A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 2, 14:45 UTC
[BUGFIX] Multimodal benchmark with MistralTokenizer (#42965)
Signed-off-by: juliendenize <julien.denize@mistral.ai> Signed-off-by: Julien Denize <40604584+juliendenize@users.noreply.github.com>
Julien Denize committed
May 28, 2026 at 14:36 UTC
02606b0b09c0b35d1c7e888f1679502c479a0ac0
1 file changed
+14
-10
vllm/benchmarks/datasets/datasets.py
+14
-10
@@ -48,6 +48,7 @@ from vllm.multimodal.image import convert_image_mode
48
from vllm.tokenizers import TokenizerLike
49
from vllm.utils.argparse_utils import FlexibleArgumentParser
50
from vllm.utils.import_utils import PlaceholderModule
51
+from vllm.utils.mistral import is_mistral_tokenizer
52
53
try:
54
from datasets import load_dataset
@@ -1219,16 +1220,19 @@ class RandomMultiModalDataset(RandomDataset):
1220
)
1221
1222
vocab_size = tokenizer.vocab_size
1222
- # Can't use tokenizer.all_special_ids since
1223
- # it returns ONLY ids from special_tokens_map.json
1224
- # We want to exclude placeholder tokens and all
1225
- # tokens that indicate start/end of image as it
1226
- # may break prompt replacement logic.
1227
- prohibited_tokens = list(
1228
- tok_id
1229
- for tok_id, token in tokenizer.added_tokens_decoder.items()
1230
- if token.special
1231
- )
1223
+ if is_mistral_tokenizer(tokenizer):
1224
+ prohibited_tokens = tokenizer.all_special_ids
1225
+ else:
1226
+ # Can't use tokenizer.all_special_ids since
1227
+ # it returns ONLY ids from special_tokens_map.json
1228
+ # We want to exclude placeholder tokens and all
1229
+ # tokens that indicate start/end of image as it
1230
+ # may break prompt replacement logic.
1231
+ prohibited_tokens = list(
1232
+ tok_id
1233
+ for tok_id, token in tokenizer.added_tokens_decoder.items()
1234
+ if token.special
1235
+ )
1236
all_tokens = np.arange(vocab_size)
1237
allowed_tokens = np.array(list(set(all_tokens) - set(prohibited_tokens)))
1238
logger.debug(