[Misc] Replace assert with proper exceptions for security and validation in pooling (#43286)

Signed-off-by: Taneem Ibrahim <taneem.ibrahim@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-authored-by: Nick Hill <nickhill123@gmail.com>
This commit is contained in:
Taneem Ibrahim
2026-05-22 18:43:33 +08:00
committed by GitHub
co-authored by Claude Opus 4.6 Nick Hill
parent d3d1cf6972
commit b3c7ffcab8
4 changed files with 37 additions and 17 deletions
+20 -6
View File
@@ -64,7 +64,11 @@ class PoolingMetadata:
for pooling_param in pooling_params
if (task := pooling_param.task) is not None
]
assert len(pooling_params) == len(tasks)
if len(pooling_params) != len(tasks):
raise ValueError(
"Every pooling param must have a task set, but got "
f"{len(tasks)} tasks for {len(pooling_params)} pooling params"
)
self.tasks = tasks
@@ -88,9 +92,11 @@ class PoolingMetadata:
self,
prompt_token_ids: torch.Tensor | None,
) -> list[torch.Tensor]:
assert prompt_token_ids is not None, (
"Please set `requires_token_ids=True` in `get_pooling_updates`"
)
if prompt_token_ids is None:
raise ValueError(
"prompt_token_ids is required but was not set. "
"Please set `requires_token_ids=True` in `get_pooling_updates`"
)
return [prompt_token_ids[i, :num] for i, num in enumerate(self.prompt_lens)]
def get_prompt_token_ids(self) -> list[torch.Tensor]:
@@ -101,7 +107,11 @@ class PoolingMetadata:
def get_pooling_cursor(self) -> PoolingCursor:
pooling_cursor = self.pooling_cursor
assert pooling_cursor is not None, "Should call `build_pooling_cursor` first"
if pooling_cursor is None:
raise RuntimeError(
"pooling_cursor has not been initialized. "
"Call `build_pooling_cursor` before accessing it"
)
return pooling_cursor
@@ -115,7 +125,11 @@ class PoolingMetadata:
n_seq = len(num_scheduled_tokens_np)
prompt_lens = self.prompt_lens
assert len(prompt_lens) == n_seq
if len(prompt_lens) != n_seq:
raise ValueError(
f"prompt_lens length ({len(prompt_lens)}) does not match "
f"the number of sequences ({n_seq})"
)
num_scheduled_tokens_cpu = torch.from_numpy(num_scheduled_tokens_np)
if query_start_loc_gpu is None: