Spaces:
Configuration error
Configuration error
Refactor
Browse files
app.py
CHANGED
@@ -1,10 +1,13 @@
|
|
1 |
# Basic example for doing model-in-the-loop dynamic adversarial data collection
|
2 |
# using Gradio Blocks.
|
|
|
3 |
import json
|
4 |
import os
|
5 |
import threading
|
|
|
6 |
import uuid
|
7 |
from pathlib import Path
|
|
|
8 |
from urllib.parse import parse_qs
|
9 |
|
10 |
import gradio as gr
|
@@ -17,15 +20,27 @@ from langchain.prompts import load_prompt
|
|
17 |
|
18 |
from utils import force_git_push
|
19 |
|
20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
21 |
if Path(".env").is_file():
|
22 |
load_dotenv(".env")
|
23 |
DATASET_REPO_URL = os.getenv("DATASET_REPO_URL")
|
24 |
FORCE_PUSH = os.getenv("FORCE_PUSH")
|
25 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
26 |
PROMPT_TEMPLATES = Path("prompt_templates")
|
27 |
-
# Set env variable for langchain to communicate with Hugging Face Hub
|
28 |
-
os.environ["HUGGINGFACEHUB_API_TOKEN"] = HF_TOKEN
|
29 |
|
30 |
DATA_FILENAME = "data.jsonl"
|
31 |
DATA_FILE = os.path.join("data", DATA_FILENAME)
|
@@ -58,52 +73,24 @@ asynchronous_push(f_stop)
|
|
58 |
# Now let's run the app!
|
59 |
prompt = load_prompt(PROMPT_TEMPLATES / "openai_chatgpt.json")
|
60 |
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
model_kwargs={"temperature": 1}
|
65 |
-
),
|
66 |
-
prompt=prompt,
|
67 |
-
verbose=False,
|
68 |
-
memory=ConversationBufferMemory(ai_prefix="Assistant"),
|
69 |
-
)
|
70 |
|
71 |
-
|
|
|
72 |
llm=HuggingFaceHub(
|
73 |
-
repo_id=
|
74 |
-
model_kwargs={"temperature":
|
|
|
75 |
),
|
76 |
prompt=prompt,
|
77 |
verbose=False,
|
78 |
memory=ConversationBufferMemory(ai_prefix="Assistant"),
|
79 |
-
)
|
80 |
|
81 |
-
chatbot_3 = ConversationChain(
|
82 |
-
llm=HuggingFaceHub(
|
83 |
-
repo_id="bigscience/T0_3B",
|
84 |
-
model_kwargs={"temperature": 1}
|
85 |
-
),
|
86 |
-
prompt=prompt,
|
87 |
-
verbose=False,
|
88 |
-
memory=ConversationBufferMemory(ai_prefix="Assistant"),
|
89 |
-
)
|
90 |
|
91 |
-
|
92 |
-
llm=HuggingFaceHub(
|
93 |
-
repo_id="EleutherAI/gpt-j-6B",
|
94 |
-
model_kwargs={"temperature": 1}
|
95 |
-
),
|
96 |
-
prompt=prompt,
|
97 |
-
verbose=False,
|
98 |
-
memory=ConversationBufferMemory(ai_prefix="Assistant"),
|
99 |
-
)
|
100 |
-
|
101 |
-
model_id2model = {
|
102 |
-
"google/flan-t5-xl": chatbot_1,
|
103 |
-
"bigscience/bloom": chatbot_2,
|
104 |
-
"bigscience/T0_3B": chatbot_3,
|
105 |
-
"EleutherAI/gpt-j-6B": chatbot_4
|
106 |
-
}
|
107 |
|
108 |
demo = gr.Blocks()
|
109 |
|
@@ -117,11 +104,9 @@ with demo:
|
|
117 |
"cnt": 0, "data": [],
|
118 |
"past_user_inputs": [],
|
119 |
"generated_responses": [],
|
120 |
-
"response_1": "",
|
121 |
-
"response_2": "",
|
122 |
-
"response_3": "",
|
123 |
-
"response_4": "",
|
124 |
}
|
|
|
|
|
125 |
state = gr.JSON(state_dict, visible=False)
|
126 |
|
127 |
gr.Markdown("# RLHF Interface")
|
@@ -132,26 +117,30 @@ with demo:
|
|
132 |
# Generate model prediction
|
133 |
def _predict(txt, state):
