-
Notifications
You must be signed in to change notification settings - Fork 0
/
kaggle_blender_run.py
178 lines (158 loc) · 6.04 KB
/
kaggle_blender_run.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
import numpy as np
import llm_blender
from llm_blender.blender.blender_utils import get_topk_candidates_from_ranks
import time
import os
import json
import datasets
import requests
from evaluate import load
def install_llms() -> list:
print('######################### Installing LLMS #########################')
os.system('ollama pull mistral')
os.system('ollama pull gemma')
os.system('ollama pull qwen2')
os.system('ollama pull llama3')
os.system('ollama pull deepseek-llm')
os.system('ollama pull orca-mini')
os.system('ollama pull phi')
return ['mistral', 'gemma', 'qwen2', 'llama3', 'deepseek-llm', 'orca-mini', 'phi']
def dataset_init_conv_questions() -> datasets.arrow_dataset.Dataset:
# to return a smaller number of question sets
# must be in the format 'dataset_name[i]['questions']'
dataset = datasets.load_dataset("conv_questions", trust_remote_code=True)
train_data = dataset['train'].select(range(13, 14))
# test_data = dataset['train'].select(range(2))
# validation_data = dataset['train'].select(range(2))
return train_data
def dataset_init_atlas_converse():
dataset_atlas = []
with open('atlas-converse\combined-convo.json', 'r') as f:
li = json.load(f)
f.close()
dataset_atlas.extend(li[:33])
with open('atlas-converse\combined-convo_2.json', 'r') as f:
li = json.load(f)
f.close()
dataset_atlas.extend(li[:33])
with open('atlas-converse\combined-convo_3.json', 'r') as f:
li = json.load(f)
f.close()
dataset_atlas.extend(li[:34])
return dataset_atlas
def generate(llm: str, prompt: str, context: str) -> str:
# url = 'https://996c-35-193-113-166.ngrok-free.app/api/generate'
url = 'http://127.0.0.1:11434/api/generate'
f = open('input.json', 'r') # /kaggle/input/treefiles/input.json - for kaggle
data = json.load(f)
data['model'] = llm
data['prompt'] = "Answer the question concisely: " + prompt + ", given the context: " + context
response = requests.post(url, json=data)
assert response.status_code == 200
dictionary = json.loads(response.text)
f.close()
return dictionary['response']
def llm_blender_Conv(llm_list, dataset, bertscore):
filename = 'blender_ans.txt'
scores_file = 'blender_scores.txt'
blender = llm_blender.Blender()
blender.loadranker("llm-blender/PairRM")
blender.loadfuser("llm-blender/gen_fuser_3b")
fused_answers = []
final_scores = []
fa = []
sa = []
for i in range(len(dataset)):
print(i)
ans_set = []
scores = []
context = ""
for j in range(len(dataset[i]['questions'])):
question = dataset[i]['questions'][j]
candidates_texts = []
for llm in llm_list:
candidates_texts.append(generate(llm, question, context))
ranks = blender.rank(
[dataset[i]['questions'][j]], [candidates_texts], return_scores=False, batch_size=1)
topk_candidates = get_topk_candidates_from_ranks(
ranks, [candidates_texts], top_k=2)
fuse_generations = blender.fuse(
[dataset[i]['questions'][j]], topk_candidates, batch_size=2)
context = context + fuse_generations[0]
ans_set.append(fuse_generations[0])
scores.append(bertscore.compute(predictions = [fuse_generations[0]], references = [dataset[i]['answer_texts'][j]], lang="en")['f1'])
fa.append(ans_set)
sa.append(scores)
if i%1==0:
with open(filename, 'a') as f:
for line in fa:
f.write(f"{line}\n")
f.close()
fused_answers.extend(fa)
fa = []
with open(scores_file, 'a') as f:
for line in sa:
f.write(f"{line}\n")
f.close()
final_scores.extend(sa)
sa = []
with open(filename, 'a') as f:
for line in fa:
f.write(f"{line}\n")
f.close()
with open(scores_file, 'a') as f:
for line in sa:
f.write(f"{line}\n")
f.close()
return fused_answers
def atlas_blender_run(llmList, dataset)->list:
blender = llm_blender.Blender()
blender.loadranker("llm-blender/PairRM")
blender.loadfuser("llm-blender/gen_fuser_3b")
filename = 'Experiment_Results\\atlas_results.txt'
# scores_files = 'Experiment_Results\\atlas_scores.txt'
fused_answers = []
# final_scores = []
fa = []
# sa = []
for i in range(len(dataset)):
ans_set = []
scores = []
context = ""
for j in range(len(dataset[i]['conversations'])):
if j+1>=len(dataset[i]['conversations']) or dataset[i]['conversations'][j]['from'] == 'AI':
continue
question = dataset[i]['conversations'][j]['value']
candidates_texts = []
for llm in llmList:
response = generate(llm, question, context)
candidates_texts.append(response)
fuse_generations = blender.fuse(
[dataset[i]['conversations'][j]['value']], [candidates_texts], batch_size=2)
ans_set.append(fuse_generations[0])
context += fuse_generations[0]
fa.append(ans_set)
if i%1==0:
with open(filename, 'a') as f:
for line in fa:
f.write(f"{line}\n")
f.close()
fused_answers.extend(fa)
fa = []
with open(filename, 'a') as f:
for line in fa:
f.write(f"{line}\n")
f.close()
return fused_answers
if __name__=='__main__':
time.sleep(15)
str = "{\"model\":\"\",\"prompt\":\"\",\"stream\":false}"
with open('input.json', 'w') as f:
f.write(str)
f.close()
llm_list = install_llms()
dataset_conv = dataset_init_conv_questions()
dataset_atlas = dataset_init_atlas_converse()
bertscore = load("bertscore")
# llm_blender_Conv(llm_list, dataset, bertscore)
atlas_blender_run(llm_list, dataset_atlas)