Skip to content

GitLab

"docs/vscode:/vscode.git/clone" did not exist on "708ea9d7e06ee09df35f66150e21abc5ab40af5f"
Commit be3dfa50 authored by jerrrrry's avatar jerrrrry
Browse files

Initial commit

parents
Pipeline #2876 failed with stages
in 0 seconds
Hide whitespace changes
Inline Side-by-side
Showing with 1329 additions and 0 deletions
opencompass/configs/datasets/apps/deprecated_apps_gen_5b4254.py 0 → 100644
View file @ be3dfa50
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import HFDataset, HumanEvalEvaluator, humaneval_postprocess
apps_reader_cfg = dict(
input_columns=['question'], output_column='problem_id', train_split='test')
# TODO: allow empty output-column
apps_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(role='HUMAN', prompt='Write a python program:\n{question}'),
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer))
apps_eval_cfg = dict(
evaluator=dict(type=HumanEvalEvaluator),
pred_role='BOT',
k=[1, 10, 100], # the parameter only for humaneval
pred_postprocessor=dict(type=humaneval_postprocess),
)
apps_datasets = [
dict(
type=HFDataset,
path='codeparrot/apps',
reader_cfg=apps_reader_cfg,
infer_cfg=apps_infer_cfg,
eval_cfg=apps_eval_cfg)
]
opencompass/configs/datasets/apps/deprecated_apps_gen_7fbb95.py 0 → 100644
View file @ be3dfa50
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import HFDataset, HumanEvalEvaluator, humaneval_postprocess
apps_reader_cfg = dict(
input_columns=['question'], output_column='problem_id', train_split='test')
# TODO: allow empty output-column
apps_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
begin=[
dict(
role='SYSTEM',
fallback_role='HUMAN',
prompt='Write a python program:'),
],
round=[
dict(role='HUMAN', prompt='{question}'),
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer))
apps_eval_cfg = dict(
evaluator=dict(type=HumanEvalEvaluator),
pred_role='BOT',
k=[1, 10, 100], # the parameter only for humaneval
pred_postprocessor=dict(type=humaneval_postprocess),
)
apps_datasets = [
dict(
type=HFDataset,
path='codeparrot/apps',
reader_cfg=apps_reader_cfg,
infer_cfg=apps_infer_cfg,
eval_cfg=apps_eval_cfg)
]
opencompass/configs/datasets/apps/deprecated_apps_gen_b4dee3.py 0 → 100644
View file @ be3dfa50
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import HFDataset, HumanEvalEvaluator, humaneval_postprocess
apps_reader_cfg = dict(
input_columns=['question'], output_column='problem_id', train_split='test')
# TODO: allow empty output-column
apps_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template='{question}'),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
apps_eval_cfg = dict(
evaluator=dict(type=HumanEvalEvaluator),
k=[1, 10, 100],
pred_postprocessor=dict(type=humaneval_postprocess),
)
apps_datasets = [
dict(
type=HFDataset,
path='codeparrot/apps',
reader_cfg=apps_reader_cfg,
infer_cfg=apps_infer_cfg,
eval_cfg=apps_eval_cfg)
]
opencompass/configs/datasets/babilong/README.md 0 → 100644
View file @ be3dfa50
# BABILong
OpenCompass now supports the brand new long-context language model evaluation benchmark — [BABILong](https://arxiv.org/pdf/2406.10149). BABILong provides an evaluation of long-context reasoning across extremely long documents, including a diverse set of 20 reasoning tasks such as fact chaining, simple induction, deduction, counting, and handling lists/sets. This benchmark is designed to test the ability of language models to reason over facts distributed in long natural text, and it allows for the construction of tasks of almost arbitrary length to adapt to the evaluation of new, more powerful models in an extensible and controllable way.
## How to Use
The BABILong dataset is available on Hugging Face: [RMT-team/babilong](https://huggingface.co/datasets/RMT-team/babilong). Opencompass provides an automatic download for BABILong dataset, due to the dataset size, we only provide the data up to 1M tokens. For longer context, you can download the dataset from Hugging Face directly.
