When running FLM2-VL-450M models on ios with flutter, it will throw the error Expected single token encoding for <|image_start|> at
|
throw std::runtime_error("Expected single token encoding for " + token_text); |
.
It looks like the tokenizer cannot recognize the image start token
verified the downloaded model, we do have <|image_start|> in tokenizer.json with special: true
{
"id": 499,
"content": "<|image_end|>",
"single_word": false,
"lstrip": false,
"rstrip": false,
"normalized": false,
"special": true
},
Model version: https://huggingface.co/Cactus-Compute/LFM2-VL-450M/commit/35f3e4525551681aa9c044cf653587337151b792
Cactus version: synced to head(10d04d7)
When running FLM2-VL-450M models on ios with flutter, it will throw the error
Expected single token encoding for <|image_start|>atcactus/cactus/models/model_lfm2vl.cpp
Line 226 in 10d04d7
It looks like the tokenizer cannot recognize the image start token
verified the downloaded model, we do have
<|image_start|>in tokenizer.json withspecial: trueModel version: https://huggingface.co/Cactus-Compute/LFM2-VL-450M/commit/35f3e4525551681aa9c044cf653587337151b792
Cactus version: synced to head(10d04d7)