Update app.py

app.py

@@ -5,7 +5,13 @@ import json
 import os
 
 title = "BLOOM"
-description = "Gradio Demo for BLOOM. To use it, simply add your text, or click one of the examples to load them. Read more at the links below."
+description = """Gradio Demo for BLOOM. To use it, simply add your text, or click one of the examples to load them.
+Tips: Do NOT talk to BLOOM as an entity, it's not a chatbot but a webpage/blog/article completion model. For the best results: MIMIC a few words of a webpage similar to the content you want to generate. Start a sentence as if YOU were writing a blog, webpage, math post, coding article and BLOOPM will generate a coherent follow-up.
+
+Generation simple selector:
+- sampling: imaginative completions (may be not super accurate e.g. math/history)
+- greedy: accurate completions (may be more boring or have repetitions)
+"""
 
 API_URL = "https://hfbloom.ngrok.io/generate"
 HF_API_TOKEN = os.getenv("HF_API_TOKEN")
@@ -14,39 +20,40 @@ hf_writer = gr.HuggingFaceDatasetSaver(HF_API_TOKEN, "huggingface/bloom_internal
 
 
 examples = [
-    ['A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is: We were traveling in Africa and we saw these very cute whatpus. To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:', 8, 0, 0, 0.1, 0, 0.9, False],
-    ['def quicksort(l):', 8, 0, 0, 0.1, 0, 0.9, False],
-    ['Q: ¿Cómo te llamas? A: What is your name? Q: ¿Qué edad tienes? A: How old are you? Q: ¿Dónde vives?', 8, 0, 0, 0.1, 0, 0.9, False]
+    ['A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is: We were traveling in Africa and we saw these very cute whatpus. To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:', 32, "Sample"],
+    ["Pour déguster un ortolan, il faut tout d'abord", 32, "Sample"],
+    ["Question: If I put cheese into the fridge, will it melt?\nAnswer:", 32, "Sample"],
+    ["Math exercise - answers:\n34+10=44\n54+20=", 16, "Greedy"],
+    ["Question: Where does the Greek Goddess Persephone spend half of the year when she is not with her mother?\nAnswer:", 24, "Greedy"],
+    ["spelling test answers.\nWhat are the letters in « language »?\nAnswer: l-a-n-g-u-a-g-e\nWhat are the letters in « Romanian »?\nAnswer:", 24, "Greedy"],
 ]
 
-def safe_text(text):
-    text = text.replace('%', '\\%25')
-    text = text.replace('#', '\\%23')
-    text = text.replace('+', '\\%2B')
-    text = text.replace('*', '\\%2A')
-    text = text.replace('&', '\\%26')
-    text = re.sub(r"([$_*\[\]()~`>\#\+\-=|\.!{}])", r"\\\1", text)
-    return f"<pre>{text}</pre>"
-
-
 def query(payload):
     print(payload)
     response = requests.request("POST", API_URL, json=payload)
     print(response)
     return json.loads(response.content.decode("utf-8"))
     
-def inference(input_sentence, max_length, no_repeat_ngram_size, num_beams, temperature,top_k, top_p, greedy_decoding, seed=42):
-    top_k = None if top_k == 0 else top_k
-    do_sample = False if num_beams > 0 else not greedy_decoding
-    num_beams = None if (greedy_decoding or num_beams == 0) else num_beams
-    no_repeat_ngram_size = None if num_beams is None else no_repeat_ngram_size
-    top_p = None if num_beams else top_p
-
-    early_stopping = None if num_beams is None else num_beams > 0
+def inference(input_sentence, max_length, sample_or_greedy, seed=42):
+    if sample_or_greedy == "Sample":
+        parameters = {"max_new_tokens": max_length,
+                      "top_p": 0.9,
+                      "do_sample": True,
+                      "seed": seed,
+                      "early_stopping": False,
+                      "length_penalty": 0.0,
+                      "eos_token_id": None}
+    else:
+        parameters = {"max_new_tokens": max_length,
+                      "do_sample": False,
+                      "seed": seed,
+                      "early_stopping": False,
+                      "length_penalty": 0.0,
+                      "eos_token_id": None}
 
     payload = {"inputs": input_sentence,
-               "parameters": {"max_new_tokens": max_length, "top_k": top_k, "top_p": top_p, "temperature": temperature,
-                              "do_sample": do_sample, "seed": seed, "early_stopping":early_stopping, "no_repeat_ngram_size":no_repeat_ngram_size, "num_beams":num_beams}}
+               "parameters": parameters}
+
     data = query(
         payload
     )
@@ -58,13 +65,8 @@ gr.Interface(
     inference, 
     [
         gr.inputs.Textbox(label="Input"),
-        gr.inputs.Slider(1, 64, default=8, step=1, label="Tokens to generate"),
-        gr.inputs.Slider(1, 10, default=2, step=1, label="No repeat N gram"),
-        gr.inputs.Slider(0, 10, default=5, step=1, label="Num beams"),
-        gr.inputs.Slider(0.0, 1.0, default=0.1, step=0.05, label="Temperature"),
-        gr.inputs.Slider(0, 64, default=0, step=1, label="Top K"),
-        gr.inputs.Slider(0.0, 10, default=0.9, step=0.05, label="Top P"),
-        gr.inputs.Checkbox(False, label="Greedy decoding"),
+        gr.inputs.Slider(1, 64, default=32, step=1, label="Tokens to generate"),
+        gr.inputs.Radio(["Sample", "Greedy"]),
     ], 
     gr.outputs.Textbox(label="Output"),
     examples=examples,