Add Mistral Small 4 (Pixtral) support (#20708)

Signed-off-by: Xinyuan Tong <xinyuantong.cs@gmail.com>
Co-authored-by: Alex Nails <alexnails@radixark.ai>
Co-authored-by: Dimitrios Bariamis <12195802+dbari@users.noreply.github.com>
Co-authored-by: dbari <dbari@users.noreply.github.com>
This commit is contained in:
Xinyuan Tong
2026-03-18 14:15:32 -07:00
committed by GitHub
co-authored by Alex Nails Dimitrios Bariamis dbari
parent df1d046de2
commit 6b8a6545b2
18 changed files with 359 additions and 123 deletions
+5 -2
View File
@@ -1216,8 +1216,11 @@ class DeepseekV2AttentionMLA(
device=get_global_server_args().device,
)
if rope_scaling:
self.scaling = compute_mla_mscale_scaling(rope_scaling, self.scaling)
if rope_scaling and rope_scaling.get("apply_yarn_scaling", True):
mscale_all_dim = rope_scaling.get("mscale_all_dim", False)
scaling_factor = rope_scaling["factor"]
mscale = yarn_get_mscale(scaling_factor, float(mscale_all_dim))
self.scaling = self.scaling * mscale * mscale
else:
self.rotary_emb = None
self.use_deepseek_yarn_rope = rope_scaling is not None