Spaces:
Sleeping
Sleeping
gera commited on
Commit ·
e46e576
1
Parent(s): 3a09f3c
added questions to first prompt to save money.
Browse files
app.py
CHANGED
|
@@ -13,12 +13,18 @@ LIMIT = 125000 # some space for answer
|
|
| 13 |
api_key = os_getenv("OPENAI_APIKEY")
|
| 14 |
client = OpenAI(api_key=api_key)
|
| 15 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
def get_prompt(books, question = None):
|
| 17 |
prompt = (
|
| 18 |
f"Read the following books.\n" +
|
| 19 |
f"Each book may have some pages at the beggining with data about the book, an index, or table of content, etc. " +
|
| 20 |
f"Pages may have a header and/or a footer. Consider all this maybe present." +
|
| 21 |
-
f"
|
|
|
|
| 22 |
f"**Title**: ...\n"
|
| 23 |
f"**Author**: ...\n"
|
| 24 |
f"**Chapter Names**: ...\n"
|
|
@@ -29,24 +35,37 @@ def get_prompt(books, question = None):
|
|
| 29 |
|
| 30 |
return prompt
|
| 31 |
|
| 32 |
-
def chat(message, history, files):
|
| 33 |
history_openai_format = []
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
|
| 38 |
-
if
|
| 39 |
-
|
| 40 |
-
raise gr.Error("First message must be empty")
|
| 41 |
-
message = history[0][0]
|
| 42 |
else:
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
|
|
|
|
|
|
| 48 |
|
| 49 |
-
|
|
|
|
| 50 |
|
| 51 |
response = client.chat.completions.create(
|
| 52 |
model=MODEL,
|
|
@@ -70,7 +89,7 @@ def get_text(filename):
|
|
| 70 |
answer = open(filename).read()
|
| 71 |
return answer
|
| 72 |
|
| 73 |
-
def files_ready(filenames):
|
| 74 |
encoder = encoding = tiktoken.encoding_for_model('gpt-4-turbo')
|
| 75 |
books = ''
|
| 76 |
for i, name in enumerate(filenames):
|
|
@@ -80,18 +99,20 @@ def files_ready(filenames):
|
|
| 80 |
prompt = get_prompt(books)
|
| 81 |
tokens = len(encoder.encode(prompt))
|
| 82 |
cost = tokens * PRICE_PER_M / 1000000 * 2 # * 2 is too much for an answer
|
|
|
|
| 83 |
|
| 84 |
if tokens > LIMIT:
|
| 85 |
raise gr.Error(f"Book is too long. It's {tokens} tokens long and can't be more than {LIMIT}.")
|
| 86 |
-
return len(prompt), tokens, f"${cost}",
|
| 87 |
|
| 88 |
-
def files_changed(filenames):
|
| 89 |
if filenames:
|
| 90 |
-
return "-", "-"
|
| 91 |
else:
|
| 92 |
-
return 0, 0, "$0"
|
| 93 |
|
| 94 |
with gr.Blocks(title="Book summarization and more") as demo:
|
|
|
|
| 95 |
with gr.Row():
|
| 96 |
files = gr.Files(file_types=["txt","doc","docx","pdf"] )
|
| 97 |
with gr.Column():
|
|
@@ -102,12 +123,12 @@ with gr.Blocks(title="Book summarization and more") as demo:
|
|
| 102 |
chat = gr.ChatInterface(
|
| 103 |
fn=chat,
|
| 104 |
title="Summarization and more",
|
| 105 |
-
additional_inputs=[files],
|
| 106 |
multimodal=False)
|
| 107 |
|
| 108 |
other = gr.Button(interactive=False)
|
| 109 |
-
files.upload(files_ready, [files], [letters, tokens, cost,
|
| 110 |
-
files.change(files_changed, files, [letters, tokens, cost])
|
| 111 |
|
| 112 |
|
| 113 |
auth=os_getenv("APP_USERS", "null")
|
|
|
|
| 13 |
api_key = os_getenv("OPENAI_APIKEY")
|
| 14 |
client = OpenAI(api_key=api_key)
|
| 15 |
|
| 16 |
+
def new_state():
|
| 17 |
+
return gr.State({
|
| 18 |
+
"prompt": "",
|
| 19 |
+
})
|
| 20 |
+
|
| 21 |
def get_prompt(books, question = None):
|
| 22 |
prompt = (
|
| 23 |
f"Read the following books.\n" +
|
