pub enum FlashAttnError {
Dispatch(DispatchError),
Fa3RequiresHopper(SmArch),
Fp8MustUseFp8Request,
EmptyBatch,
ZeroSeqlen,
SeqlenOverflow,
EmptyPagedCache,
InvalidPagedBlockSize(u32),
PagedHeadDimMismatch {
cache: u32,
req: u32,
},
ChunkIndexOutOfRange {
index: u32,
total: u32,
},
MockMode,
}Expand description
Errors surfaced by the FlashAttention crate. Most are construction-
time validation failures; a small set are runtime launch errors
produced by the actor (and kept here so callers can pattern-match
without depending on the rest of atomr-accel-cuda).
Variants§
Dispatch(DispatchError)
Validation against the dispatch table failed.
Fa3RequiresHopper(SmArch)
A FlashAttention v3 request targeted a non-Hopper arch.
Fp8MustUseFp8Request
An fp8 dtype was passed to a non-fp8 request type, or vice versa.
EmptyBatch
Variable-length / paged batch is empty.
ZeroSeqlen
Sequence length is zero.
SeqlenOverflow
Cumulative seqlens overflow batch_size * max_seqlen.
EmptyPagedCache
Paged KV cache is empty / zero-sized.
InvalidPagedBlockSize(u32)
Paged KV-cache block size not in the supported set.
PagedHeadDimMismatch
Paged cache head_dim doesn’t match the request head_dim.
ChunkIndexOutOfRange
Chunked-prefill chunk index is out of range.
MockMode
Mock-mode actor saw a launch it can’t honour.
Trait Implementations§
Source§impl Clone for FlashAttnError
impl Clone for FlashAttnError
Source§fn clone(&self) -> FlashAttnError
fn clone(&self) -> FlashAttnError
1.0.0 (const: unstable) · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
source. Read moreSource§impl Debug for FlashAttnError
impl Debug for FlashAttnError
Source§impl Display for FlashAttnError
impl Display for FlashAttnError
Source§impl Error for FlashAttnError
impl Error for FlashAttnError
Source§fn source(&self) -> Option<&(dyn Error + 'static)>
fn source(&self) -> Option<&(dyn Error + 'static)>
1.0.0 · Source§fn description(&self) -> &str
fn description(&self) -> &str
use the Display impl or to_string()