branch 'main' into disable-flash-attn-if-env
alvarobartt Aug 14, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions backends/candle/src/lib.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -423,6 +423,10 @@ impl CandleBackend {
if dtype != DType::F16
|| !cfg!(feature = "flash-attn")
|| get_runtime_compute_cap().unwrap() < 80
|| &std::env::var("USE_FLASH_ATTENTION")
.unwrap_or("True".to_string())
.to_lowercase()
!= "true"
{
return Err(BackendError::Start("Mistral is only supported on Cuda devices in fp16 with flash attention v2 enabled".to_string()));
}
Expand All @@ -435,6 +439,10 @@ impl CandleBackend {
(Config::Gte(config), Device::Cuda(_)) => {
if dtype != DType::F16
|| !cfg!(any(feature = "flash-attn", feature = "flash-attn-v1"))
|| &std::env::var("USE_FLASH_ATTENTION")
.unwrap_or("True".to_string())
.to_lowercase()
!= "true"
{
tracing::info!("Starting GTE model on {:?}", device);
Ok(Box::new(GTEModel::load(vb, &config, model_type).s()?))
Expand All @@ -447,6 +455,10 @@ impl CandleBackend {
(Config::Qwen2(config), Device::Cuda(_)) => {
if dtype != DType::F16
|| !cfg!(any(feature = "flash-attn", feature = "flash-attn-v1"))
|| &std::env::var("USE_FLASH_ATTENTION")
.unwrap_or("True".to_string())
.to_lowercase()
!= "true"
{
return Err(BackendError::Start("Qwen2 is only supported on Cuda devices in fp16 with flash attention v2 enabled".to_string()));
}
Expand All @@ -459,6 +471,10 @@ impl CandleBackend {
(Config::Qwen3(config), Device::Cuda(_)) => {
if dtype != DType::F16
|| !cfg!(any(feature = "flash-attn", feature = "flash-attn-v1"))
|| &std::env::var("USE_FLASH_ATTENTION")
.unwrap_or("True".to_string())
.to_lowercase()
!= "true"
{
tracing::info!("Starting Qwen3 model on {:?}", device);
Ok(Box::new(Qwen3Model::load(vb, &config, model_type).s()?))
Expand Down
Loading