Skip to content

issue/429 - feat: adjust warmup #430

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
spike-zhu wants to merge 1 commit into
base: main
Choose a base branch
from
Open

issue/429 - feat: adjust warmup #430

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 24 additions & 13 deletions examples/bench.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -331,32 +331,43 @@ def run(
warmup_steps = 1

# warmup cache capacity
warmup_cache_len = 128
warmup_batch = len(test.input_ids_list)

test.model.reset_cache(
StaticKVCacheConfig(
warmup_case = next(iter(cases_dict.values()))
warmup_batch = warmup_case["batch_size"]
warmup_input_len = warmup_case["input_len"]
warmup_decode_len = 5

if enable_paged_attn:
warmup_num_blocks = (
(warmup_input_len + warmup_decode_len + paged_kv_block_size - 1)
// paged_kv_block_size
) * warmup_batch
warmup_cache_config = PagedKVCacheConfig(
warmup_num_blocks, paged_kv_block_size
)
else:
warmup_cache_config = StaticKVCacheConfig(
max_batch_size=warmup_batch,
max_cache_len=warmup_cache_len,
max_cache_len=warmup_input_len + warmup_decode_len,
)
)

avg_prompt_len = min(64, max(len(ids) for ids in test.input_ids_list))
test.model.reset_cache(warmup_cache_config)

warmup_ids = [
ids[:avg_prompt_len] if len(ids) >= avg_prompt_len else ids
for ids in test.input_ids_list
]
warmup_prompt_ids = repeat_prompt(test.input_ids_list[0], warmup_input_len)
warmup_ids = [warmup_prompt_ids] * warmup_batch

input_ids_infini = infinicore.from_list(warmup_ids, dtype=infinicore.int64)

print(
f"\033[93m[warmup] batch={warmup_batch}, input_len={warmup_input_len}, "
f"will prefill + {warmup_decode_len} decode steps\033[0m"
)
print("=================== warmup start ===================")

for _ in range(warmup_steps):
_ = test.model.generate(
input_ids_infini,
GenerationConfig(
max_new_tokens=5, # decode kernel warmup
max_new_tokens=warmup_decode_len, # decode kernel warmup
temperature=cfg.temperature,
top_k=cfg.top_k,
top_p=cfg.top_p,
Expand Down
Loading