BABILong paper provides in total 20 tasks, we provide 10 tasks configurations in OpenCompass and they are organized by different context sizes. You can create your own configurations by following the examples in `opencompass/configs/datasets/babilong/babilong_1m_gen.py`.
Opencompass provides a demo for evaluating language models on the BABILong dataset.
```bash
opencompass configs/eval_babilong.py
```
OpenCompass provides the results of some models on the BABILong dataset. The evaluation results are run with LMDeploy with default model settings.
| dataset | version | metric | mode | internlm2_5-7b-chat-turbomind | qwen2.5-7b-instruct-turbomind | llama-3_1-8b-instruct-turbomind | ministral-8B-instruct-2410-turbomind |
|----- | ----- | ----- | ----- | ----- | ----- | ----- | -----|
| babilong_0k | - | naive_average | gen | 76.51 | 80.25 | 76.44 | 76.40 |
| babilong_4k | - | naive_average | gen | 67.55 | 70.35 | 67.41 | 67.92 |
| babilong_16k | - | naive_average | gen | 53.78 | 65.83 | 60.26 | 56.58 |
| babilong_32k | - | naive_average | gen | 50.86 | 62.66 | 59.56 | 53.52 |
| babilong_128k | - | naive_average | gen | 39.33 | 27.79 | 52.01 | 3.20 |
| babilong_256k | - | naive_average | gen | 17.31 | 7.30 | 23.35 | 9.50 |
## Citation
```bibtex
@misc{kuratov2024babilong,
title={BABILong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack},
author={Yuri Kuratov and Aydar Bulatov and Petr Anokhin and Ivan Rodkin and Dmitry Sorokin and Artyom Sorokin and Mikhail Burtsev},
year={2024},
eprint={2406.10149},
archivePrefix={arXiv}
}
```
\ No newline at end of file
opencompass/configs/datasets/babilong/babilong_0k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_0k_datasets = []
split_name='0k'
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_0k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_128k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_128k_datasets = []
split_name='128k'
max_seq_len = 128*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_seq_len=max_seq_len),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_128k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_16k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_16k_datasets = []
split_name='16k'
max_seq_len = 16*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_seq_len=max_seq_len),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_16k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_1m_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_1m_datasets = []
split_name='1m'
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_1m_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_256k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_256k_datasets = []
split_name='256k'
max_seq_len = 256*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_seq_len=max_seq_len ),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_256k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_2k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_2k_datasets = []
