-
Notifications
You must be signed in to change notification settings - Fork 188
/
constants.py
233 lines (195 loc) · 7.83 KB
/
constants.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
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
import ast
import getpass
import os
from functools import partial
from pathlib import Path
import datasets
from huggingface_hub import hf_hub_download
CURRENT_DIR = Path(__file__).parent.absolute()
BASE_DIR = Path(__file__).parents[2].absolute()
### API specific ###
API_MAX_CONCURRENCY = int(os.environ.get("API_MAX_CONCURRENCY", 5))
OPENAI_MAX_CONCURRENCY = int(os.environ.get("OPENAI_MAX_CONCURRENCY", 5))
OPENAI_CLIENT_CONFIG_PATH = os.environ.get("OPENAI_CLIENT_CONFIG_PATH", BASE_DIR / "client_configs/openai_configs.yaml")
# the following is for backward compatibility, the recommended way is to use OPENAI_CLIENT_CONFIG_PATH
OPENAI_API_KEYS = os.environ.get("OPENAI_API_KEYS", os.environ.get("OPENAI_API_KEY", None))
if isinstance(OPENAI_API_KEYS, str):
OPENAI_API_KEYS = OPENAI_API_KEYS.split(",")
OPENAI_ORGANIZATION_IDS = os.environ.get("OPENAI_ORGANIZATION_IDS", None)
if isinstance(OPENAI_ORGANIZATION_IDS, str):
OPENAI_ORGANIZATION_IDS = OPENAI_ORGANIZATION_IDS.split(",")
#
ANTHROPIC_API_KEY = os.environ.get("ANTHROPIC_API_KEY", None)
ANTHROPIC_MAX_CONCURRENCY = int(os.environ.get("ANTHROPIC_MAX_CONCURRENCY", 4))
GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY", None)
COHERE_API_KEY = os.environ.get("COHERE_API_KEY", None)
DATASETS_TOKEN = os.environ.get("DATASETS_TOKEN", None)
HUGGINGFACEHUB_API_TOKEN = os.environ.get("HUGGINGFACEHUB_API_TOKEN", None)
DATASETS_FORCE_DOWNLOAD = os.environ.get("DATASETS_FORCE_DOWNLOAD", False)
########################
IS_ALPACA_EVAL_2 = ast.literal_eval(os.environ.get("IS_ALPACA_EVAL_2", "True"))
ANNOTATOR_CONFIG_AE1 = "alpaca_eval_gpt4"
ANNOTATOR_CONFIG_AE2 = "weighted_alpaca_eval_gpt4_turbo"
DEFAULT_ANNOTATOR_CONFIG = ANNOTATOR_CONFIG_AE2 if IS_ALPACA_EVAL_2 else ANNOTATOR_CONFIG_AE1
DEFAULT_CACHE_DIR = None
EVALUATORS_CONFIG_DIR = CURRENT_DIR / "evaluators_configs"
MODELS_CONFIG_DIR = CURRENT_DIR / "models_configs"
MINIMAL_EVALUATORS = (
ANNOTATOR_CONFIG_AE2,
ANNOTATOR_CONFIG_AE1,
"aviary_gpt4",
"gpt4",
"claude",
"text_davinci_003",
"chatgpt",
"lmsys_gpt4",
"humans",
"alpaca_farm_greedy_gpt4",
)
VERIFIED_EVALUATORS = tuple(
list(MINIMAL_EVALUATORS)
+ [
"claude_ranking",
"improved_aviary_gpt4",
"improved_lmsys_gpt4",
"lmsys_gpt4",
"cohere",
"alpaca_farm",
"alpaca_farm_greedy_gpt4",
"guanaco_33b",
"longest",
]
)
# order matters i => i+1 when filtering
ORDERED_LEADERBOARD_MODES = ["minimal", "verified", "community", "dev"]
def get_alpaca_eval_data(dataset="alpaca_eval_gpt4_baseline"):
dataset = datasets.load_dataset(
"tatsu-lab/alpaca_eval",
dataset,
cache_dir=DEFAULT_CACHE_DIR,
token=DATASETS_TOKEN,
download_mode="force_redownload" if DATASETS_FORCE_DOWNLOAD else None,
trust_remote_code=True,
)["eval"]
return dataset
ALPACAEVAL_REFERENCE_OUTPUTS_2 = get_alpaca_eval_data
ALPACAEVAL_REFERENCE_OUTPUTS_1 = partial(get_alpaca_eval_data, dataset="alpaca_eval")
ALPACAEVAL_REFERENCE_OUTPUTS = ALPACAEVAL_REFERENCE_OUTPUTS_2 if IS_ALPACA_EVAL_2 else ALPACAEVAL_REFERENCE_OUTPUTS_1
def ALPACAEVAL_INSTRUCTION_PARAMETERS():
out = hf_hub_download(
repo_id="tatsu-lab/alpaca_eval",
filename="instruction_difficulty.csv",
repo_type="dataset",
force_download=DATASETS_FORCE_DOWNLOAD,
cache_dir=DEFAULT_CACHE_DIR,
token=DATASETS_TOKEN,
)
pd.read_csv(out, index_col=0).squeeze()
return df
def ALPACAFARM_GOLD_CROSSANNOTATIONS():
df = datasets.load_dataset(
"tatsu-lab/alpaca_eval",
"alpaca_farm_human_crossannotations",
cache_dir=DEFAULT_CACHE_DIR,
token=DATASETS_TOKEN,
download_mode="force_redownload" if DATASETS_FORCE_DOWNLOAD else None,
