chat-gpt-ui / app.py
coraKong's picture
Add gpt-3.5-turbo-16k
368c3ea
raw
history blame
26 kB
# import gradio as gr
import gradio
# import lmdb
# import base64
# import io
# import random
# import time
import json
import copy
# import sqlite3
from urllib.parse import urljoin
import openai
DEFAULT_PROMPT = [
["system", "You(assistant) are a helpful AI assistant."],
]
# def get_settings(old_state):
# db_path = './my_app_state'
# env = lmdb.open(db_path, max_dbs=2*1024*1024)
# # print(env.stat())
# txn = env.begin()
# saved_api_key = txn.get(key=b'api_key').decode('utf-8') or ''
# txn.commit()
# env.close()
# new_state = copy.deepcopy(old_state) or {}
# new_state['api_key'] = saved_api_key
# return new_state, saved_api_key
# def save_settings(old_state, api_key_text):
# db_path = './my_app_state'
# env = lmdb.open(db_path, max_dbs=2*1024*1024)
# # print(env.stat())
# txn = env.begin(write=True)
# txn.put(key=b'api_key', value=api_key_text.encode('utf-8'))
# # 提交事务
# txn.commit()
# return get_settings(old_state)
def on_click_send_btn(
global_state_json, api_key_text, chat_input_role, chat_input, prompt_table, chat_use_prompt, chat_use_history, chat_log,
chat_model, temperature, top_p, choices_num, stream, max_tokens, presence_penalty, frequency_penalty, logit_bias,
):
old_state = json.loads(global_state_json or "{}")
print('\n\n\n\n\n')
print(prompt_table)
prompt_table = prompt_table or []
chat_log = chat_log or []
chat_log_md = ''
if chat_use_prompt:
chat_log_md += '<center>(prompt)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", prompt_table)])
chat_log_md += '\n---\n'
if True:
chat_log_md += '<center>(history)</center>\n\n' if chat_use_history else '<center>(not used history)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", chat_log)])
chat_log_md += '\n---\n'
# if chat_input=='':
# return json.dumps(old_state), chat_log, chat_log_md, chat_log_md, None, None, chat_input
print('\n')
print(chat_input)
print('')
try:
logit_bias_json = json.dumps(logit_bias) if logit_bias else None
except:
return json.dumps(old_state), chat_log, chat_log_md, chat_log_md, None, None, chat_input
new_state = copy.deepcopy(old_state) or {}
req_hist = copy.deepcopy(prompt_table) if chat_use_prompt else []
if chat_use_history:
for hh in (chat_log or []):
req_hist.append(hh)
if chat_input and chat_input!="":
req_hist.append([(chat_input_role or 'user'), chat_input])
openai.api_key = api_key_text
props = {
'model': chat_model,
'messages': [xx for xx in map(lambda it: {'role':it[0], 'content':it[1]}, req_hist)],
'temperature': temperature,
'top_p': top_p,
'n': choices_num,
'stream': stream,
'presence_penalty': presence_penalty,
'frequency_penalty': frequency_penalty,
}
if max_tokens>0:
props['max_tokens'] = max_tokens
if logit_bias_json is not None:
props['logit_bias'] = logit_bias_json
props_json = json.dumps(props)
try:
completion = openai.ChatCompletion.create(**props)
print('')
chat_log_md = ''
if chat_use_prompt:
chat_log_md += '<center>(prompt)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", prompt_table)])
chat_log_md += '\n---\n'
if True:
chat_log_md += '<center>(history)</center>\n\n' if chat_use_history else '<center>(not used history)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", chat_log)])
chat_log_md += '\n---\n'
if chat_input and chat_input!="":
chat_log.append([(chat_input_role or 'user'), chat_input])
chat_log_md += f"##### `{(chat_input_role or 'user')}`\n\n{chat_input}\n\n"
partial_words = ""
counter=0
if stream:
the_response = ''
the_response_role = ''
for chunk in completion:
#Skipping first chunk
if counter == 0:
the_response_role = chunk.choices[0].delta.role
chat_log_md += f"##### `{the_response_role}`\n\n"
counter += 1
continue
# print(('chunk', chunk))
if chunk.choices[0].finish_reason is None:
the_response_chunk = chunk.choices[0].delta.content
the_response += the_response_chunk
chat_log_md += f"{the_response_chunk}"
yield json.dumps(new_state), chat_log, chat_log_md, chat_log_md, "{}", props_json, ''
else:
chat_log.append([the_response_role, the_response])
chat_log_md += f"\n\n"
yield json.dumps(new_state), chat_log, chat_log_md, chat_log_md, '{"msg": "stream模式不支持显示"}', props_json, ''
