[Model] Add native Dots3 NOTE multimodal support (#51255)

Signed-off-by: KurodaKanbei <mistergalahad@gmail.com>
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: Jee Jee Li <jeejeelee@inferact.ai>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
范裕达
2026-08-12 23:13:20 +08:00
committed by GitHub
co-authored by OpenAI Codex Jee Jee Li Jee Jee Li
parent b1b752042f
commit 9035151d6c
28 changed files with 6468 additions and 7 deletions
+3
View File
@@ -78,6 +78,7 @@ def test_mla_kv_cache_spec_uses_layer_cache_dtype(
kv_cache_dtype=cache_dtype,
head_size=576,
non_causal_multi_token_decode=False,
sliding_window=None,
)
vllm_config = SimpleNamespace(
cache_config=SimpleNamespace(block_size=64), model_config=None
@@ -460,6 +461,7 @@ class MockSparseMLAAttentionLayer:
self.qk_nope_head_dim = qk_nope_head_dim
self.qk_rope_head_dim = qk_rope_head_dim
self.v_head_dim = v_head_dim
self.q_lora_rank = None
self.kv_lora_rank = kv_lora_rank
# Compute weight matrices in the format expected by forward_impl
@@ -583,6 +585,7 @@ class MockMLAAttentionLayer(MLAAttention):
self.qk_nope_head_dim = qk_nope_head_dim
self.qk_rope_head_dim = qk_rope_head_dim
self.v_head_dim = v_head_dim
self.q_lora_rank = None
self.kv_lora_rank = kv_lora_rank
# Compute weight matrices from kv_b_proj (like MLAAttention does)