Add disclaimers

app.py

@@ -91,15 +91,18 @@ def creativity(index_desc, query):
 with gr.Blocks() as demo:
     with gr.Column():
         gr.HTML(
-            '''<h1 text-align="center">Creativity Index</h1>
+            f'''<h1 text-align="center">Creativity Index</h1>
 
             <p style='font-size: 16px;'>Compute the <a href="">Creativity Index</a> of a piece of text.</p>
-            <p style='font-size: 16px;'>The computed Creativity Index is based on verbatim match and is supported by <a href="https://infini-gram.io">infini-gram</a>.</p>
+            <p style='font-size: 16px;'>The Creativity Index is computed based on verbatim matching against massive text corpora and is powered by <a href="https://infini-gram.io">infini-gram</a>. It is defined as the ratio of tokens not covered by n-grams (n >= L) that can be found in the corpus, averaged across {NGRAM_LEN_MIN} <= L <= {NGRAM_LEN_MAX}. You can view the covered tokens (highlighted in red background) for each value of L.</p>
+            <p style='font-size: 16px;'><b>Note:</b> The input text is limited to {MAX_QUERY_CHARS} characters. Each query has a timeout of {MAX_TIMEOUT_IN_SECONDS} seconds. If you have waited 30 seconds and receive an error, you can try submitted the same query and it's more likely to work on the second try.</p>
+            <p style='font-size: 16px;'><b>Disclaimer 1:</b> The Creativity Index of text that appear exactly in the corpora may be deflated. In our paper, we remove exact duplicates (including quotations and citations) from the corpus before computing the Creativity Index. However, deduplication is not applied in this demo.</p>
+            <p style='font-size: 16px;'><b>Disclaimer 2:</b> The Creativity Index of text generated by latest models (e.g., GPT-4) may be inflated. This is because we don't have all the data that these models are trained on, and our supported corpora have a earlier cutoff date (Dolma-v1.7 is Oct 2023, RedPajama is Mar 2023, Pile is 2020).</p>
             '''
         )
         with gr.Row():
             with gr.Column(scale=1, min_width=240):
-                index_desc = gr.Radio(choices=INDEX_DESCS, label='Corpus', value=INDEX_DESCS[0])
+                index_desc = gr.Radio(choices=INDEX_DESCS, label='Corpus', value=INDEX_DESCS[2])
 
             with gr.Column(scale=3):
                 creativity_query = gr.Textbox(placeholder='Enter a piece of text here', label='Input', interactive=True, lines=10)
@@ -112,8 +115,8 @@ with gr.Blocks() as demo:
                 creativity_ci = gr.Label(value='', label='Creativity Index')
                 creativity_htmls = []
                 for n in range(NGRAM_LEN_MIN, NGRAM_LEN_MAX + 1):
-                    with gr.Tab(label=f'n={n}'):
-                        creativity_htmls.append(gr.HTML(value='', label=f'n={n}'))
+                    with gr.Tab(label=f'L={n}'):
+                        creativity_htmls.append(gr.HTML(value='', label=f'L={n}'))
 
             creativity_clear.add([creativity_query, creativity_latency, creativity_ci] + creativity_htmls)
             creativity_submit.click(creativity, inputs=[index_desc, creativity_query], outputs=[creativity_latency, creativity_ci] + creativity_htmls, api_name=False)

constants.py

@@ -15,7 +15,7 @@ MAX_QUERY_CHARS = int(os.environ.get('MAX_QUERY_CHARS', 1000))
 NGRAM_LEN_MIN = int(os.environ.get('NGRAM_LEN_MIN', 5))
 NGRAM_LEN_MAX = int(os.environ.get('NGRAM_LEN_MAX', 11))
 MAX_DISP_CHARS_PER_LINE = int(os.environ.get('MAX_DISP_CHARS_PER_LINE', 90))
-MAX_TIMEOUT_IN_SECONDS = int(os.environ.get('MAX_TIMEOUT_IN_SECONDS', 60))
+MAX_TIMEOUT_IN_SECONDS = int(os.environ.get('MAX_TIMEOUT_IN_SECONDS', 30))
 
 # HF demo
 API_URL = os.environ.get('API_URL', None)