| 24 |
f"Each book may have some pages at the beggining with data about the book, an index, or table of content, etc. " +
|
| 25 |
f"Pages may have a header and/or a footer. Consider all this maybe present." +
|
| 26 |
+
f"For each book, please answer, all below in the suggested format and also answer all the questions at the end in detail, if present.\n"
|
| 27 |
+
f"Answer in the language of the book:\n"+
|
| 28 |
f"**Title**: ...\n"
|
| 29 |
f"**Author**: ...\n"
|
| 30 |
f"**Chapter Names**: ...\n"
|
|
|
|
| 35 |
|
| 36 |
return prompt
|
| 37 |
|
| 38 |
+
def chat(message, history, files, state):
|
| 39 |
history_openai_format = []
|
| 40 |
|
| 41 |
+
prompt = state["prompt"]
|
| 42 |
+
|
| 43 |
+
if not message:
|
| 44 |
+
if len(history) > 0:
|
| 45 |
+
gr.Error("You sent an empty question. It's expensive, don't do it")
|
| 46 |
+
return ''
|
| 47 |
+
|
| 48 |
+
if not prompt:
|
| 49 |
+
gr.Error("First upload a book")
|
| 50 |
+
return ''
|
| 51 |
+
|
| 52 |
+
if (not history) and message:
|
| 53 |
+
prompt += f"**Questions**:{message}"
|
| 54 |
+
state["prompt"] = prompt
|
| 55 |
|
| 56 |
+
if history:
|
| 57 |
+
history[0] = (prompt, history[0][1])
|
|
|
|
|
|
|
| 58 |
else:
|
| 59 |
+
message = prompt
|
| 60 |
+
|
| 61 |
+
for human, assistant in history:
|
| 62 |
+
if human:
|
| 63 |
+
history_openai_format.append({"role": "user", "content": human })
|
| 64 |
+
if assistant:
|
| 65 |
+
history_openai_format.append({"role": "assistant", "content":assistant})
|
| 66 |
|
| 67 |
+
if message:
|
| 68 |
+
history_openai_format.append({"role": "user", "content": message})
|
| 69 |
|
| 70 |
response = client.chat.completions.create(
|
| 71 |
model=MODEL,
|
|
|
|
| 89 |
answer = open(filename).read()
|
| 90 |
return answer
|
| 91 |
|
| 92 |
+
def files_ready(filenames, state):
|
| 93 |
encoder = encoding = tiktoken.encoding_for_model('gpt-4-turbo')
|
| 94 |
books = ''
|
| 95 |
for i, name in enumerate(filenames):
|
|
|
|
| 99 |
prompt = get_prompt(books)
|
| 100 |
tokens = len(encoder.encode(prompt))
|
| 101 |
cost = tokens * PRICE_PER_M / 1000000 * 2 # * 2 is too much for an answer
|
| 102 |
+
state["prompt"] = prompt
|
| 103 |
|
| 104 |
if tokens > LIMIT:
|
| 105 |
raise gr.Error(f"Book is too long. It's {tokens} tokens long and can't be more than {LIMIT}.")
|
| 106 |
+
return len(prompt), tokens, f"${cost}", state
|
| 107 |
|
| 108 |
+
def files_changed(filenames, state):
|
| 109 |
if filenames:
|
| 110 |
+
return "-", "-", "-", state
|
| 111 |
else:
|
| 112 |
+
return 0, 0, "$0", new_state()
|
| 113 |
|
| 114 |
with gr.Blocks(title="Book summarization and more") as demo:
|
| 115 |
+
state = new_state()
|
| 116 |
with gr.Row():
|
| 117 |
files = gr.Files(file_types=["txt","doc","docx","pdf"] )
|
| 118 |
with gr.Column():
|
|
|
|
| 123 |
chat = gr.ChatInterface(
|
| 124 |
fn=chat,
|
| 125 |
title="Summarization and more",
|
| 126 |
+
additional_inputs=[files, state],
|
| 127 |
multimodal=False)
|
| 128 |
|
| 129 |
other = gr.Button(interactive=False)
|
| 130 |
+
files.upload(files_ready, [files, state], [letters, tokens, cost, state])
|
| 131 |
+
files.change(files_changed, [files, state], [letters, tokens, cost, state])
|
| 132 |
|
| 133 |
|
| 134 |
auth=os_getenv("APP_USERS", "null")
|