split_name='2k'
max_seq_len = 2*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer， max_seq_len=max_seq_len),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_2k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_32k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_32k_datasets = []
split_name='32k'
max_seq_len = 32*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_seq_len=max_seq_len),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_32k_datasets.append(tmp_dataset)
opencompass/configs/datasets/babilong/babilong_4k_gen.py 0 → 100644
View file @ be3dfa50
from opencompass.datasets.babilong.babilong import BabiLongDataset, BabiLongEvaluator
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
babiLong_4k_datasets = []
split_name='4k'
max_seq_len=4*1024
tasks = ['qa1', 'qa2', 'qa3', 'qa4', 'qa5', 'qa6', 'qa7', 'qa8', 'qa9', 'qa10']
for task in tasks:
tmp_dataset = {
'abbr': f'babilong_{task}_{split_name}',
'type': BabiLongDataset,
'path': 'opencompass/babilong',
'task': task,
'split_name': split_name,
'reader_cfg': dict(input_columns=['prompt'], output_column='answer'),
'infer_cfg': dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(
round=[
dict(role='HUMAN', prompt='{prompt}'),
dict(role='BOT', prompt='{answer}\n'),
]
),
),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_seq_len=max_seq_len),
),
'eval_cfg': dict(
evaluator=dict(type=BabiLongEvaluator),
),
}
babiLong_4k_datasets.append(tmp_dataset)
opencompass/configs/datasets/bbh/README.md 0 → 100644
View file @ be3dfa50
# BBH
```bash
python3 run.py --models hf_internlm2_7b --datasets bbh_gen_98fba6 --debug
python3 run.py --models hf_internlm2_chat_7b --datasets bbh_gen_5b92b0 --debug
```
## Base Models
| model | bbh |
|:------------------------:|------:|
| llama-7b-turbomind | 33.34 |
| llama-13b-turbomind | 37.99 |
| llama-30b-turbomind | 49.86 |
| llama-65b-turbomind | 58.26 |
| llama-2-7b-turbomind | 38.27 |
| llama-2-13b-turbomind | 45.68 |
| llama-2-70b-turbomind | 64.78 |
| llama-3-8b-turbomind | 59.69 |
| llama-3-70b-turbomind | 79.16 |
| internlm2-1.8b-turbomind | 36.03 |
| internlm2-7b-turbomind | 63.56 |
| internlm2-20b-turbomind | 71.29 |
| qwen-1.8b-turbomind | 22.53 |
| qwen-7b-turbomind | 45.89 |
| qwen-14b-turbomind | 56.75 |
| qwen-72b-turbomind | 63.35 |
| qwen1.5-0.5b-hf | 20.54 |
| qwen1.5-1.8b-hf | 27.01 |
| qwen1.5-4b-hf | 34.81 |
| qwen1.5-7b-hf | 39.87 |
| qwen1.5-14b-hf | 50.38 |
| qwen1.5-32b-hf | 67.47 |
| qwen1.5-72b-hf | 58.81 |
| qwen1.5-moe-a2-7b-hf | 39.46 |
| mistral-7b-v0.1-hf | 56.71 |
| mistral-7b-v0.2-hf | 57.32 |
| mixtral-8x7b-v0.1-hf | 68.46 |
| mixtral-8x22b-v0.1-hf | 79.48 |
| yi-6b-hf | 44.82 |
| yi-34b-hf | 66.37 |
| deepseek-7b-base-hf | 42.88 |
| deepseek-67b-base-hf | 71.86 |
