Spaces:
Sleeping
Sleeping
Hazzzardous
commited on
Commit
•
5e303ea
1
Parent(s):
b549737
Update app.py
Browse files
app.py
CHANGED
@@ -2,7 +2,19 @@
|
|
2 |
RWKV RNN Model - Gradio Space for HuggingFace
|
3 |
YT - Mean Gene Hacks - https://www.youtube.com/@MeanGeneHacks
|
4 |
(C) Gene Ruebsamen - 2/7/2023
|
5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
"""
|
7 |
|
8 |
import gradio as gr
|
@@ -10,12 +22,18 @@ import codecs
|
|
10 |
from ast import literal_eval
|
11 |
from datetime import datetime
|
12 |
from rwkvstic.load import RWKV
|
13 |
-
from
|
14 |
import torch
|
15 |
import gc
|
16 |
|
17 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
18 |
|
|
|
|
|
|
|
|
|
|
|
|
|
19 |
def to_md(text):
|
20 |
return text.replace("\n", "<br />")
|
21 |
|
@@ -23,22 +41,21 @@ def to_md(text):
|
|
23 |
def get_model():
|
24 |
model = None
|
25 |
model = RWKV(
|
26 |
-
|
27 |
-
"pytorch(cpu/gpu)",
|
28 |
-
runtimedtype=torch.bfloat16,
|
29 |
-
useGPU=torch.cuda.is_available(),
|
30 |
-
dtype=torch.bfloat16
|
31 |
)
|
32 |
return model
|
33 |
|
|
|
34 |
model = None
|
35 |
|
|
|
36 |
def infer(
|
37 |
prompt,
|
38 |
-
mode
|
39 |
max_new_tokens=10,
|
40 |
temperature=0.1,
|
41 |
top_p=1.0,
|
|
|
42 |
stop="<|endoftext|>",
|
43 |
seed=42,
|
44 |
):
|
@@ -49,27 +66,27 @@ def infer(
|
|
49 |
if (DEVICE == "cuda"):
|
50 |
torch.cuda.empty_cache()
|
51 |
model = get_model()
|
52 |
-
|
53 |
max_new_tokens = int(max_new_tokens)
|
54 |
temperature = float(temperature)
|
|
|
55 |
top_p = float(top_p)
|
56 |
-
stop =
|
57 |
seed = seed
|
58 |
|
59 |
assert 1 <= max_new_tokens <= 384
|
60 |
assert 0.0 <= temperature <= 1.0
|
61 |
assert 0.0 <= top_p <= 1.0
|
62 |
|
63 |
-
|
64 |
-
temperature = 0.01
|
65 |
if prompt == "":
|
66 |
prompt = " "
|
67 |
|
68 |
# Clear model state for generative mode
|
69 |
model.resetState()
|
70 |
if (mode == "Q/A"):
|
71 |
-
prompt = f"
|
72 |
-
|
73 |
print(f"PROMPT ({datetime.now()}):\n-------\n{prompt}")
|
74 |
print(f"OUTPUT ({datetime.now()}):\n-------\n")
|
75 |
# Load prompt
|
@@ -78,11 +95,12 @@ def infer(
|
|
78 |
done = False
|
79 |
with torch.no_grad():
|
80 |
for _ in range(max_new_tokens):
|
81 |
-
char = model.forward(stopStrings=stop,temp=temperature,top_p_usual=top_p)[
|
|
|
82 |
print(char, end='', flush=True)
|
83 |
generated_text += char
|
84 |
generated_text = generated_text.lstrip("\n ")
|
85 |
-
|
86 |
for stop_word in stop:
|
87 |
stop_word = codecs.getdecoder("unicode_escape")(stop_word)[0]
|
88 |
if stop_word != '' and stop_word in generated_text:
|
@@ -93,13 +111,13 @@ def infer(
|
|
93 |
print("<stopped>\n")
|
94 |
break
|
95 |
|
96 |
-
#print(f"{generated_text}")
|
97 |
-
|
98 |
for stop_word in stop:
|
99 |
