File size: 9,281 Bytes
651b7a3 b7d8a7d 651b7a3 98b4762 b7d8a7d 98b4762 651b7a3 98b4762 651b7a3 b7d8a7d 40a88e8 651b7a3 c10563c 651b7a3 043c386 651b7a3 043c386 b7d8a7d 40a88e8 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 |
"""
Test module for sharegpt integration w chatml
"""
import pytest
from datasets import Dataset
from tokenizers import AddedToken
from transformers import AutoTokenizer
from axolotl.datasets import TokenizedPromptDataset
from axolotl.prompt_strategies.sharegpt import (
GlaiveShareGPTPromptTokenizingStrategy,
SimpleShareGPTPromptTokenizingStrategy,
register_chatml_template,
)
from axolotl.prompters import ShareGPTPrompterV2
register_chatml_template()
@pytest.fixture(name="sharegpt_dataset")
def fixture_sharegpt_dataset():
return Dataset.from_list(
[
{
"conversations": [
{
"from": "system",
"value": "repeat",
},
{
"from": "human",
"value": "hello",
},
{
"from": "gpt",
"value": "hello",
},
{
"from": "human",
"value": "goodbye",
},
{
"from": "gpt",
"value": "goodbye",
},
]
}
]
)
@pytest.fixture(name="glaive_dataset")
def fixture_sharegpt_glaive_dataset():
return Dataset.from_list(
[
{
"system": "SYSTEM: This is a system prompt",
"chat": "USER: Can you book a flight for me from New York to London? ASSISTANT: I'm sorry, but I don't have the capability to book flights. <|endoftext|>",
}
]
)
@pytest.fixture(name="multi_role_dataset")
def fixture_multi_role_dataset():
return Dataset.from_list(
[
{
"conversations": [
{
"from": "system",
"value": "use get_weather(city) to get the weather for a city",
},
{
"from": "human",
"value": "hello, what's the weather in New York?",
},
{
"from": "gpt",
"value": "let me get that for you",
},
{
"from": "tool",
"value": "get_weather(New York)",
},
{
"from": "gpt",
"value": "the weather in New York is 70 degrees and sunny",
},
]
}
]
)
@pytest.fixture(name="tokenizer")
def fixture_tokenizer():
tokenizer = AutoTokenizer.from_pretrained(
"casperhansen/mistral-7b-instruct-v0.1-awq"
)
tokenizer.add_special_tokens(
{
"eos_token": AddedToken(
"<|im_end|>", rstrip=False, lstrip=False, normalized=False
)
}
)
tokenizer.add_tokens(
[
AddedToken("<|im_start|>", rstrip=False, lstrip=False, normalized=False),
]
)
return tokenizer
class TestSharegpt:
"""
Test class for sharegpt prompter
"""
def test_no_double_im_end(self, sharegpt_dataset, tokenizer):
strategy = SimpleShareGPTPromptTokenizingStrategy(
ShareGPTPrompterV2(
conversation="chatml",
role_key_model=None,
role_key_human=None,
),
tokenizer,
False, # train_on_inputs
2048, # sequence_len
)
dataset_wrapper = TokenizedPromptDataset(
strategy, sharegpt_dataset, process_count=1
)
input_ids = dataset_wrapper[0]["input_ids"]
# fmt: off
assert input_ids == [
# 28705, 13, is " \n"
1, # bos
32001, 1587, 13, 25997, 32000, 28705, 13, # system
32001, 2188, 13, 21558, 32000, 28705, 13, # human
32001, 13892, 13, 21558, 32000, 28705, 13, # gpt
32001, 2188, 13, 12684, 17664, 32000, 28705, 13, # human
32001, 13892, 13, 12684, 17664, 32000, 28705, 13, # gpt
]
# fmt: on
def test_w_train_on_input(self, sharegpt_dataset, tokenizer):
strategy = SimpleShareGPTPromptTokenizingStrategy(
ShareGPTPrompterV2(
conversation="chatml",
role_key_model=None,
role_key_human=None,
),
tokenizer,
False, # train_on_inputs
2048, # sequence_len
)
dataset_wrapper = TokenizedPromptDataset(
strategy, sharegpt_dataset, process_count=1