### Details
| model | temporal_sequences | disambiguation_qa | date_understanding | tracking_shuffled_objects_three_objects | penguins_in_a_table | geometric_shapes | snarks | ruin_names | tracking_shuffled_objects_seven_objects |
|:------------------------:|---------------------:|--------------------:|---------------------:|------------------------------------------:|----------------------:|-------------------:|---------:|-------------:|------------------------------------------:|
| llama-7b-turbomind | 23.60 | 46.00 | 44.80 | 36.40 | 30.14 | 0.00 | 46.07 | 21.60 | 15.20 |
| llama-13b-turbomind | 16.80 | 50.00 | 56.80 | 36.40 | 43.15 | 0.00 | 60.67 | 29.20 | 15.20 |
| llama-30b-turbomind | 33.60 | 60.00 | 76.40 | 29.20 | 57.53 | 0.00 | 59.55 | 62.40 | 17.20 |
| llama-65b-turbomind | 84.00 | 76.00 | 84.40 | 50.00 | 65.75 | 0.00 | 62.92 | 69.60 | 31.60 |
| llama-2-7b-turbomind | 12.00 | 46.80 | 60.00 | 34.00 | 32.19 | 0.00 | 49.44 | 32.80 | 18.40 |
| llama-2-13b-turbomind | 24.00 | 40.80 | 73.20 | 36.00 | 45.89 | 0.00 | 55.06 | 37.60 | 22.40 |
| llama-2-70b-turbomind | 75.60 | 66.80 | 88.80 | 73.60 | 69.86 | 0.00 | 73.60 | 60.80 | 57.60 |
| llama-3-8b-turbomind | 65.60 | 42.00 | 78.80 | 56.80 | 69.86 | 0.00 | 56.18 | 66.00 | 30.80 |
| llama-3-70b-turbomind | 100.00 | 82.80 | 91.60 | 100.00 | 86.30 | 0.00 | 81.46 | 77.20 | 94.40 |
| internlm2-1.8b-turbomind | 31.20 | 44.00 | 60.00 | 36.00 | 35.62 | 0.00 | 44.94 | 27.20 | 12.80 |
| internlm2-7b-turbomind | 94.80 | 75.60 | 86.40 | 53.60 | 69.18 | 0.00 | 59.55 | 68.00 | 46.00 |
| internlm2-20b-turbomind | 98.40 | 83.60 | 84.00 | 72.00 | 71.92 | 0.00 | 81.46 | 78.40 | 74.40 |
| qwen-1.8b-turbomind | 26.40 | 39.60 | 33.20 | 28.40 | 28.08 | 0.00 | 44.94 | 21.60 | 12.40 |
| qwen-7b-turbomind | 38.80 | 42.80 | 64.40 | 30.80 | 45.89 | 0.00 | 55.62 | 44.00 | 14.40 |
| qwen-14b-turbomind | 57.60 | 59.20 | 67.20 | 46.40 | 67.12 | 0.00 | 51.12 | 63.60 | 30.40 |
| qwen-72b-turbomind | 72.00 | 66.80 | 77.60 | 81.20 | 84.93 | 0.00 | 78.09 | 67.20 | 63.60 |
| qwen1.5-0.5b-hf | 15.20 | 37.20 | 20.40 | 30.40 | 18.49 | 8.40 | 44.94 | 11.20 | 14.00 |
| qwen1.5-1.8b-hf | 27.60 | 40.80 | 36.00 | 24.40 | 32.19 | 0.00 | 50.56 | 20.80 | 11.20 |
| qwen1.5-4b-hf | 10.40 | 44.40 | 47.20 | 36.80 | 44.52 | 24.80 | 46.63 | 20.80 | 14.80 |
| qwen1.5-7b-hf | 37.20 | 42.40 | 52.00 | 52.40 | 56.85 | 6.80 | 48.31 | 23.60 | 18.40 |
| qwen1.5-14b-hf | 38.80 | 62.80 | 73.60 | 24.80 | 69.86 | 26.80 | 66.29 | 52.80 | 2.00 |
| qwen1.5-32b-hf | 93.60 | 77.20 | 68.40 | 70.00 | 82.88 | 36.80 | 47.75 | 70.40 | 71.20 |
| qwen1.5-72b-hf | 75.60 | 66.00 | 78.80 | 72.80 | 80.82 | 0.00 | 75.84 | 64.80 | 44.40 |
| qwen1.5-moe-a2-7b-hf | 23.20 | 59.60 | 43.20 | 27.60 | 46.58 | 25.20 | 48.88 | 16.80 | 13.20 |
| mistral-7b-v0.1-hf | 73.60 | 53.60 | 76.40 | 45.20 | 56.85 | 28.00 | 64.04 | 66.00 | 21.60 |
| mistral-7b-v0.2-hf | 76.80 | 42.00 | 73.20 | 47.20 | 60.27 | 26.00 | 66.85 | 60.80 | 26.40 |
| mixtral-8x7b-v0.1-hf | 89.60 | 70.80 | 84.80 | 81.20 | 70.55 | 25.60 | 66.29 | 71.20 | 58.80 |
