[Feature] Triton INT4 per-token-head KV cache quantization (#40835)

Signed-off-by: JartX <sagformas@epdcenter.es>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
JartX
2026-06-24 10:21:25 +00:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent d20dbf921b
commit 70749fdcca
10 changed files with 1462 additions and 111 deletions
+1
View File
@@ -39,6 +39,7 @@ STR_DTYPE_TO_TORCH_DTYPE = {
"fp8_e4m3": torch.uint8,
"fp8_e5m2": torch.uint8,
"int8": torch.int8,
"int4_per_token_head": torch.uint8,
"int8_per_token_head": torch.int8,
"fp8_per_token_head": torch.uint8,
"fp8_inc": torch.float8_e4m3fn,