[fix] baseline4 yaml
This commit is contained in:
@@ -34,7 +34,7 @@ command:
|
|||||||
- half
|
- half
|
||||||
env:
|
env:
|
||||||
- name: VLLM_ENGINE_ITERATION_TIMEOUT_S
|
- name: VLLM_ENGINE_ITERATION_TIMEOUT_S
|
||||||
value: 3600
|
value: '3600'
|
||||||
# --- MoE kernel selection ---
|
# --- MoE kernel selection ---
|
||||||
- name: BI100_MOE_COREX_DIRECT_ROUTED
|
- name: BI100_MOE_COREX_DIRECT_ROUTED
|
||||||
value: '1'
|
value: '1'
|
||||||
@@ -43,17 +43,10 @@ env:
|
|||||||
# --- GDN kernel selection ---
|
# --- GDN kernel selection ---
|
||||||
- name: BI100_GDN_COREX_PACKED_DECODE
|
- name: BI100_GDN_COREX_PACKED_DECODE
|
||||||
value: '1'
|
value: '1'
|
||||||
- name: BI100_GDN_COMBINED_QK_NORM
|
|
||||||
value: '1'
|
|
||||||
# --- Hybrid KV/GDN cache ---
|
# --- Hybrid KV/GDN cache ---
|
||||||
- name: BI100_HYBRID_KV_ACCOUNTING
|
- name: BI100_HYBRID_KV_ACCOUNTING
|
||||||
value: full_attention
|
value: full_attention
|
||||||
- name: BI100_GDN_CACHE_POLICY
|
- name: BI100_GDN_CACHE_POLICY
|
||||||
value: admission64
|
value: admission64
|
||||||
- name: BI100_GDN_RESTORE_MODE
|
- name: BI100_GDN_RESTORE_MODE
|
||||||
value: hybrid64
|
value: hybrid64
|
||||||
# --- Profiling (event mode: no sync overhead) ---
|
|
||||||
- name: BI100_PROFILE
|
|
||||||
value: '1'
|
|
||||||
- name: BI100_PROFILE_MODE
|
|
||||||
value: event
|
|
||||||
Reference in New Issue
Block a user