|
|
@@ -142,7 +142,7 @@ args = [
|
|
|
# SPECULATIVE config, not the global flag (eagle/utils.py:
|
|
|
# _make_eagle_draft_vllm_config) — B12X here mirrors the main model's.
|
|
|
"--speculative-config", """
|
|
|
-'{"method":"draft_model","model":"incoai/GLM-5.3-Flash-DFlash2","num_speculative_tokens":7,"attention_backend":"FLASH_ATTN"}'""",
|
|
|
+'{"method":"draft_model","model":"incoai/GLM-5.3-Flash-DFlash2","num_speculative_tokens":7,"attention_backend":"FLASH_ATTN","kv_cache_dtype":"auto"}'""",
|
|
|
"--language-model-only",
|
|
|
# MTP + CUDA graphs degrades OUTPUT QUALITY on the glm53 day-0 dev build
|
|
|
# (forum #381350/#44, CONFIRMED by A/B bench 2026-08-28: same seed/flags,
|