trust_remote_code=True,
)["validation"].to_pandas()
# turkers took around 9 min for 15 examples in AlpacaFarm
df["time_per_example"] = 9.2 * 60 / 15
df["price_per_example"] = 0.3 # price we paid for each example
return df
def ALPACAFARM_GOLD_ANNOTATIONS():
df = datasets.load_dataset(
"tatsu-lab/alpaca_eval",
"alpaca_farm_human_annotations",
cache_dir=DEFAULT_CACHE_DIR,
token=DATASETS_TOKEN,
download_mode="force_redownload" if DATASETS_FORCE_DOWNLOAD else None,
trust_remote_code=True,
)["validation"].to_pandas()
# turkers took around 9 min for 15 examples in AlpacaFarm
df["time_per_example"] = 9.2 * 60 / 15
df["price_per_example"] = 0.3 # price we paid for each example
return df
ALPACAEVAL_2_LEADERBOARD_PATHS = CURRENT_DIR / f"leaderboards/data_AlpacaEval_2"
ALPACAEVAL_1_LEADERBOARD_PATHS = CURRENT_DIR / f"leaderboards/data_AlpacaEval"
ALPACAEVAL_LEADERBOARD_PATHS = ALPACAEVAL_2_LEADERBOARD_PATHS if IS_ALPACA_EVAL_2 else ALPACAEVAL_1_LEADERBOARD_PATHS
PRECOMPUTED_LEADERBOARDS = {
(str(ALPACAEVAL_REFERENCE_OUTPUTS_1), "claude"): ALPACAEVAL_1_LEADERBOARD_PATHS / "claude_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_1), ANNOTATOR_CONFIG_AE1): ALPACAEVAL_1_LEADERBOARD_PATHS
/ f"{ANNOTATOR_CONFIG_AE1}_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_1), "chatgpt_fn"): ALPACAEVAL_1_LEADERBOARD_PATHS / "chatgpt_fn_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_2), ANNOTATOR_CONFIG_AE2): ALPACAEVAL_2_LEADERBOARD_PATHS
/ f"{ANNOTATOR_CONFIG_AE2}_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_2), "weighted_alpaca_eval_gpt4_turbo"): ALPACAEVAL_2_LEADERBOARD_PATHS
/ f"weighted_alpaca_eval_gpt4_turbo_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_2), "mistral-large-2402_ranking"): ALPACAEVAL_2_LEADERBOARD_PATHS
/ f"mistral-large-2402_ranking_leaderboard.csv",
(str(ALPACAEVAL_REFERENCE_OUTPUTS_2), "claude_3_opus_ranking"): ALPACAEVAL_2_LEADERBOARD_PATHS
/ f"claude_3_opus_ranking_leaderboard.csv",
# (str(ALPACAEVAL_REFERENCE_OUTPUTS_2), "gpt-3.5-turbo-1106_ranking"): ALPACAEVAL_2_LEADERBOARD_PATHS
# / f"gpt-3.5-turbo-1106_ranking_leaderboard.csv",
# (str(ALPACAEVAL_REFERENCE_OUTPUTS_2), "alpaca_eval_cot_gpt4_turbo_fn"): ALPACAEVAL_2_LEADERBOARD_PATHS
# / f"alpaca_eval_cot_gpt4_turbo_fn_leaderboard.csv",
}
HUMAN_ANNOTATED_MODELS_TO_KEEP = (
"GPT-4 300 characters",
"GPT-4",
"AlpacaFarm PPO sim (step 40)",
"ChatGPT",
"ChatGPT 300 characters",
"AlpacaFarm best-of-16 human",
"AlpacaFarm PPO sim (gpt4 greedy, step 30)",
"Davinci003",
"AlpacaFarm ExpIter human (n=128)",
"AlpacaFarm SFT 10K",
"AlpacaFarm PPO human (10k, step 40)",
"Alpaca 7B",
"AlpacaFarm FeedMe human",
"Davinci001",
"LLaMA 7B",
)
EVALUATORS_LEADERBOARD_COLS_TO_PRIORITIZE = [
"Human agreement",
"Price [$/1000 examples]",
"Time [seconds/1000 examples]",
"Spearman corr.",
"Pearson corr.",
"Bias",
"Variance",
"Proba. prefer longer",
"Proba. prefer lists",
"Proba. prefer 1",
]
MINIMAL_MODELS_FOR_NEW_LEADERBOARD = [
"gpt4_turbo",
"gpt4",
"tulu-2-dpo-70b",
"Yi-34B-Chat",
"llama-2-70b-chat-hf",
"claude-2.1",
"cohere",
"chatgpt",
"gemini-pro",
"Mixtral-8x7B-Instruct-v0.1",
"Mistral-7B-Instruct-v0.2",
"vicuna-33b-v1.3",
"alpaca-7b",
]
EVALUATORS_LEADERBOARD_COLS_TO_PRINT = EVALUATORS_LEADERBOARD_COLS_TO_PRIORITIZE[:8]
CURRENT_USER = getpass.getuser()
if CURRENT_USER in ["yanndubs"]:
DEFAULT_CACHE_DIR = "/juice5/scr5/nlp/crfm/human-feedback/cache"
def ALPACAFARM_ALL_OUTPUTS():
if IS_ALPACA_EVAL_2:
return [f"results/{m}/model_outputs.json" for m in MINIMAL_MODELS_FOR_NEW_LEADERBOARD]
else:
return datasets.load_dataset(
"tatsu-lab/alpaca_eval",
"alpaca_eval_all_outputs",
cache_dir=DEFAULT_CACHE_DIR,
token=DATASETS_TOKEN,
download_mode="force_redownload" if DATASETS_FORCE_DOWNLOAD else None,
trust_remote_code=True,
)["eval"]