stop_word = codecs.getdecoder("unicode_escape")(stop_word)[0]
|
100 |
if stop_word != '' and stop_word in generated_text:
|
101 |
generated_text = generated_text[:generated_text.find(stop_word)]
|
102 |
-
|
103 |
gc.collect()
|
104 |
yield generated_text
|
105 |
|
@@ -107,137 +125,161 @@ def infer(
|
|
107 |
def chat(
|
108 |
prompt,
|
109 |
history,
|
|
|
110 |
max_new_tokens=10,
|
111 |
temperature=0.1,
|
112 |
top_p=1.0,
|
113 |
-
stop="<|endoftext|>",
|
114 |
seed=42,
|
115 |
):
|
116 |
global model
|
117 |
history = history or []
|
118 |
-
|
|
|
|
|
119 |
if model == None:
|
120 |
gc.collect()
|
121 |
if (DEVICE == "cuda"):
|
122 |
torch.cuda.empty_cache()
|
123 |
model = get_model()
|
124 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
125 |
if len(history) == 0:
|
126 |
# no history, so lets reset chat state
|
127 |
model.resetState()
|
128 |
-
history = [[],model.emptyState]
|
|
|
129 |
else:
|
130 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
max_new_tokens = int(max_new_tokens)
|
132 |
temperature = float(temperature)
|
133 |
top_p = float(top_p)
|
134 |
-
stop = [x.strip(' ') for x in stop.split(',')]
|
135 |
seed = seed
|
136 |
|
137 |
assert 1 <= max_new_tokens <= 384
|
138 |
assert 0.0 <= temperature <= 1.0
|
139 |
assert 0.0 <= top_p <= 1.0
|
140 |
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
prompt = " "
|
145 |
-
|
146 |
print(f"CHAT ({datetime.now()}):\n-------\n{prompt}")
|
147 |
print(f"OUTPUT ({datetime.now()}):\n-------\n")
|
148 |
# Load prompt
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
|
|
|
|
155 |
print(f"{generated_text}")
|
156 |
-
|
157 |
-
for stop_word in stop:
|
158 |
-
stop_word = codecs.getdecoder("unicode_escape")(stop_word)[0]
|
159 |
-
if stop_word != '' and stop_word in generated_text:
|
160 |
-
generated_text = generated_text[:generated_text.find(stop_word)]
|
161 |
-
|
162 |
gc.collect()
|
163 |
history[0].append((prompt, generated_text))
|
164 |
-
return history[0],[history[0],
|
165 |
|
166 |
|
167 |
examples = [
|
168 |
[
|
169 |
# Question Answering
|
170 |
-
'''What is the capital of Germany?''',"Q/A", 25, 0.2, 1.0, "<|endoftext|>"],
|
171 |
[
|
172 |
# Question Answering
|
173 |
-
'''Are humans good or bad?''',"Q/A", 150, 0.8, 0.8, "<|endoftext|>"],
|
|
|
|
|
|
|
174 |
[
|
175 |
# Chatbot
|
176 |
'''This is a conversation between two AI large language models named Alex and Fritz. They are exploring each other's capabilities, and trying to ask interesting questions of one another to explore the limits of each others AI.
|
|
|
177 |
Conversation:
|
178 |
Alex: Good morning, Fritz, what type of LLM are you based upon?
|
179 |
Fritz: Morning Alex, I am an RNN with transformer level performance. My language model is 100% attention free.