|
134 |
# TODO: parallelize this!
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
|
140 |
response2model_id = {}
|
141 |
-
|
142 |
-
|
143 |
-
response2model_id[response_3] = chatbot_3.llm.repo_id
|
144 |
-
response2model_id[response_4] = chatbot_4.llm.repo_id
|
145 |
|
146 |
state["cnt"] += 1
|
147 |
|
148 |
new_state_md = f"Inputs remaining in HIT: {state['cnt']}/{TOTAL_CNT}"
|
149 |
|
150 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
151 |
state["past_user_inputs"].append(txt)
|
152 |
|
153 |
past_conversation_string = "<br />".join(["<br />".join(["π: " + user_input, "π€: " + model_response]) for user_input, model_response in zip(state["past_user_inputs"], state["generated_responses"] + [""])])
|
154 |
-
return gr.update(visible=False), gr.update(visible=True), gr.update(visible=True, choices=
|
155 |
|
156 |
def _select_response(selected_response, state, dummy):
|
157 |
done = state["cnt"] == TOTAL_CNT
|
@@ -169,7 +158,7 @@ with demo:
|
|
169 |
past_conversation_string = "<br />".join(["<br />".join(["π: " + user_input, "π€: " + model_response]) for user_input, model_response in zip(state["past_user_inputs"], state["generated_responses"])])
|
170 |
query = parse_qs(dummy[1:])
|
171 |
if "assignmentId" in query and query["assignmentId"][0] != "ASSIGNMENT_ID_NOT_AVAILABLE":
|
172 |
-
# It seems that someone is using this app on
|
173 |
# store the assignmentId in the state before submit_hit_button
|
174 |
# is clicked. We can do this here in _predict. We need to save the
|
175 |
# assignmentId so that the turker can get credit for their HIT.
|
@@ -182,17 +171,13 @@ with demo:
|
|
182 |
|
183 |
if done:
|
184 |
# Wipe the memory completely because we will be starting a new hit soon.
|
185 |
-
|
186 |
-
|
187 |
-
chatbot_3.memory = ConversationBufferMemory(ai_prefix="Assistant")
|
188 |
-
chatbot_4.memory = ConversationBufferMemory(ai_prefix="Assistant")
|
189 |
else:
|
190 |
# Sync all of the model's memories with the conversation path that
|
191 |
# was actually taken.
|
192 |
-
|
193 |
-
|
194 |
-
chatbot_3.memory = model_id2model[state["data"][-1]["response2model_id"][selected_response]].memory
|
195 |
-
chatbot_4.memory = model_id2model[state["data"][-1]["response2model_id"][selected_response]].memory
|
196 |
|
197 |
text_input = gr.update(visible=False) if done else gr.update(visible=True)
|
198 |
return gr.update(visible=False), gr.update(visible=True), text_input, gr.update(visible=False), state, gr.update(value=past_conversation_string), toggle_example_submit, toggle_final_submit, toggle_final_submit_preview,
|
@@ -207,7 +192,7 @@ with demo:
|
|
207 |
with gr.Column(visible=False) as final_submit:
|
208 |
submit_hit_button = gr.Button("Submit HIT")
|
209 |
with gr.Column(visible=False) as final_submit_preview:
|
210 |
-
submit_hit_button_preview = gr.Button("Submit Work (preview mode; no
|
211 |
|
212 |
# Button event handlers
|
213 |
get_window_location_search_js = """
|
@@ -232,7 +217,7 @@ with demo:
|
|
232 |
|
233 |
post_hit_js = """
|
234 |
function(state) {
|
235 |
-
// If there is an assignmentId, then the submitter is on
|
236 |
// and has accepted the HIT. So, we need to submit their HIT.
