[Bugfix] Fix Random Dataset Prefix Length Inaccuracy (#33907)
Signed-off-by: frankwang28 <frank.wbb@hotmail.com> Co-authored-by: Roger Wang <hey@rogerw.io>
This commit is contained in:
+29
-10
@@ -380,7 +380,7 @@ def gen_prompt_decode_to_target_len(
|
|||||||
max_retry: int = 10,
|
max_retry: int = 10,
|
||||||
add_special_tokens: bool = False,
|
add_special_tokens: bool = False,
|
||||||
rng: np.random.Generator | None = None,
|
rng: np.random.Generator | None = None,
|
||||||
) -> tuple[str, list[int]]:
|
) -> tuple[str, list[int], int]:
|
||||||
"""
|
"""
|
||||||
Ensure decoded-then-encoded prompt length matches the target token length.
|
Ensure decoded-then-encoded prompt length matches the target token length.
|
||||||
|
|
||||||
@@ -392,7 +392,9 @@ def gen_prompt_decode_to_target_len(
|
|||||||
[6880, 6881] -> ['Ġcalls', 'here'] ->
|
[6880, 6881] -> ['Ġcalls', 'here'] ->
|
||||||
[1650, 939, 486] -> ['Ġcall', 'sh', 'ere']
|
[1650, 939, 486] -> ['Ġcall', 'sh', 'ere']
|
||||||
|
|
||||||
Returns a tuple of the final prompt string and the adjusted token sequence.
|
Returns a tuple of the final prompt string, the adjusted token sequence,
|
||||||
|
and the token mismatch (final_len - target_token_len) if the retry budget
|
||||||
|
is exhausted.
|
||||||
"""
|
"""
|
||||||
remain_num_try = max_retry
|
remain_num_try = max_retry
|
||||||
token_mismatch = 0
|
token_mismatch = 0
|
||||||
@@ -499,7 +501,7 @@ class RandomDataset(BenchmarkDataset):
|
|||||||
allowed_tokens = np.array(list(set(all_tokens) - set(prohibited_tokens)))
|
allowed_tokens = np.array(list(set(all_tokens) - set(prohibited_tokens)))
|
||||||
|
|
||||||
# Generate prefix once
|
# Generate prefix once
|
||||||
prefix_token_ids = self.get_prefix(allowed_tokens, prefix_len)
|
prefix_token_ids = self.get_prefix(tokenizer, allowed_tokens, prefix_len)
|
||||||
|
|
||||||
requests = []
|
requests = []
|
||||||
token_mismatch_total = 0
|
token_mismatch_total = 0
|
||||||
@@ -554,19 +556,36 @@ class RandomDataset(BenchmarkDataset):
|
|||||||
|
|
||||||
def get_prefix(
|
def get_prefix(
|
||||||
self,
|
self,
|
||||||
|
tokenizer: TokenizerLike,
|
||||||
allowed_tokens: np.ndarray,
|
allowed_tokens: np.ndarray,
|
||||||
prefix_len: int,
|
prefix_len: int,
|
||||||
) -> list[int]:
|
) -> list[int]:
|
||||||
"""
|
"""
|
||||||
Get the prefix for the dataset.
|
Get the prefix for the dataset.
|
||||||
"""
|
"""
|
||||||
return (
|
if prefix_len <= 0:
|
||||||
allowed_tokens[
|
return []
|
||||||
self._rng.integers(0, len(allowed_tokens), size=prefix_len)
|
|
||||||
].tolist()
|
prefix_tokens = allowed_tokens[
|
||||||
if prefix_len > 0
|
self._rng.integers(0, len(allowed_tokens), size=prefix_len)
|
||||||
else []
|
].tolist()
|
||||||
|
_, adjusted_tokens, token_mismatch = gen_prompt_decode_to_target_len(
|
||||||
|
tokenizer=tokenizer,
|
||||||
|
token_sequence=prefix_tokens,
|
||||||
|
target_token_len=prefix_len,
|
||||||
|
add_special_tokens=False,
|
||||||
|
rng=self._rng,
|
||||||
)
|
)
|
||||||
|
if token_mismatch != 0:
|
||||||
|
sign = "more" if token_mismatch > 0 else "fewer"
|
||||||
|
logger.warning(
|
||||||
|
"Prefix tokenization produced %d %s tokens than expected "
|
||||||
|
"after decoding and re-encoding. This is expected due to "
|
||||||
|
"the imperfect nature of the sampling procedure",
|
||||||
|
abs(token_mismatch),
|
||||||
|
sign,
|
||||||
|
)
|
||||||
|
return adjusted_tokens
|
||||||
|
|
||||||
def get_sampling_params(
|
def get_sampling_params(
|
||||||
self,
|
self,
|
||||||
@@ -1128,7 +1147,7 @@ class RandomMultiModalDataset(RandomDataset):
|
|||||||
"Sampling from %d out of %d (vocab size)", len(allowed_tokens), vocab_size
|
"Sampling from %d out of %d (vocab size)", len(allowed_tokens), vocab_size
|
||||||
)
|
)
|
||||||
# Generate prefix once
|
# Generate prefix once
|
||||||
prefix_token_ids = self.get_prefix(allowed_tokens, prefix_len)
|
prefix_token_ids = self.get_prefix(tokenizer, allowed_tokens, prefix_len)
|
||||||
# Add synthetic multimodal items to each request
|
# Add synthetic multimodal items to each request
|
||||||
mm_requests = []
|
mm_requests = []
|
||||||
token_mismatch_total = 0
|
token_mismatch_total = 0
|
||||||
|
|||||||
Reference in New Issue
Block a user