Spaces:
Sleeping
Sleeping
Camila Salinas Camacho
commited on
Commit
•
081b46f
1
Parent(s):
98e913c
Update app.py
Browse files
app.py
CHANGED
@@ -1,246 +1,25 @@
|
|
1 |
import gradio as gr
|
2 |
-
import
|
3 |
-
import
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
iface = gr.Interface(fn=greet, inputs="text", outputs="text")
|
23 |
-
iface.launch()
|
24 |
-
|
25 |
-
# determinant vs. determiner
|
26 |
-
# https://wikidiff.com/determiner/determinant
|
27 |
-
ents_prompt = [
|
28 |
-
'Noun',
|
29 |
-
'Verb',
|
30 |
-
'Adjective',
|
31 |
-
'Adverb',
|
32 |
-
'Preposition/Subord',
|
33 |
-
'Coordinating Conjunction',
|
34 |
-
# 'Cardinal Number',
|
35 |
-
'Determiner',
|
36 |
-
'Noun Phrase',
|
37 |
-
'Verb Phrase',
|
38 |
-
'Adjective Phrase',
|
39 |
-
'Adverb Phrase',
|
40 |
-
'Preposition Phrase',
|
41 |
-
'Conjunction Phrase',
|
42 |
-
'Coordinate Phrase',
|
43 |
-
'Quantitave Phrase',
|
44 |
-
'Complex Nominal',
|
45 |
-
'Clause',
|
46 |
-
'Dependent Clause',
|
47 |
-
'Fragment Clause',
|
48 |
-
'T-unit',
|
49 |
-
'Complex T-unit',
|
50 |
-
# 'Fragment T-unit',
|
51 |
-
]
|
52 |
-
ents = ['NN', 'VB', 'JJ', 'RB', 'IN', 'CC', 'DT', 'NP', 'VP', 'ADJP', 'ADVP', 'PP', 'CONJP', 'CP', 'QP', 'CN', 'C', 'DC', 'FC', 'T', 'CT']
|
53 |
-
|
54 |
-
|
55 |
-
model_mapping = {
|
56 |
-
# 'gpt3': 'gpt-3',
|
57 |
-
'gpt3.5': 'gpt-3.5-turbo-0613',
|
58 |
-
'vicuna-7b': 'lmsys/vicuna-7b-v1.3',
|
59 |
-
'vicuna-13b': 'lmsys/vicuna-13b-v1.3',
|
60 |
-
'vicuna-33b': 'lmsys/vicuna-33b-v1.3',
|
61 |
-
'fastchat-t5': 'lmsys/fastchat-t5-3b-v1.0',
|
62 |
-
# 'llama2': 'meta-llama/Llama-2-7b-chat-hf',
|
63 |
-
'llama-7b': '/data/jiali/llama/hf/7B',
|
64 |
-
'llama-13b': '/data/jiali/llama/hf/13B',
|
65 |
-
'llama-30b': '/data/jiali/llama/hf/30B',
|
66 |
-
'llama-65b': '/data/jiali/llama/hf/65B',
|
67 |
-
'alpaca': '/data/jiali/alpaca-7B',
|
68 |
-
# 'koala-7b': 'koala-7b',
|
69 |
-
# 'koala-13b': 'koala-13b',
|
70 |
-
}
|
71 |
-
|
72 |
-
for m in model_mapping.keys():
|
73 |
-
for eid, ent in enumerate(ents):
|
74 |
-
os.makedirs(f'result/openai_result/{m}/ptb/per_ent/{ent}', exist_ok=True)
|
75 |
-
os.makedirs(f'result/structured_prompt/{m}/ptb', exist_ok=True)
|
76 |
-
|
77 |
-
|
78 |
-
# s = int(sys.argv[1])
|
79 |
-
# e = int(sys.argv[2])
|
80 |
-
|
81 |
-
s = 0
|
82 |
-
e = 1000
|
83 |
-
with open('ptb_corpus/sample_uniform_1k_2.txt', 'r') as f:
|
84 |
-
selected_idx = f.readlines()
|
85 |
-
selected_idx = [int(i.strip()) for i in selected_idx][s:e]
|
86 |
-
|
87 |
-
|
88 |
-
ptb = []
|
89 |
-
with open('./ptb_corpus/ptb.jsonl', 'r') as f:
|
90 |
-
for l in f:
|
91 |
-
ptb.append(json.loads(l))
|
92 |
-
|
93 |
-
|
94 |
-
## Prompt 1
|
95 |
-
template_all = '''Please output the <Noun, Verb, Adjective, Adverb, Preposition/Subord, Coordinating Conjunction, Cardinal Number, Determiner, Noun Phrase, Verb Phrase, Adjective Phrase, Adverb Phrase, Preposition Phrase, Conjunction Phrase, Coordinate Phrase, Quantitave Phrase, Complex Nominal, Clause, Dependent Clause, Fragment Clause, T-unit, Complex T-unit, Fragment T-unit> in the following sentence without any additional text in json format: "{}"'''