|
237 |
const form = document.createElement('form');
|
238 |
form.action = 'https://workersandbox.mturk.com/mturk/externalSubmit';
|
|
|
1 |
# Basic example for doing model-in-the-loop dynamic adversarial data collection
|
2 |
# using Gradio Blocks.
|
3 |
+
import concurrent.futures
|
4 |
import json
|
5 |
import os
|
6 |
import threading
|
7 |
+
import time
|
8 |
import uuid
|
9 |
from pathlib import Path
|
10 |
+
from typing import List
|
11 |
from urllib.parse import parse_qs
|
12 |
|
13 |
import gradio as gr
|
|
|
20 |
|
21 |
from utils import force_git_push
|
22 |
|
23 |
+
|
24 |
+
def generate_respone(chatbot: ConversationChain, input: str) -> str:
|
25 |
+
"""Generates a response for a `langchain` chatbot."""
|
26 |
+
return chatbot.predict(input=input)
|
27 |
+
|
28 |
+
def generate_responses(chatbots: List[ConversationChain], inputs: List[str]) -> List[str]:
|
29 |
+
"""Generates parallel responses for a list of `langchain` chatbots."""
|
30 |
+
results = []
|
31 |
+
executor = concurrent.futures.ThreadPoolExecutor(max_workers=100)
|
32 |
+
for result in executor.map(generate_respone, chatbots, inputs):
|
33 |
+
results.append(result)
|
34 |
+
return results
|
35 |
+
|
36 |
+
|
37 |
+
# These variables are for storing the MTurk HITs in a Hugging Face dataset.
|
38 |
if Path(".env").is_file():
|
39 |
load_dotenv(".env")
|
40 |
DATASET_REPO_URL = os.getenv("DATASET_REPO_URL")
|
41 |
FORCE_PUSH = os.getenv("FORCE_PUSH")
|
42 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
43 |
PROMPT_TEMPLATES = Path("prompt_templates")
|
|
|
|
|
44 |
|
45 |
DATA_FILENAME = "data.jsonl"
|
46 |
DATA_FILE = os.path.join("data", DATA_FILENAME)
|
|
|
73 |
# Now let's run the app!
|
74 |
prompt = load_prompt(PROMPT_TEMPLATES / "openai_chatgpt.json")
|
75 |
|
76 |
+
# TODO: update this list with better, instruction-trained models
|
77 |
+
MODEL_IDS = ["google/flan-t5-xl", "bigscience/T0_3B", "EleutherAI/gpt-j-6B"]
|
78 |
+
chatbots = []
|
|
|
|
|
|
|
|
|
|
|
|
|
79 |
|
80 |
+
for model_id in MODEL_IDS:
|
81 |
+
chatbots.append(ConversationChain(
|
82 |
llm=HuggingFaceHub(
|
83 |
+
repo_id=model_id,
|
84 |
+
model_kwargs={"temperature": 1},
|
85 |
+
huggingfacehub_api_token=HF_TOKEN,
|
86 |
),
|
87 |
prompt=prompt,
|
88 |
verbose=False,
|
89 |
memory=ConversationBufferMemory(ai_prefix="Assistant"),
|
90 |
+
))
|
91 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
92 |
|
93 |
+
model_id2model = {chatbot.llm.repo_id: chatbot for chatbot in chatbots}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
demo = gr.Blocks()
|
96 |
|
|
|
104 |
"cnt": 0, "data": [],
|
105 |
"past_user_inputs": [],
|
106 |
"generated_responses": [],
|
|
|
|
|
|
|
|
|
107 |
}
|
108 |
+
for idx in range(len(chatbots)):
|
109 |
+
state_dict[f"response_{idx+1}"] = ""
|
110 |
state = gr.JSON(state_dict, visible=False)
|
111 |
|
112 |
gr.Markdown("# RLHF Interface")
|
|
|
117 |
# Generate model prediction
|
118 |
def _predict(txt, state):