# chat_last_resp = json.dumps(completion.__dict__)
# chat_last_resp_dict = json.loads(chat_last_resp)
# chat_last_resp_dict['api_key'] = "hidden by UI"
# chat_last_resp_dict['organization'] = "hidden by UI"
# chat_last_resp = json.dumps(chat_last_resp_dict)
else:
the_response_role = completion.choices[0].message.role
the_response = completion.choices[0].message.content
print(the_response)
print('')
chat_log.append([the_response_role, the_response])
chat_log_md += f"##### `{the_response_role}`\n\n{the_response}\n\n"
chat_last_resp = json.dumps(completion.__dict__)
chat_last_resp_dict = json.loads(chat_last_resp)
chat_last_resp_dict['api_key'] = "hidden by UI"
chat_last_resp_dict['organization'] = "hidden by UI"
chat_last_resp = json.dumps(chat_last_resp_dict)
yield json.dumps(new_state), chat_log, chat_log_md, chat_log_md, chat_last_resp, props_json, ''
except Exception as error:
print(error)
print('error!!!!!!')
chat_log_md = ''
if chat_use_prompt:
chat_log_md += '<center>(prompt)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", prompt_table)])
chat_log_md += '\n---\n'
if True:
chat_log_md += '<center>(history)</center>\n\n' if chat_use_history else '<center>(not used history)</center>\n\n'
chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", chat_log)])
chat_log_md += '\n---\n'
# chat_log_md = ''
# chat_log_md = "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", prompt_table)]) if chat_use_prompt else ''
# chat_log_md += "\n".join([xx for xx in map(lambda it: f"##### `{it[0]}`\n\n{it[1]}\n\n", hist)])
chat_log_md += "\n"
chat_log_md += str(error)
yield json.dumps(new_state), chat_log, chat_log_md, chat_log_md, None, props_json, chat_input
def clear_history():
return [], ""
def copy_history(txt):
# print('\n\n copying')
# print(txt)
# print('\n\n')
pass
def update_saved_prompt_titles(global_state_json, selected_saved_prompt_title):
print('')
global_state = json.loads(global_state_json or "{}")
print(global_state)
print(selected_saved_prompt_title)
saved_prompts = global_state.get('saved_prompts') or []
print(saved_prompts)
the_choices = [(it.get('title') or '[untitled]') for it in saved_prompts]
print(the_choices)
print('')
return gradio.Dropdown.update(choices=the_choices)
def save_prompt(global_state_json, saved_prompts, prompt_title, prompt_table):
the_choices = []
global_state = json.loads(global_state_json or "{}")
saved_prompts = global_state.get('saved_prompts') or []
if len(saved_prompts):
the_choices = [it.get('title') or '[untitled]' for it in saved_prompts]
pass
return global_state_json, gradio.Dropdown.update(choices=the_choices, value=prompt_title), prompt_title, prompt_table
def load_saved_prompt(title):
pass
header_intro = """
Try our new ChatGPT Batch Tool: [Here](https://huggingface.co/spaces/hugforziio/chat-gpt-batch)
"""
css = """
.table-wrap .cell-wrap input {min-width:80%}
#api-key-textbox textarea {filter:blur(8px); transition: filter 0.25s}
#api-key-textbox textarea:focus {filter:none}
#chat-log-md hr {margin-top: 1rem; margin-bottom: 1rem;}
"""
with gradio.Blocks(title="ChatGPT", css=css) as demo:
global_state_json = gradio.Textbox(visible=False)
# https://gradio.app/docs
# https://platform.openai.com/docs/api-reference/chat/create
with gradio.Tab("ChatGPT"):
gradio.Markdown(header_intro)
with gradio.Row():
with gradio.Box():
with gradio.Column(scale=12):
with gradio.Row():
api_key_text = gradio.Textbox(label="Your API key", elem_id="api-key-textbox")
with gradio.Row():
with gradio.Column(scale=2):
api_key_refresh_btn = gradio.Button("🔄 Load from browser storage")
api_key_refresh_btn.click(
# get_settings,
None,
inputs=[],
outputs=[api_key_text],
api_name="load-settings",
_js="""()=>{
const the_api_key = localStorage?.getItem?.('[gradio][chat-gpt-ui][api_key_text]') ?? '';
return the_api_key;
}""",
)
with gradio.Column(scale=2):
api_key_save_btn = gradio.Button("💾 Save to browser storage")
api_key_save_btn.click(
# save_settings,
None,
inputs=[api_key_text],
outputs=[api_key_text],
api_name="save-settings",
_js="""(api_key_text)=>{
localStorage.setItem('[gradio][chat-gpt-ui][api_key_text]', api_key_text);
return api_key_text;
}""",
)
with gradio.Row():
gradio.Markdown("Go to https://platform.openai.com/account/api-keys to get your API key.")
