File size: 8,486 Bytes
933d496 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 |
import datasets
from datasets import load_dataset
import json
from tqdm import tqdm
fw = open("/apdcephfs_gy2/share_303094202/bazzfeng/data/math_sft_bigbig.jsonl", "w+")
fw2 = open("/apdcephfs_gy2/share_303094202/bazzfeng/data/science_sft_bigbig.jsonl", "w+")
udict = {}
mydata = load_dataset('TIGER-Lab/WebInstruct-verified')
for item in mydata['train']:
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0} if item['category']=='Mathematics' else {"loss_mask": [0,1], "topic": "科学", "is_business": 0}
answer = item['answer']
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item['question'], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
if item['category']=='Mathematics':
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
else:
fw2.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in mydata['test']:
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0} if item['category']=='Mathematics' else {"loss_mask": [0,1], "topic": "科学", "is_business": 0}
answer = item['answer']
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item['question'], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
if item['category']=='Mathematics':
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
else:
fw2.write(json.dumps(new_d, ensure_ascii=False)+"\n")
mydata = load_dataset('Skywork/Skywork-OR1-RL-Data')
#fw = open("/apdcephfs_gy2/share_303094202/bazzfeng/data/skywork_deepmath.jsonl", "w+")
for item in mydata['math']:
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
answer = item['reward_model']['ground_truth']
new_d['ref_answer'] = answer
new_d['messages'] = item['prompt']
new_d['messages'].append({"content": answer, "role": "assistant"})
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
#break
mydata = load_dataset('zwhe99/DeepMath-103K')
for item in mydata['train']:
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
answer = item['final_answer']
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
#break
lans = ['ar', 'bn', 'de', 'en', 'es', 'fr', 'id', 'it', 'ja', 'ko', 'ms', 'pt', 'ru', 'sw', 'te', 'th', 'vi', 'zh']
for lan in lans:
mydata = load_dataset('Qwen/PolyMath', lan)
for item in mydata['top']:
answer = item['answer']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in mydata['high']:
answer = item['answer']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in mydata['medium']:
answer = item['answer']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in mydata['low']:
answer = item['answer']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
mydata = load_dataset('nvidia/OpenMathReasoning')
for item in tqdm(mydata['cot']):
answer = item['generated_solution']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = item['expected_answer']
new_d['messages'] = [{"content": item["problem"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in tqdm(mydata['tir']):
answer = item['generated_solution']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = item['expected_answer']
new_d['messages'] = [{"content": item["problem"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in tqdm(mydata['genselect']):
answer = item['generated_solution']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = item['expected_answer']
new_d['messages'] = [{"content": item["problem"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
"""
mydata = load_dataset('nvidia/Nemotron-CrossThink')
for item in mydata['train_math']:
answer = item['reward_model']['ground_truth']
new_d = {"loss_mask": [0,1], "topic": "数学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["meta_data"]["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw.write(json.dumps(new_d, ensure_ascii=False)+"\n")
for item in mydata['train_qa']:
answer = item['reward_model']['ground_truth']
new_d = {"loss_mask": [0,1], "topic": "科学", "is_business": 0}
new_d['ref_answer'] = answer
new_d['messages'] = [{"content": item["meta_data"]["question"], "role": "user"}, {"content": answer, "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw2.write(json.dumps(new_d, ensure_ascii=False)+"\n")
"""
#mydata = load_dataset('FreedomIntelligence/medical-o1-reasoning-SFT')
lans = ['en','en_mix','zh','zh_mix']
for lan in lans:
mydata = load_dataset('FreedomIntelligence/medical-o1-reasoning-SFT', lan)
for item in mydata['train']:
new_d = {"loss_mask": [0,1], "topic": "科学", "is_business": 0}
new_d['ref_answer'] = item['Response']
new_d['messages'] = [{"content": item['Question'], "role": "user"}, {"content": item['Response'], "role": "assistant"}]
if new_d['messages'][0]['content'][:50] in udict:
continue
else:
udict[new_d['messages'][0]['content'][:50]] = 1
fw2.write(json.dumps(new_d, ensure_ascii=False)+"\n")
#new_d['messages'].append({"content": item['output'], "role": "assistant"})
#print(mydata)
|