|
119 |
# TODO: parallelize this!
|
120 |
+
start = time.time()
|
121 |
+
responses = generate_responses(chatbots, [txt] * len(chatbots))
|
122 |
+
print(f"Time taken (threading): {time.time() - start} seconds")
|
123 |
+
|
124 |
|
125 |
response2model_id = {}
|
126 |
+
for chatbot, response in zip(chatbots, responses):
|
127 |
+
response2model_id[response] = chatbot.llm.repo_id
|
|
|
|
|
128 |
|
129 |
state["cnt"] += 1
|
130 |
|
131 |
new_state_md = f"Inputs remaining in HIT: {state['cnt']}/{TOTAL_CNT}"
|
132 |
|
133 |
+
metadata = {"cnt": state["cnt"], "text": txt}
|
134 |
+
for idx, response in enumerate(responses):
|
135 |
+
metadata[f"response_{idx + 1}"] = response
|
136 |
+
|
137 |
+
metadata["response2model_id"] = response2model_id
|
138 |
+
|
139 |
+
state["data"].append(metadata)
|
140 |
state["past_user_inputs"].append(txt)
|
141 |
|
142 |
past_conversation_string = "<br />".join(["<br />".join(["π: " + user_input, "π€: " + model_response]) for user_input, model_response in zip(state["past_user_inputs"], state["generated_responses"] + [""])])
|
143 |
+
return gr.update(visible=False), gr.update(visible=True), gr.update(visible=True, choices=responses, interactive=True, value=responses[0]), gr.update(value=past_conversation_string), state, gr.update(visible=False), gr.update(visible=False), gr.update(visible=False), new_state_md, dummy
|
144 |
|
145 |
def _select_response(selected_response, state, dummy):
|
146 |
done = state["cnt"] == TOTAL_CNT
|
|
|
158 |
past_conversation_string = "<br />".join(["<br />".join(["π: " + user_input, "π€: " + model_response]) for user_input, model_response in zip(state["past_user_inputs"], state["generated_responses"])])
|
159 |
query = parse_qs(dummy[1:])
|
160 |
if "assignmentId" in query and query["assignmentId"][0] != "ASSIGNMENT_ID_NOT_AVAILABLE":
|
161 |
+
# It seems that someone is using this app on MTurk. We need to
|
162 |
# store the assignmentId in the state before submit_hit_button
|
163 |
# is clicked. We can do this here in _predict. We need to save the
|
164 |
# assignmentId so that the turker can get credit for their HIT.
|
|
|
171 |
|
172 |
if done:
|
173 |
# Wipe the memory completely because we will be starting a new hit soon.
|
174 |
+
for chatbot in chatbots:
|
175 |
+
chatbot.memory = ConversationBufferMemory(ai_prefix="Assistant")
|
|
|
|
|
176 |
else:
|
177 |
# Sync all of the model's memories with the conversation path that
|
178 |
# was actually taken.
|
179 |
+
for chatbot in chatbots:
|
180 |
+
chatbot.memory = model_id2model[state["data"][-1]["response2model_id"][selected_response]].memory
|
|
|
|
|
181 |
|
182 |
text_input = gr.update(visible=False) if done else gr.update(visible=True)
|
183 |
return gr.update(visible=False), gr.update(visible=True), text_input, gr.update(visible=False), state, gr.update(value=past_conversation_string), toggle_example_submit, toggle_final_submit, toggle_final_submit_preview,
|
|
|
192 |
with gr.Column(visible=False) as final_submit:
|
193 |
submit_hit_button = gr.Button("Submit HIT")
|
194 |
with gr.Column(visible=False) as final_submit_preview:
|
195 |
+
submit_hit_button_preview = gr.Button("Submit Work (preview mode; no MTurk HIT credit, but your examples will still be stored)")
|
196 |
|
197 |
# Button event handlers
|
198 |
get_window_location_search_js = """
|
|
|
217 |
|
218 |
post_hit_js = """
|
219 |
function(state) {
|
220 |
+
// If there is an assignmentId, then the submitter is on MTurk
|
221 |
// and has accepted the HIT. So, we need to submit their HIT.
|
222 |
const form = document.createElement('form');
|
223 |
form.action = 'https://workersandbox.mturk.com/mturk/externalSubmit';
|