biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Add the support for the qwen3 next model (a hybrid attention model). (#24526)

Browse Source 
Signed-off-by: Tao He <linzhu.ht@alibaba-inc.com>
Co-authored-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Tao He
2025-09-11 15:32:09 +08:00
committed by GitHub
parent 2048c4e379
commit e93f4cc9e3
29 changed files with 2476 additions and 61 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/model_executor/models/registry.py
View File

@@ -74,6 +74,7 @@ _TEXT_GENERATION_MODELS = {
"Gemma2ForCausalLM": ("gemma2", "Gemma2ForCausalLM"),
"Gemma3ForCausalLM": ("gemma3", "Gemma3ForCausalLM"),
"Gemma3nForCausalLM": ("gemma3n", "Gemma3nForCausalLM"),
"Qwen3NextForCausalLM": ("qwen3_next", "Qwen3NextForCausalLM"),
"GlmForCausalLM": ("glm", "GlmForCausalLM"),
"Glm4ForCausalLM": ("glm4", "Glm4ForCausalLM"),
"Glm4MoeForCausalLM": ("glm4_moe", "Glm4MoeForCausalLM"),
@@ -285,6 +286,7 @@ _SPECULATIVE_DECODING_MODELS = {
"ErnieMTPModel": ("ernie_mtp", "ErnieMTP"),
"Glm4MoeMTPModel": ("glm4_moe_mtp", "Glm4MoeMTP"),
"MedusaModel": ("medusa", "Medusa"),
"Qwen3NextMTP": ("qwen3_next_mtp", "Qwen3NextMTP"),
# Temporarily disabled.
# # TODO(woosuk): Re-enable this once the MLP Speculator is supported in V1.
# "MLPSpeculatorPreTrainedModel": ("mlp_speculator", "MLPSpeculator"),