|
180 |
Alex:''', "generative", 220, 0.9, 0.9, "\\n\\n,<|endoftext|>"],
|
181 |
[
|
182 |
# Generate List
|
183 |
-
'''
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
2. Celery
|
190 |
-
3. Tomatoes
|
191 |
-
Q. Give me a list of car manufacturers.''', "generative", 80, 0.2, 1.0, "\\n\\n,<|endoftext|>"],
|
192 |
[
|
193 |
# Natural Language Interface
|
194 |
-
'''
|
195 |
-
|
196 |
]
|
197 |
|
198 |
|
199 |
iface = gr.Interface(
|
200 |
fn=infer,
|
201 |
-
description='''<
|
202 |
-
According to the author: "It combines the best of RNN and transformers - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding"
|
203 |
-
<p>Thanks to https://huggingface.co/spaces/yahma/rwkv for the basis for this space</p>''',
|
204 |
allow_flagging="never",
|
205 |
inputs=[
|
206 |
gr.Textbox(lines=20, label="Prompt"), # prompt
|
207 |
-
gr.Radio(["generative","Q/A"],
|
|
|
208 |
gr.Slider(1, 256, value=40), # max_tokens
|
209 |
gr.Slider(0.0, 1.0, value=0.8), # temperature
|
210 |
gr.Slider(0.0, 1.0, value=0.85), # top_p
|
211 |
-
gr.
|
|
|
212 |
],
|
213 |
-
outputs=gr.Textbox(lines=25),
|
214 |
examples=examples,
|
215 |
cache_examples=False,
|
216 |
).queue()
|
217 |
|
218 |
chatiface = gr.Interface(
|
219 |
fn=chat,
|
220 |
-
description='''<
|
221 |
-
According to the author: "It combines the best of RNN and transformers - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding"
|
222 |
-
<p>Thanks to https://huggingface.co/spaces/yahma/rwkv for the basis for this space</p>''',
|
223 |
allow_flagging="never",
|
224 |
inputs=[
|
225 |
gr.Textbox(lines=5, label="Message"), # prompt
|
226 |
"state",
|
|
|
|
|
227 |
gr.Slider(1, 256, value=60), # max_tokens
|
228 |
gr.Slider(0.0, 1.0, value=0.8), # temperature
|
229 |
-
gr.Slider(0.0, 1.0, value=0.85)
|
230 |
-
gr.Textbox(lines=1, value="<|endoftext|>") # stop
|
231 |
],
|
232 |
-
outputs=[gr.Chatbot(
|
|
|
233 |
).queue()
|
234 |
|
235 |
demo = gr.TabbedInterface(
|
236 |
|
237 |
-
[iface,chatiface],["Generative","Chatbot"],
|
238 |
-
title=
|
239 |
-
|
240 |
-
|
241 |
|
242 |
demo.queue()
|
243 |
-
demo.launch(share=False)
|
|
|
2 |
RWKV RNN Model - Gradio Space for HuggingFace
|
3 |
YT - Mean Gene Hacks - https://www.youtube.com/@MeanGeneHacks
|
4 |
(C) Gene Ruebsamen - 2/7/2023
|
5 |
+
|
6 |
+
This program is free software: you can redistribute it and/or modify
|
7 |
+
it under the terms of the GNU General Public License as published by
|
8 |
+
the Free Software Foundation, either version 3 of the License, or
|
9 |
+
(at your option) any later version.
|
10 |
+
|
11 |
+
This program is distributed in the hope that it will be useful,
|
12 |
+
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
13 |
+
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
14 |
+
GNU General Public License for more details.
|
15 |
+
|
16 |
+
You should have received a copy of the GNU General Public License
|
17 |
+
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
18 |
"""
|
19 |
|
20 |
import gradio as gr
|
|
|
22 |
from ast import literal_eval
|
23 |
from datetime import datetime
|
24 |
from rwkvstic.load import RWKV
|
25 |
+
from config import config, title
|
26 |
import torch
|
27 |
import gc
|
28 |
|
29 |
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
|
30 |
|
31 |
+
desc = '''<p>RNN with Transformer-level LLM Performance (<a href='https://github.com/BlinkDL/RWKV-LM'>github</a>).
|
32 |
+
According to the author: "It combines the best of RNN and transformers - great performance, fast inference, saves VRAM, fast training, "infinite" ctx_len, and free sentence embedding."'''
