[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
@@ -31,7 +31,8 @@ from ..utils import check_logprobs_close
],
)
@pytest.mark.parametrize(
"kv_cache_dtype", ["int8_per_token_head", "fp8_per_token_head"]
"kv_cache_dtype",
["int4_per_token_head", "int8_per_token_head", "fp8_per_token_head"],
)
@pytest.mark.parametrize("max_tokens", [4])
@pytest.mark.parametrize("enforce_eager", [True])