|
96 |
-
template_single = '''Please output any <{}> in the following sentence one per line without any additional text: "{}"'''
|
97 |
-
|
98 |
-
## Prompt 2
|
99 |
-
with open('ptb_corpus/structured_prompting_demonstration_42.txt', 'r') as f:
|
100 |
-
demonstration = f.read()
|
101 |
-
|
102 |
-
|
103 |
-
def para(m):
|
104 |
-
c = 0
|
105 |
-
for n, p in m.named_parameters():
|
106 |
-
c += p.numel()
|
107 |
-
return c
|
108 |
-
|
109 |
-
def main(args=None):
|
110 |
-
|
111 |
-
if 'gpt3' in args.model:
|
112 |
-
pass
|
113 |
-
|
114 |
-
else:
|
115 |
-
path = model_mapping[args.model]
|
116 |
-
model, tokenizer = load_model(
|
117 |
-
path,
|
118 |
-
args.device,
|
119 |
-
args.num_gpus,
|
120 |
-
args.max_gpu_memory,
|
121 |
-
args.load_8bit,
|
122 |
-
args.cpu_offloading,
|
123 |
-
revision=args.revision,
|
124 |
-
debug=args.debug,
|
125 |
-
)
|
126 |
-
|
127 |
-
if args.prompt == 1:
|
128 |
-
for gid in tqdm(selected_idx, desc='Query'):
|
129 |
-
text = ptb[gid]['text']
|
130 |
-
|
131 |
-
for eid, ent in enumerate(ents):
|
132 |
-
# if os.path.exists(f'result/openai_result/{args.model}/ptb/per_ent/{ent}/{gid}.pkl') or \
|
133 |
-
# os.path.exists(f'result/openai_result/{args.model}/ptb/per_ent/{ent}/{gid}.txt'):
|
134 |
-
# print(gid, ent, 'skip')
|
135 |
-
# continue
|
136 |
-
|
137 |
-
## Get prompt
|
138 |
-
msg = template_single.format(ents_prompt[eid], text)
|
139 |
-
|
140 |
-
if 'gpt' in args.model:
|
141 |
-
prompt = msg
|
142 |
-
|
143 |
-
elif 'vicuna' in args.model or 'alpaca' in args.model or 'fastchat-t5' in args.model:
|
144 |
-
conv = get_conversation_template(args.model)
|
145 |
-
conv.append_message(conv.roles[0], msg)
|
146 |
-
conv.append_message(conv.roles[1], None)
|
147 |
-
conv.system = ''
|
148 |
-
prompt = conv.get_prompt().strip()
|
149 |
-
|
150 |
-
elif 'llama-' in args.model:
|
151 |
-
prompt = '### Human: ' + msg + ' ### Assistant:'
|
152 |
-
|
153 |
-
|
154 |
-
## Run
|
155 |
-
if 'gpt3' in args.model:
|
156 |
-
outputs = gpt3(prompt)
|
157 |
-
|
158 |
-
else:
|
159 |
-
outputs = fastchat(prompt, model, tokenizer)
|
160 |
-
|
161 |
-
with open(f'result/openai_result/{args.model}/ptb/per_ent/{ent}/{gid}.txt', 'w') as f:
|
162 |
-
f.write(outputs)
|
163 |
-
|
164 |
-
|
165 |
-
if args.prompt == 2:
|
166 |
-
for gid in tqdm(selected_idx, desc='Query'):
|
167 |
-
text = ptb[gid]['text']
|
168 |
-
|
169 |
-
if os.path.exists(f'result/structured_prompt/{args.model}/ptb/{gid}.pkl') or \
|
170 |
-
os.path.exists(f'result/structured_prompt/{args.model}/ptb/{gid}.txt'):
|
171 |
-
print(gid, 'skip')
|
172 |
-
continue
|
173 |
-
|
174 |
-
prompt = demonstration + '\n' + text
|
175 |
-
|
176 |
-
if 'gpt3' in args.model:
|
177 |
-
outputs = gpt3(prompt)
|
178 |
-
|
179 |
-
else:
|
180 |
-
outputs = fastchat(prompt, model, tokenizer)
|
181 |
-
|
182 |
-
with open(f'result/structured_prompt/{args.model}/ptb/{gid}.txt', 'w') as f:
|
183 |
-
f.write(outputs)
|
184 |
-
|
185 |
-
|
186 |
-
def fastchat(prompt, model, tokenizer):
|
187 |
-
input_ids = tokenizer([prompt]).input_ids
|
188 |
-
output_ids = model.generate(
|
189 |
-
