update

compression_app.py

@@ -37,7 +37,7 @@ The encoding and decoding process can be formulated as
 - **Lossless** <br>
 Lossless tokenization preserves the exact original text, i.e. `decoded_text = input_text`.
 
-  - Most lossy tokenizers get many out-of-vocabulary tokens. 👉 Check the [oov of bert-base-uncased](https://huggingface.co/spaces/eson/tokenizer-arena/raw/main/stats/google-bert.bert-base-cased%20@%20cc100.zh-Hans.diff.json).
+  - Most lossy tokenizers get many out-of-vocabulary tokens. 👉 Check the [oov of bert-base-uncased](https://huggingface.co/spaces/eson/tokenizer-arena/blob/main/stats/compression_rate/google-bert.bert-base-cased%20%40%20cc100.zh-Hans.diff.json).
   - Some other tokenizers have no oov, but still be lossy due to text normalization. For example qwen performs [unicode normalization](https://github.com/huggingface/transformers/blob/v4.42.3/src/transformers/models/qwen2/tokenization_qwen2.py#L338),
     which may bring some [slight difference](https://huggingface.co/spaces/eson/tokenizer-arena/raw/main/stats/compression_rate/Qwen.Qwen1.5-1.8B%20@%20cc100.ja.diff.json) to the reconstructed text. 
 

requirements.txt

@@ -1,4 +1,4 @@
-gradio>=4.32.0
+gradio>=4.38.1
 transformers
 sentencepiece
 tiktoken