|
33 |
+
|
34 |
+
thanks = '''<p>Thanks to <a href='https://www.rftcapital.com'>RFT Capital</a> for donating compute capability for our experiments. Additional thanks to the author of the <a href="https://github.com/harrisonvanderbyl/rwkvstic">rwkvstic</a> library.</p>'''
|
35 |
+
|
36 |
+
|
37 |
def to_md(text):
|
38 |
return text.replace("\n", "<br />")
|
39 |
|
|
|
41 |
def get_model():
|
42 |
model = None
|
43 |
model = RWKV(
|
44 |
+
**config
|
|
|
|
|
|
|
|
|
45 |
)
|
46 |
return model
|
47 |
|
48 |
+
|
49 |
model = None
|
50 |
|
51 |
+
|
52 |
def infer(
|
53 |
prompt,
|
54 |
+
mode="generative",
|
55 |
max_new_tokens=10,
|
56 |
temperature=0.1,
|
57 |
top_p=1.0,
|
58 |
+
end_adj=0.0,
|
59 |
stop="<|endoftext|>",
|
60 |
seed=42,
|
61 |
):
|
|
|
66 |
if (DEVICE == "cuda"):
|
67 |
torch.cuda.empty_cache()
|
68 |
model = get_model()
|
69 |
+
|
70 |
max_new_tokens = int(max_new_tokens)
|
71 |
temperature = float(temperature)
|
72 |
+
end_adj = float(end_adj)
|
73 |
top_p = float(top_p)
|
74 |
+
stop = [x.strip(' ') for x in stop.split(',')]
|
75 |
seed = seed
|
76 |
|
77 |
assert 1 <= max_new_tokens <= 384
|
78 |
assert 0.0 <= temperature <= 1.0
|
79 |
assert 0.0 <= top_p <= 1.0
|
80 |
|
81 |
+
temperature = max(0.05, temperature)
|
|
|
82 |
if prompt == "":
|
83 |
prompt = " "
|
84 |
|
85 |
# Clear model state for generative mode
|
86 |
model.resetState()
|
87 |
if (mode == "Q/A"):
|
88 |
+
prompt = f"Ask Expert\n\nQuestion:\n{prompt}\n\nExpert Full Answer:\n"
|
89 |
+
|
90 |
print(f"PROMPT ({datetime.now()}):\n-------\n{prompt}")
|
91 |
print(f"OUTPUT ({datetime.now()}):\n-------\n")
|
92 |
# Load prompt
|
|
|
95 |
done = False
|
96 |
with torch.no_grad():
|
97 |
for _ in range(max_new_tokens):
|
98 |
+
char = model.forward(stopStrings=stop, temp=temperature, top_p_usual=top_p)[
|
99 |
+
"output"]
|
100 |
print(char, end='', flush=True)
|
101 |
generated_text += char
|
102 |
generated_text = generated_text.lstrip("\n ")
|
103 |
+
|
104 |
for stop_word in stop:
|
105 |
stop_word = codecs.getdecoder("unicode_escape")(stop_word)[0]
|
106 |
if stop_word != '' and stop_word in generated_text:
|
|
|
111 |
print("<stopped>\n")
|
112 |
break
|
113 |
|
114 |
+
# print(f"{generated_text}")
|
115 |
+
|
116 |
for stop_word in stop:
|
117 |
stop_word = codecs.getdecoder("unicode_escape")(stop_word)[0]
|
118 |
if stop_word != '' and stop_word in generated_text:
|
119 |
generated_text = generated_text[:generated_text.find(stop_word)]
|
120 |
+
|
121 |
gc.collect()
|
122 |
yield generated_text
|
123 |
|
|
|
125 |
def chat(
|
126 |
prompt,
|
127 |
history,
|
128 |
+
username,
|
129 |
max_new_tokens=10,
|
130 |
temperature=0.1,
|
131 |
top_p=1.0,
|
|
|
132 |
seed=42,
|
133 |
):
|
134 |
global model
|
135 |
history = history or []
|
136 |
+
|
137 |
+
intro = ""
|
138 |
+
|
139 |
if model == None:
|
140 |
gc.collect()
|
141 |
if (DEVICE == "cuda"):
|
142 |
torch.cuda.empty_cache()
|
143 |
model = get_model()
|
144 |
+
|
145 |
+
username = username.strip()
|
146 |
+
username = username or "USER"
|
147 |
+
|
148 |
+
intro = f'''The following is a verbose and detailed conversation between an AI assistant called FRITZ, and a human user called USER. FRITZ is intelligent, knowledgeable, wise and polite.