with gradio.Row():
with gradio.Box():
gradio.Markdown("**Prompt**")
with gradio.Column(scale=12):
with gradio.Row():
with gradio.Column(scale=6):
prompt_title = gradio.Textbox(label='Prompt title (only for saving)')
with gradio.Column(scale=6):
selected_saved_prompt_title = gradio.Dropdown(label='Select prompt from saved list (click ♻️ then 🔄)')
with gradio.Row():
with gradio.Column(scale=1, min_width=100):
saved_prompts_refresh_btn = gradio.Button("♻️")
with gradio.Column(scale=1, min_width=100):
saved_prompts_save_btn = gradio.Button("💾")
with gradio.Column(scale=1, min_width=100):
saved_prompts_delete_btn = gradio.Button("🗑")
with gradio.Column(scale=1, min_width=100):
saved_prompts_list_refresh_btn = gradio.Button("🔄")
with gradio.Column(scale=1, min_width=100):
copy_prompt = gradio.Button("📑")
with gradio.Column(scale=1, min_width=100):
paste_prompt = gradio.Button("📋")
with gradio.Row():
gradio.Markdown("""Buttons above: ♻️ then 🔄: Load prompts from browser storage. 💾 then 🔄: Save current prompt to browser storage, overwrite the prompt with the same title. 🗑 then 🔄: Delete prompt with the same title from browser storage. 🔄 : Update the selector list. 📑 : Copy current prompt to clipboard. 📋 : Paste prompt from clipboard (need [permission](https://developer.mozilla.org/en-US/docs/Web/API/Clipboard/readText#browser_compatibility)).""")
with gradio.Row():
prompt_table = gradio.Dataframe(
type='array',
label='Prompt content', col_count=(2, 'fixed'), max_cols=2,
value=DEFAULT_PROMPT, headers=['role', 'content'], interactive=True,
)
with gradio.Row():
gradio.Markdown("The Table above is editable. The content will be added to the beginning of the conversation (if you check 'send with prompt' as `√`). See https://platform.openai.com/docs/guides/chat/introduction .")
