Support fixed ubatch batch sweep at 16K context
This commit is contained in:
1 parent
dfdd4bb6a8
commit
dc69e2b4b4
2 files changed
+25
-22
No files matched your search
@@ -65,12 +65,10 @@ TEMPLATE_FILE="$REPO_ROOT/scripts/chat-template.jinja"
|
||||
|
||||
@test "large ubatch bench defaults to the active model and requested sweep" {
|
||||
local bench="$REPO_ROOT/scripts/bench-ubatch-rocmi4.sh"
|
||||
grep -q ': "${UBATCH_MIN:=256}"' "$bench"
|
||||
grep -q ': "${UBATCH_MAX:=6144}"' "$bench"
|
||||
grep -q 'UBATCH_VALUES.*256 512 768 1024 1280 1536 2048 3072 4096 5120 6144' "$bench"
|
||||
grep -q ': "${UBATCH_FIXED:=512}"' "$bench"
|
||||
grep -q 'BATCH_VALUES.*512 1024 1536 2048' "$bench"
|
||||
grep -q ': "${STOPSVC:=1}"' "$bench"
|
||||
grep -q ': "${CTX:=32768}"' "$bench"
|
||||
grep -q ': "${BENCH_BATCH:=32768}"' "$bench"
|
||||
grep -q ': "${CTX:=16384}"' "$bench"
|
||||
grep -q 'MODEL_FILE' "$bench"
|
||||
grep -q 'REASONING_FORMAT' "$bench"
|
||||
}
|
||||
@@ -82,6 +80,13 @@ TEMPLATE_FILE="$REPO_ROOT/scripts/chat-template.jinja"
|
||||
grep -qF 'eval time =\s' "$bench"
|
||||
}
|
||||
|
||||
@test "large ubatch bench supports fixed ubatch batch sweep" {
|
||||
local bench="$REPO_ROOT/scripts/bench-ubatch-rocmi4.sh"
|
||||
grep -q ': "${UBATCH_FIXED:=512}"' "$bench"
|
||||
grep -q 'BATCH_VALUES' "$bench"
|
||||
grep -q -- '-b "$batch" -ub "$ubatch"' "$bench"
|
||||
}
|
||||
|
||||
@test "frank-default-large uses the tracked Qwen-Sharp template on :7770" {
|
||||
grep -q '^CHAT_TEMPLATE_FILE=' "$ENV_FILE"
|
||||
grep -q -- '--jinja' "$SERVE_FILE"
|
||||
|
||||
Reference in new issue
Block a user