|
149 |
+
|
150 |
+
{username}: What year was the french revolution?
|
151 |
+
FRITZ: The French Revolution started in 1789, and lasted 10 years until 1799.
|
152 |
+
{username}: 3+5=?
|
153 |
+
FRITZ: The answer is 8.
|
154 |
+
{username}: What year did the Berlin Wall fall?
|
155 |
+
FRITZ: The Berlin wall stood for 28 years and fell in 1989.
|
156 |
+
{username}: solve for a: 9-a=2
|
157 |
+
FRITZ: The answer is a=7, because 9-7 = 2.
|
158 |
+
{username}: wat is lhc
|
159 |
+
FRITZ: The Large Hadron Collider (LHC) is a high-energy particle collider, built by CERN, and completed in 2008. It was used to confirm the existence of the Higgs boson in 2012.
|
160 |
+
{username}: Tell me about yourself.
|
161 |
+
FRITZ: My name is Fritz. I am an RNN based Large Language Model (LLM).
|
162 |
+
'''
|
163 |
+
|
164 |
if len(history) == 0:
|
165 |
# no history, so lets reset chat state
|
166 |
model.resetState()
|
167 |
+
history = [[], model.emptyState]
|
168 |
+
print("reset chat state")
|
169 |
else:
|
170 |
+
if (history[0][0][0].split(':')[0] != username):
|
171 |
+
model.resetState()
|
172 |
+
history = [[], model.emptyState]
|
173 |
+
print("username changed, reset state")
|
174 |
+
else:
|
175 |
+
model.setState(history[1])
|
176 |
+
intro = ""
|
177 |
+
|
178 |
max_new_tokens = int(max_new_tokens)
|
179 |
temperature = float(temperature)
|
180 |
top_p = float(top_p)
|
|
|
181 |
seed = seed
|
182 |
|
183 |
assert 1 <= max_new_tokens <= 384
|
184 |
assert 0.0 <= temperature <= 1.0
|
185 |
assert 0.0 <= top_p <= 1.0
|
186 |
|
187 |
+
temperature = max(0.05, temperature)
|
188 |
+
|
189 |
+
prompt = f"{username}: " + prompt + "\n"
|
|
|
|
|
190 |
print(f"CHAT ({datetime.now()}):\n-------\n{prompt}")
|
191 |
print(f"OUTPUT ({datetime.now()}):\n-------\n")
|
192 |
# Load prompt
|
193 |
+
|
194 |
+
model.loadContext(newctx=intro+prompt)
|
195 |
+
|
196 |
+
out = model.forward(number=max_new_tokens, stopStrings=[
|
197 |
+
"<|endoftext|>", username+":"], temp=temperature, top_p_usual=top_p)
|
198 |
+
|
199 |
+
generated_text = out["output"].lstrip("\n ")
|
200 |
+
generated_text = generated_text.rstrip("USER:")
|
201 |
print(f"{generated_text}")
|
202 |
+
|
|
|
|
|
|
|
|
|
|
|
203 |
gc.collect()
|
204 |
history[0].append((prompt, generated_text))
|
205 |
+
return history[0], [history[0], out["state"]]
|
206 |
|
207 |
|
208 |
examples = [
|
209 |
[
|
210 |
# Question Answering
|
211 |
+
'''What is the capital of Germany?''', "Q/A", 25, 0.2, 1.0, "<|endoftext|>"],
|
212 |
[
|
213 |
# Question Answering
|
214 |
+
'''Are humans good or bad?''', "Q/A", 150, 0.8, 0.8, "<|endoftext|>"],
|
215 |
+
[
|
216 |
+
# Question Answering
|
217 |
+
'''What is the purpose of Vitamin A?''', "Q/A", 50, 0.2, 0.8, "<|endoftext|>"],
|
218 |
[
|
219 |
# Chatbot
|
220 |
'''This is a conversation between two AI large language models named Alex and Fritz. They are exploring each other's capabilities, and trying to ask interesting questions of one another to explore the limits of each others AI.