copy_prompt.click(None, inputs=[prompt_title, prompt_table], outputs=[prompt_title, prompt_table], _js="""(prompt_title, prompt_table)=>{
try {
const txt = JSON.stringify({
title: prompt_title,
content: prompt_table,
}, null, 2);
console.log(txt);
const promise = navigator?.clipboard?.writeText?.(txt);
} catch(error) {console?.log?.(error);};
return [prompt_title, prompt_table];
}""")
paste_prompt.click(None, inputs=[prompt_title, prompt_table], outputs=[prompt_title, prompt_table], _js="""async (prompt_title, prompt_table)=>{
console.log("flag1");
try {
const promise = navigator?.clipboard?.readText?.();
console.log(promise);
console.log("flag1 p");
const result = await promise?.then?.((txt)=>{
console.log("flag1 t");
const json = JSON.parse(txt);
const title = json?.title ?? "";
console.log("flag1 0");
console.log(title);
const content = json?.content ?? {data: [], headers: ['role', 'content']};
console.log(content);
const result = [title, content];
console.log("flag1 1");
console.log(result);
console.log("flag1 2");
return result;
});
console.log("flag1 3");
if (result!=null) {
return result;
};
} catch(error) {console?.log?.(error);};
console.log("flag2");
try {
const promise = navigator?.clipboard?.read?.();
console.log(promise);
promise?.then?.((data)=>{
console.log(data);
});
} catch(error) {console?.log?.(error);};
console.log("flag3");
return [prompt_title, prompt_table];
}""")
saved_prompts_refresh_btn.click(None, inputs=[global_state_json, selected_saved_prompt_title], outputs=[global_state_json, selected_saved_prompt_title], _js="""(global_state_json, saved_prompts)=>{
try {
if(global_state_json=="") {global_state_json=null;};
console.log('global_state_json:\\n', global_state_json);
const global_state = JSON.parse(global_state_json??"{ }")??{ };
const saved = (JSON.parse(localStorage?.getItem?.('[gradio][chat-gpt-ui][prompts]') ?? '[]'));
console.log('saved:\\n', saved);
global_state['saved_prompts'] = saved;
global_state['selected_saved_prompt_title'] = saved.map(it=>it?.title??"[untitled]")[0];
const results = [JSON.stringify(global_state), global_state['selected_saved_prompt_title']];
console.log(results);
return results;
} catch(error) {
console.log(error);
return ["{ }", ""];
};
}""")
saved_prompts_list_refresh_btn.click(
update_saved_prompt_titles, inputs=[global_state_json, selected_saved_prompt_title], outputs=[selected_saved_prompt_title],
)
selected_saved_prompt_title.change(None, inputs=[global_state_json, selected_saved_prompt_title], outputs=[global_state_json, prompt_title, prompt_table], _js="""(global_state_json, selected_saved_prompt_title)=>{
if(global_state_json=="") {global_state_json=null;};
const global_state = JSON.parse(global_state_json??"{ }")??{ };
const found = (global_state?.['saved_prompts']??[]).find(it=>it?.title==selected_saved_prompt_title);
return [JSON.stringify(global_state), found?.title??'', found?.content??{data:[], headers:["role", "content"]}];
}""")
saved_prompts_delete_btn.click(None, inputs=[global_state_json, selected_saved_prompt_title, prompt_title, prompt_table], outputs=[global_state_json, selected_saved_prompt_title, prompt_title, prompt_table], _js="""(global_state_json, saved_prompts, prompt_title, prompt_table)=>{
if(prompt_title==""||!prompt_title){
return [global_state_json, selected_saved_prompt_title, prompt_title, prompt_table];
};
console.log('global_state_json:\\n', global_state_json);
if(global_state_json=="") {global_state_json=null;};
const global_state = JSON.parse(global_state_json??"{ }")??{ };
console.log(global_state);
const saved = (JSON.parse(localStorage?.getItem?.('[gradio][chat-gpt-ui][prompts]') ?? '[]'));
console.log('saved:\\n', saved);
global_state['saved_prompts'] = saved?.filter?.(it=>it.title!=prompt_title)??[];
global_state['selected_saved_prompt_title'] = "";
console.log(global_state);
localStorage?.setItem?.('[gradio][chat-gpt-ui][prompts]', JSON.stringify(global_state['saved_prompts']));
return [JSON.stringify(global_state), "", "", {data: [], headers: ['role', 'content']}];
}""")
saved_prompts_save_btn.click(None, inputs=[global_state_json, selected_saved_prompt_title, prompt_title, prompt_table], outputs=[global_state_json, selected_saved_prompt_title, prompt_title, prompt_table], _js="""(global_state_json, saved_prompts, prompt_title, prompt_table)=>{
if(prompt_title==""||!prompt_title){
return [global_state_json, selected_saved_prompt_title, prompt_title, prompt_table];
};
console.log('global_state_json:\\n', global_state_json);
if(global_state_json=="") {global_state_json=null;};
const global_state = JSON.parse(global_state_json??"{ }")??{ };
console.log(global_state);
const saved = (JSON.parse(localStorage?.getItem?.('[gradio][chat-gpt-ui][prompts]') ?? '[]'));
console.log('saved:\\n', saved);
const new_prompt_obj = {
title: prompt_title, content: prompt_table,
};
global_state['saved_prompts'] = saved?.filter?.(it=>it.title!=prompt_title)??[];
global_state['saved_prompts'].unshift(new_prompt_obj);
global_state['selected_saved_prompt_title'] = prompt_title;
console.log(global_state);
localStorage?.setItem?.('[gradio][chat-gpt-ui][prompts]', JSON.stringify(global_state['saved_prompts']));
return [JSON.stringify(global_state), prompt_title, prompt_title, prompt_table];
}""")
with gradio.Row():
with gradio.Column(scale=4):
with gradio.Box():
gradio.Markdown("See https://platform.openai.com/docs/api-reference/chat/create .")
