[AMD][DSV4] Enable hicache on deepseek-v4 fp8 unified attn (#37778)

This commit is contained in:
Thomas Wang
2026-09-17 11:29:23 -07:00
committed by GitHub
parent 1f60ddef5d
commit 6c73368c32
9 changed files with 246 additions and 24 deletions
@@ -247,6 +247,8 @@ fn pool_name_str(name: PoolName) -> &'static str {
PoolName::DeepseekV4C2 => "deepseek_v4_c2",
PoolName::DeepseekV4C2Indexer => "deepseek_v4_c2_indexer",
PoolName::DeepseekV4C2IndexerScale => "deepseek_v4_c2_indexer_scale",
PoolName::DeepseekV4C4Rope => "deepseek_v4_c4_rope",
PoolName::DeepseekV4C128Rope => "deepseek_v4_c128_rope",
PoolName::DeepseekV4C4State => "deepseek_v4_c4_state",
PoolName::DeepseekV4C4IndexerState => "deepseek_v4_c4_indexer_state",
PoolName::DeepseekV4C128State => "deepseek_v4_c128_state",
@@ -273,6 +275,8 @@ fn parse_pool_name(name: &str) -> PyResult<PoolName> {
"deepseek_v4_c2" => Ok(PoolName::DeepseekV4C2),
"deepseek_v4_c2_indexer" => Ok(PoolName::DeepseekV4C2Indexer),
"deepseek_v4_c2_indexer_scale" => Ok(PoolName::DeepseekV4C2IndexerScale),
"deepseek_v4_c4_rope" => Ok(PoolName::DeepseekV4C4Rope),
"deepseek_v4_c128_rope" => Ok(PoolName::DeepseekV4C128Rope),
"deepseek_v4_c4_state" => Ok(PoolName::DeepseekV4C4State),
"deepseek_v4_c4_indexer_state" => Ok(PoolName::DeepseekV4C4IndexerState),
"deepseek_v4_c128_state" => Ok(PoolName::DeepseekV4C128State),
@@ -348,6 +348,8 @@ pub enum PoolName {
DeepseekV4C2,
DeepseekV4C2Indexer,
DeepseekV4C2IndexerScale,
DeepseekV4C4Rope,
DeepseekV4C128Rope,
DeepseekV4C4State,
DeepseekV4C4IndexerState,
DeepseekV4C128State,