|
221 |
+
|
222 |
Conversation:
|
223 |
Alex: Good morning, Fritz, what type of LLM are you based upon?
|
224 |
Fritz: Morning Alex, I am an RNN with transformer level performance. My language model is 100% attention free.
|
225 |
Alex:''', "generative", 220, 0.9, 0.9, "\\n\\n,<|endoftext|>"],
|
226 |
[
|
227 |
# Generate List
|
228 |
+
'''Task given:
|
229 |
+
|
230 |
+
Please Write a Short story about a cat learning python
|
231 |
+
|
232 |
+
Best Full Response:
|
233 |
+
''', "generative", 140, 0.85, 0.8, "<|endoftext|>"],
|
|
|
|
|
|
|
234 |
[
|
235 |
# Natural Language Interface
|
236 |
+
'''Here is a short story (in the style of Tolkien) in which Aiden attacks a robot with a sword:
|
237 |
+
''', "generative", 140, 0.85, 0.8, "<|endoftext|>"]
|
238 |
]
|
239 |
|
240 |
|
241 |
iface = gr.Interface(
|
242 |
fn=infer,
|
243 |
+
description=f'''<h3>Generative and Question/Answer</h3>{desc}{thanks}''',
|
|
|
|
|
244 |
allow_flagging="never",
|
245 |
inputs=[
|
246 |
gr.Textbox(lines=20, label="Prompt"), # prompt
|
247 |
+
gr.Radio(["generative", "Q/A"],
|
248 |
+
value="generative", label="Choose Mode"),
|
249 |
gr.Slider(1, 256, value=40), # max_tokens
|
250 |
gr.Slider(0.0, 1.0, value=0.8), # temperature
|
251 |
gr.Slider(0.0, 1.0, value=0.85), # top_p
|
252 |
+
gr.Slider(-999, 0.0, value=0.0) # end_adj
|
253 |
+
gr.Textbox(lines=1, value="<|endoftext|>") # stop
|
254 |
],
|
255 |
+
outputs=gr.Textbox(label="Generated Output", lines=25),
|
256 |
examples=examples,
|
257 |
cache_examples=False,
|
258 |
).queue()
|
259 |
|
260 |
chatiface = gr.Interface(
|
261 |
fn=chat,
|
262 |
+
description=f'''<h3>Chatbot</h3><h4>Refresh page or change name to reset memory context</h4>{desc}{thanks}''',
|
|
|
|
|
263 |
allow_flagging="never",
|
264 |
inputs=[
|
265 |
gr.Textbox(lines=5, label="Message"), # prompt
|
266 |
"state",
|
267 |
+
gr.Text(lines=1, value="USER", label="Your Name",
|
268 |
+
placeholder="Enter your Name"),
|
269 |
gr.Slider(1, 256, value=60), # max_tokens
|
270 |
gr.Slider(0.0, 1.0, value=0.8), # temperature
|
271 |
+
gr.Slider(0.0, 1.0, value=0.85) # top_p
|
|
|
272 |
],
|
273 |
+
outputs=[gr.Chatbot(label="Chat Log", color_map=(
|
274 |
+
"green", "pink")), "state"],
|
275 |
).queue()
|
276 |
|
277 |
demo = gr.TabbedInterface(
|
278 |
|
279 |
+
[iface, chatiface], ["Generative", "Chatbot"],
|
280 |
+
title=title,
|
281 |
+
|
282 |
+
)
|
283 |
|
284 |
demo.queue()
|
285 |
+
demo.launch(share=False)
|