torch.as_tensor(input_ids).cuda(),
|
190 |
-
do_sample=True,
|
191 |
-
temperature=args.temperature,
|
192 |
-
repetition_penalty=args.repetition_penalty,
|
193 |
-
max_new_tokens=args.max_new_tokens,
|
194 |
-
)
|
195 |
-
|
196 |
-
if model.config.is_encoder_decoder:
|
197 |
-
output_ids = output_ids[0]
|
198 |
-
else:
|
199 |
-
output_ids = output_ids[0][len(input_ids[0]) :]
|
200 |
-
outputs = tokenizer.decode(
|
201 |
-
output_ids, skip_special_tokens=True, spaces_between_special_tokens=False
|
202 |
-
)
|
203 |
-
|
204 |
-
# print('Empty system message')
|
205 |
-
# print(f"{conv.roles[0]}: {msg}")
|
206 |
-
# print(f"{conv.roles[1]}: {outputs}")
|
207 |
-
|
208 |
-
return outputs
|
209 |
-
|
210 |
-
|
211 |
-
def gpt3(prompt):
|
212 |
-
try:
|
213 |
-
response = openai.ChatCompletion.create(
|
214 |
-
model=args.model, messages=[{"role": "user", "content": prompt}])
|
215 |
-
|
216 |
-
return response
|
217 |
-
|
218 |
-
except Exception as err:
|
219 |
-
print('Error')
|
220 |
-
print(err)
|
221 |
-
|
222 |
-
# time.sleep(1)
|
223 |
-
raise
|
224 |
-
|
225 |
|
226 |
if __name__ == "__main__":
|
227 |
-
|
228 |
-
|
229 |
-
parser.add_argument("--temperature", type=float, default=0.7)
|
230 |
-
parser.add_argument("--repetition_penalty", type=float, default=1.0)
|
231 |
-
parser.add_argument("--max-new-tokens", type=int, default=512)
|
232 |
-
parser.add_argument("--debug", action="store_true")
|
233 |
-
parser.add_argument("--message", type=str, default="Hello! Who are you?")
|
234 |
-
parser.add_argument("--start", type=int, default=0)
|
235 |
-
parser.add_argument("--end", type=int, default=1)
|
236 |
-
parser.add_argument("--model", required=True, type=str, default=None)
|
237 |
-
parser.add_argument("--prompt", required=True, type=int, default=None)
|
238 |
-
args = parser.parse_args()
|
239 |
-
|
240 |
-
# Reset default repetition penalty for T5 models.
|
241 |
-
if "t5" in args.model and args.repetition_penalty == 1.0:
|
242 |
-
args.repetition_penalty = 1.2
|
243 |
-
|
244 |
-
main(args)
|
245 |
-
|
246 |
-
|
|
|
1 |
import gradio as gr
|
2 |
+
import subprocess
|
3 |
+
from gradio.mix import Parallel
|
4 |
+
|
5 |
+
def qa_prompting(model):
|
6 |
+
# Call your `run_llm.py` script for QA-Based Prompting with the selected model
|
7 |
+
output = subprocess.check_output([sys.executable, "run_llm.py", "--model", model, ...], text=True)
|
8 |
+
return output
|
9 |
+
|
10 |
+
def strategy_1_interface():
|
11 |
+
model_names = ["ChatGPT", "LLaMA", "Vicuna", "Alpaca", "Flan-T5"]
|
12 |
+
interfaces = []
|
13 |
+
for model_name in model_names:
|
14 |
+
interfaces.append(gr.Interface(
|
15 |
+
fn=qa_prompting,
|
16 |
+
inputs=gr.inputs.Textbox(label=f"{model_name} Input"),
|
17 |
+
outputs=gr.outputs.Textbox(label=f"{model_name} Output"),
|
18 |
+
title=f"Strategy 1 - QA-Based Prompting: {model_name}",
|
19 |
+
))
|
20 |
+
|
21 |
+
return Parallel(*interfaces)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
22 |
|
23 |
if __name__ == "__main__":
|
24 |
+
iface = strategy_1_interface()
|
25 |
+
iface.launch()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|