[CI] Fit small KV-offload evals within shared memory (#52496)
Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
This commit is contained in:
@@ -133,6 +133,7 @@ MODELS = [
|
||||
connector="OffloadingConnector",
|
||||
# Baseline ~0.49 on 200 questions (measured on GB200).
|
||||
accuracy_threshold=0.39,
|
||||
cpu_offload_gib=1,
|
||||
),
|
||||
OffloadingModelConfig(
|
||||
id="offloading-gemma-4-e4b-it",
|
||||
@@ -140,6 +141,7 @@ MODELS = [
|
||||
connector="OffloadingConnector",
|
||||
# Baseline ~0.64 on 200 questions (measured on GB200).
|
||||
accuracy_threshold=0.55,
|
||||
cpu_offload_gib=1,
|
||||
),
|
||||
OffloadingModelConfig(
|
||||
id="offloading-qwen3.5-35b",
|
||||
|
||||
Reference in New Issue
Block a user