chat_model = gradio.Dropdown(label="model", choices=[
"gpt-3.5-turbo", "gpt-3.5-turbo-0301", "gpt-3.5-turbo-16k",
"gpt-4", "gpt-4-0314", "gpt-4-32k", "gpt-4-32k-0314",
], value="gpt-3.5-turbo")
chat_temperature = gradio.Slider(label="temperature", value=1, minimum=0, maximum=2)
chat_top_p = gradio.Slider(label="top_p", value=1, minimum=0, maximum=1)
chat_choices_num = gradio.Slider(label="choices num(n)", value=1, minimum=1, maximum=20)
chat_stream = gradio.Checkbox(label="stream", value=True, visible=True)
chat_max_tokens = gradio.Slider(label="max_tokens", value=-1, minimum=-1, maximum=4096)
chat_presence_penalty = gradio.Slider(label="presence_penalty", value=0, minimum=-2, maximum=2)
chat_frequency_penalty = gradio.Slider(label="frequency_penalty", value=0, minimum=-2, maximum=2)
chat_logit_bias = gradio.Textbox(label="logit_bias", visible=False)
pass
with gradio.Column(scale=8):
with gradio.Row():
with gradio.Column(scale=10):
chat_log = gradio.State()
with gradio.Box():
with gradio.Column(scale=10):
chat_log_box = gradio.Markdown(label='chat history', value="<center>(empty)</center>", elem_id="chat-log-md")
real_md_box = gradio.Textbox(value="", visible=False)
with gradio.Row():
chat_copy_history_btn = gradio.Button("Copy all (as HTML)")
chat_copy_history_md_btn = gradio.Button("Copy all (as Markdown)")
chat_copy_history_btn.click(
copy_history, inputs=[chat_log_box],
_js="""(txt)=>{
console.log(txt);
try {let promise = navigator?.clipboard?.writeText?.(txt);}
catch(error) {console?.log?.(error);};
}""",
)
chat_copy_history_md_btn.click(
copy_history, inputs=[real_md_box],
_js="""(txt)=>{
console.log(txt);
try {let promise = navigator?.clipboard?.writeText?.(txt);}
catch(error) {console?.log?.(error);};
}""",
)
chat_input_role = gradio.Dropdown(label='role', choices=['user', 'system', 'assistant'], value='user')
chat_input = gradio.Textbox(lines=4, label='input')
with gradio.Row():
chat_clear_history_btn = gradio.Button("clear history")
chat_clear_history_btn.click(clear_history, inputs=[], outputs=[chat_log, chat_log_box])
chat_use_prompt = gradio.Checkbox(label='send with prompt', value=True)
chat_use_history = gradio.Checkbox(label='send with history', value=True)
chat_send_btn = gradio.Button("send")
pass
with gradio.Row():
chat_last_req = gradio.JSON(label='last request')
chat_last_resp = gradio.JSON(label='last response')
chat_send_btn.click(
on_click_send_btn,
inputs=[
global_state_json, api_key_text, chat_input_role, chat_input, prompt_table, chat_use_prompt, chat_use_history, chat_log,
chat_model, chat_temperature, chat_top_p, chat_choices_num, chat_stream, chat_max_tokens, chat_presence_penalty, chat_frequency_penalty, chat_logit_bias,
],
outputs=[global_state_json, chat_log, chat_log_box, real_md_box, chat_last_resp, chat_last_req, chat_input],
api_name="click-send-btn",
)
pass
with gradio.Tab("Settings"):
gradio.Markdown('Currently nothing.')
pass
if __name__ == "__main__":
demo.queue(concurrency_count=20).launch()