[CI] Organize speculative decoding E2E tests by coverage (#50330)
Signed-off-by: mgoin <mgoin64@gmail.com> Signed-off-by: Michael Goin <mgoin64@gmail.com> Co-authored-by: OpenAI Codex <noreply@openai.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
OpenAI Codex
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
b40d859c7b
commit
fcdc7c2e9c
@@ -19,6 +19,7 @@ def load_dspark_model(target_model: nn.Module, vllm_config: VllmConfig) -> nn.Mo
|
||||
|
||||
from vllm.compilation.backends import set_model_tag
|
||||
from vllm.model_executor.models.qwen3_dflash import dflash_has_any_non_causal
|
||||
from vllm.model_executor.models.utils import get_draft_quant_config
|
||||
|
||||
draft_vllm_config = replace(
|
||||
vllm_config,
|
||||
@@ -36,6 +37,9 @@ def load_dspark_model(target_model: nn.Module, vllm_config: VllmConfig) -> nn.Mo
|
||||
else vllm_config.cache_config
|
||||
),
|
||||
)
|
||||
# VllmConfig post-init restores the target's quant config because the target
|
||||
# config is retained for DSpark's target-layer metadata, so we must override it.
|
||||
draft_vllm_config.quant_config = get_draft_quant_config(vllm_config)
|
||||
|
||||
with set_model_tag("dspark_head"):
|
||||
draft_model = get_model(
|
||||
|
||||
Reference in New Issue
Block a user