)
labels = dataset_wrapper[0]["labels"]
# fmt: off
assert labels == [
-100, # bos
-100, -100, -100, -100, -100, -100, -100, # system
-100, -100, -100, -100, -100, -100, -100, # human
-100, -100, 13, 21558, 32000, 28705, 13, # gpt
-100, -100, -100, -100, -100, -100, -100, -100, # human
-100, -100, 13, 12684, 17664, 32000, 28705, 13, # gpt
]
# fmt: on
def test_no_train_on_input(self, sharegpt_dataset, tokenizer):
strategy = SimpleShareGPTPromptTokenizingStrategy(
ShareGPTPrompterV2(
conversation="chatml",
role_key_model=None,
role_key_human=None,
),
tokenizer,
True, # train_on_inputs
2048, # sequence_len
)
dataset_wrapper = TokenizedPromptDataset(
strategy, sharegpt_dataset, process_count=1
)
labels = dataset_wrapper[0]["labels"]
# fmt: off
assert labels == [
1, # bos
32001, 1587, 13, 25997, 32000, 28705, 13, # system
32001, 2188, 13, 21558, 32000, 28705, 13, # human
32001, 13892, 13, 21558, 32000, 28705, 13, # gpt
32001, 2188, 13, 12684, 17664, 32000, 28705, 13, # human
32001, 13892, 13, 12684, 17664, 32000, 28705, 13, # gpt
]
# fmt: on
def test_chatml_glaive(self, glaive_dataset, tokenizer):
strategy = GlaiveShareGPTPromptTokenizingStrategy(
ShareGPTPrompterV2(
conversation="chatml",
role_key_model=None,
role_key_human=None,
),
tokenizer,
True, # train_on_inputs
2048, # sequence_len
)
dataset_wrapper = TokenizedPromptDataset(
strategy, glaive_dataset, process_count=1
)
labels = dataset_wrapper[0]["labels"]
# fmt: off
assert labels == [
1, # bos
32001, 1587, 13, 3260, 349, 264, 1587, 11510, 32000, 28705, 13, # system
32001, 2188, 13, 6325, 368, 1820, 264, 9314, 354, 528, 477, 1450, 2726, 298, 4222, 28804, 32000, 28705, 13, # human
32001, 13892, 13, 28737, 28742, 28719, 7371, 28725, 562, 315, 949, 28742, 28707, 506, 272, 21368, 298, 1820, 22447, 28723, 28705, 523, 28766, 416, 1009, 772, 28766, 28767, 32000, 28705, 13 # gpt
]
# fmt: on
def test_multi_role_dataset(self, multi_role_dataset, tokenizer):
strategy = SimpleShareGPTPromptTokenizingStrategy(
ShareGPTPrompterV2(conversation="chatml", roles={"input": ["tool"]}),
tokenizer,
False, # train_on_inputs
2048, # sequence_len
)
dataset_wrapper = TokenizedPromptDataset(
strategy, multi_role_dataset, process_count=1
)
input_ids = dataset_wrapper[0]["input_ids"]
# fmt: off
assert input_ids == [
1, # bos
32001, 1587, 13, 1730, 625, 28730, 769, 1223, 28732, 18373, 28731, 298, 625, 272, 8086, 354, 264, 2990, 32000, 28705, 13, # system
32001, 2188, 13, 21558, 28725, 767, 28742, 28713, 272, 8086, 297, 1450, 2726, 28804, 32000, 28705, 13, # human
32001, 13892, 13, 895, 528, 625, 369, 354, 368, 32000, 28705, 13, # gpt
32001, 3921, 13, 527, 28730, 769, 1223, 28732, 2972, 2726, 28731, 32000, 28705, 13, # tool
32001, 13892, 13, 1237, 8086, 297, 1450, 2726, 349, 28705, 28787, 28734, 11182, 304, 4376, 1780, 32000, 28705, 13 # gpt
]
# fmt: on
labels = dataset_wrapper[0]["labels"]
# fmt: off
assert labels == [
-100, # bos
-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, # system
-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, # human
-100, -100, 13, 895, 528, 625, 369, 354, 368, 32000, 28705, 13, # gpt
-100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, -100, # tool
-100, -100, 13, 1237, 8086, 297, 1450, 2726, 349, 28705, 28787, 28734, 11182, 304, 4376, 1780, 32000, 28705, 13 # gpt
]
# fmt: on
|