[Bugfix][Model] Fix K3 DSpark config for 96-head drafts (#54005)

Signed-off-by: zixi-qi <zixi@inferact.ai>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
qizixi
2026-08-27 15:58:28 -07:00
committed by GitHub
co-authored by OpenAI Codex
parent 7c877062ac
commit e5377e6727
2 changed files with 19 additions and 8 deletions
@@ -18,8 +18,8 @@ def _write_dspark_config(path, **overrides):
"hidden_size": 7168,
"intermediate_size": 14336,
"num_hidden_layers": 5,
"num_attention_heads": 64,
"num_key_value_heads": 64,
"num_attention_heads": 96,
"num_key_value_heads": 96,
"q_lora_rank": 1536,
"kv_lora_rank": 512,
"qk_nope_head_dim": 128,
@@ -97,9 +97,20 @@ def test_dspark_mla_rejects_unsupported_checkpoint_options(tmp_path, overrides):
get_config(draft_path, trust_remote_code=False)
def test_dspark_mla_uses_latent_kv_geometry(tmp_path):
@pytest.mark.parametrize(
("num_attention_heads", "expected_local_heads"),
[(64, 8), (96, 12)],
ids=["released-64-head", "block5-96-head"],
)
def test_dspark_mla_uses_latent_kv_geometry(
tmp_path, num_attention_heads, expected_local_heads
):
draft_path = tmp_path / "draft"
_write_dspark_config(draft_path)
_write_dspark_config(
draft_path,
num_attention_heads=num_attention_heads,
num_key_value_heads=num_attention_heads,
)
model_config = ModelConfig(
model=str(draft_path),
tokenizer_mode="skip",
@@ -116,7 +127,7 @@ def test_dspark_mla_uses_latent_kv_geometry(tmp_path):
tensor_parallel_size=8, distributed_executor_backend="external_launcher"
)
assert model_config.get_num_kv_heads(parallel_config) == 1
assert model_config.get_num_attention_heads(parallel_config) == 8
assert model_config.get_num_attention_heads(parallel_config) == expected_local_heads
assert model_config.get_num_experts() == 0
+3 -3
View File
@@ -1,10 +1,10 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
from transformers import DeepseekV2Config
from transformers import DeepseekV3Config
class K3DSparkConfig(DeepseekV2Config):
class K3DSparkConfig(DeepseekV3Config):
"""Configuration for a dense MLA DSpark draft model."""
model_type = "k3_dspark"
@@ -18,7 +18,7 @@ class K3DSparkConfig(DeepseekV2Config):
rope_theta: float = 50000.0,
**kwargs,
) -> None:
# DeepseekV2Config defaults to a MoE topology. Zero these fields so
# DeepseekV3Config defaults to a MoE topology. Zero these fields so
# generic vLLM config logic also recognizes this draft as dense.
kwargs.setdefault("n_routed_experts", 0)
kwargs.setdefault("n_shared_experts", 0)