[Model] Support Qwen3.8-Flash-Next (#53896)

Signed-off-by: huanghaoyan.hhy <huanghaoyan.hhy@alibaba-inc.com>
Signed-off-by: andyluo7 <andy.luo@amd.com>
Signed-off-by: lishunyang12 <lishunyang03@gmail.com>
Signed-off-by: Harry Huang <huanghaoyan.hhy@alibaba-inc.com>
Signed-off-by: Thien Tran <gau.nernst@yahoo.com.sg>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai>
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: zjy0516 <riverclouds.zhu@qq.com>
Co-authored-by: KungYork <firelikest@163.com>
Co-authored-by: Thien Tran <gau.nernst@yahoo.com.sg>
Co-authored-by: Yifan Qiao <yifanqiao@inferact.ai>
Co-authored-by: andyluo7 <43718156+andyluo7@users.noreply.github.com>
Co-authored-by: Isotr0py <mozf@inferact.ai>
Co-authored-by: andyluo7 <andy.luo@amd.com>
Co-authored-by: OpenAI Codex <noreply@openai.com>
Co-authored-by: lishunyang12 <lishunyang03@gmail.com>
Co-authored-by: Vadim Gimpelson <156319763+vadiklyutiy@users.noreply.github.com>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Lucas Wilkinson <lwilkins@redhat.com>
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: OpenAI Codex <codex@openai.com>
This commit is contained in:
Harry Huang
2026-08-31 05:57:56 +00:00
committed by GitHub
co-authored by zjy0516 KungYork Thien Tran Yifan Qiao andyluo7 Isotr0py andyluo7 OpenAI Codex lishunyang12 Vadim Gimpelson Isotr0py Lucas Wilkinson Claude Opus 5 OpenAI Codex
parent 687db59744
commit e126687a9a
124 changed files with 19424 additions and 450 deletions
+80
View File
@@ -839,6 +839,83 @@ class Qwen3_5ForCausalLMConfig(Qwen3_5ForConditionalGenerationConfig):
rope_parameters.pop("mrope_interleaved", None)
def _strip_qwen4_exp_mrope(model_config: "ModelConfig") -> None:
configs = {
id(config): config
for config in (
getattr(model_config, "hf_config", None),
model_config.hf_text_config,
)
if config is not None
}
for config in configs.values():
rope_parameters = getattr(config, "rope_parameters", None)
if rope_parameters is not None:
rope_parameters.pop("mrope_section", None)
rope_parameters.pop("mrope_interleaved", None)
class Qwen4ExpForConditionalGenerationConfig(Qwen3_5ForConditionalGenerationConfig):
"""Apply the Qwen3.5 hybrid-cache contract to Qwen4Exp."""
@staticmethod
def verify_and_update_config(vllm_config: "VllmConfig") -> None:
Qwen3_5ForConditionalGenerationConfig.verify_and_update_config(vllm_config)
text_config = vllm_config.model_config.hf_text_config
if text_config.hc_count <= 1:
raise ValueError("Qwen4Exp requires hc_count > 1")
parallel_config = vllm_config.parallel_config
uses_ple_or_qsa = bool(text_config.ple_layer_ids) or (
getattr(text_config, "indexer_n_heads", None) is not None
)
if uses_ple_or_qsa and (
parallel_config.enable_dbo or parallel_config.ubatch_size > 1
):
raise NotImplementedError(
"Qwen4Exp PLE/QSA does not support dual-batch overlap or microbatching"
)
# Checked again in Qwen4ExpModelState; rejecting it here keeps the
# engine from loading weights first.
if text_config.ple_layer_ids and parallel_config.pipeline_parallel_size > 1:
raise NotImplementedError(
"Qwen4Exp N-gram PLE embedding requires pipeline_parallel_size=1 "
"because non-first pipeline ranks do not receive the raw input_ids "
"it needs. Please run with PP=1."
)
multimodal_config = vllm_config.model_config.multimodal_config
if multimodal_config is not None and multimodal_config.language_model_only:
_strip_qwen4_exp_mrope(vllm_config.model_config)
spec_config = vllm_config.speculative_config
if spec_config is not None and spec_config.method not in {
"mtp",
"ngram",
"ngram_gpu",
}:
raise NotImplementedError(
"Qwen4Exp speculative decoding supports only its native MTP "
"checkpoint and linear n-gram proposers"
)
class Qwen4ExpForCausalLMConfig(Qwen4ExpForConditionalGenerationConfig):
@staticmethod
def verify_and_update_config(vllm_config: "VllmConfig") -> None:
Qwen4ExpForConditionalGenerationConfig.verify_and_update_config(vllm_config)
_strip_qwen4_exp_mrope(vllm_config.model_config)
class Qwen4ExpMTPConfig(Qwen4ExpForConditionalGenerationConfig):
"""Preserve MRoPE for a VL target and use 1D RoPE for a text target."""
@staticmethod
def verify_and_update_config(vllm_config: "VllmConfig") -> None:
Qwen4ExpForConditionalGenerationConfig.verify_and_update_config(vllm_config)
if hasattr(vllm_config.model_config.hf_config, "vision_config"):
return
_strip_qwen4_exp_mrope(vllm_config.model_config)
class ColQwen3_5Config(Qwen3_5ForConditionalGenerationConfig):
"""Apply the attention contract declared by a ColQwen3.5 checkpoint."""
@@ -963,6 +1040,9 @@ MODELS_CONFIG_MAP: dict[str, type[VerifyAndUpdateConfig]] = {
"Qwen3_5ForConditionalGeneration": Qwen3_5ForConditionalGenerationConfig,
"Qwen3_5MoeForCausalLM": Qwen3_5ForCausalLMConfig,
"Qwen3_5MoeForConditionalGeneration": Qwen3_5ForConditionalGenerationConfig,
"Qwen4ExpForCausalLM": Qwen4ExpForCausalLMConfig,
"Qwen4ExpForConditionalGeneration": (Qwen4ExpForConditionalGenerationConfig),
"Qwen4ExpMTP": Qwen4ExpMTPConfig,
"UnlimitedOCRForCausalLM": UnlimitedOCRForCausalLMConfig,
"VoyageQwen3BidirectionalEmbedModel": VoyageQwen3BidirectionalEmbedModelConfig,
"XLMRobertaModel": JinaRobertaModelConfig,