| mixtral-8x22b-v0.1-hf | 98.80 | 77.60 | 92.00 | 98.80 | 83.56 | 35.60 | 80.34 | 79.20 | 82.00 |
| yi-6b-hf | 32.80 | 46.40 | 64.40 | 34.40 | 47.26 | 28.80 | 60.11 | 45.60 | 14.00 |
| yi-34b-hf | 86.00 | 76.00 | 84.80 | 54.80 | 67.81 | 24.80 | 73.60 | 66.00 | 65.60 |
| deepseek-7b-base-hf | 27.60 | 42.00 | 64.40 | 31.20 | 40.41 | 33.60 | 52.25 | 46.00 | 13.20 |
| deepseek-67b-base-hf | 95.60 | 75.60 | 86.40 | 86.40 | 76.71 | 39.20 | 76.40 | 77.20 | 82.00 |
| model | tracking_shuffled_objects_five_objects | logical_deduction_three_objects | hyperbaton | logical_deduction_five_objects | logical_deduction_seven_objects | movie_recommendation | salient_translation_error_detection | reasoning_about_colored_objects | multistep_arithmetic_two |
|:------------------------:|-----------------------------------------:|----------------------------------:|-------------:|---------------------------------:|----------------------------------:|-----------------------:|--------------------------------------:|----------------------------------:|---------------------------:|
| llama-7b-turbomind | 18.40 | 42.80 | 58.00 | 23.20 | 13.20 | 40.00 | 16.40 | 30.40 | 0.00 |
| llama-13b-turbomind | 16.00 | 48.80 | 53.60 | 30.40 | 16.40 | 61.60 | 11.20 | 44.80 | 0.80 |
| llama-30b-turbomind | 22.40 | 66.40 | 73.20 | 43.60 | 31.60 | 84.40 | 43.60 | 57.60 | 2.80 |
| llama-65b-turbomind | 41.60 | 79.20 | 74.40 | 48.40 | 39.20 | 91.20 | 40.40 | 67.20 | 20.00 |
| llama-2-7b-turbomind | 17.20 | 54.80 | 51.60 | 32.80 | 23.60 | 74.40 | 19.60 | 45.60 | 1.20 |
| llama-2-13b-turbomind | 23.20 | 63.60 | 52.40 | 46.00 | 42.00 | 68.00 | 21.60 | 62.00 | 2.00 |
| llama-2-70b-turbomind | 72.40 | 86.40 | 84.40 | 55.20 | 43.20 | 95.60 | 50.80 | 76.80 | 20.80 |
| llama-3-8b-turbomind | 40.80 | 76.40 | 93.20 | 45.20 | 36.80 | 88.80 | 53.60 | 72.80 | 30.80 |
| llama-3-70b-turbomind | 99.20 | 94.00 | 98.00 | 58.40 | 42.80 | 93.60 | 63.60 | 88.40 | 79.20 |
| internlm2-1.8b-turbomind | 16.80 | 47.60 | 63.60 | 21.60 | 12.00 | 69.20 | 16.80 | 45.20 | 5.60 |
| internlm2-7b-turbomind | 51.20 | 78.80 | 90.40 | 52.00 | 41.20 | 95.60 | 58.80 | 74.40 | 44.40 |
| internlm2-20b-turbomind | 81.20 | 95.60 | 83.60 | 62.40 | 48.00 | 94.80 | 57.60 | 75.60 | 72.80 |
| qwen-1.8b-turbomind | 14.80 | 35.60 | 51.20 | 22.40 | 15.20 | 31.20 | 12.40 | 22.00 | 3.20 |
| qwen-7b-turbomind | 20.80 | 54.80 | 76.00 | 37.60 | 27.60 | 74.80 | 41.20 | 57.60 | 23.60 |
| qwen-14b-turbomind | 35.60 | 81.20 | 78.40 | 45.20 | 40.80 | 80.00 | 44.80 | 70.40 | 65.60 |
| qwen-72b-turbomind | 66.40 | 89.20 | 90.40 | 60.00 | 50.80 | 81.60 | 56.40 | 88.00 | 70.40 |
| qwen1.5-0.5b-hf | 20.00 | 34.80 | 46.80 | 18.80 | 15.60 | 24.40 | 15.20 | 16.00 | 1.20 |
| qwen1.5-1.8b-hf | 18.00 | 32.80 | 66.00 | 18.80 | 11.20 | 24.80 | 13.60 | 27.60 | 4.80 |
| qwen1.5-4b-hf | 18.40 | 56.40 | 56.80 | 30.00 | 20.80 | 40.80 | 46.80 | 44.80 | 41.20 |
| qwen1.5-7b-hf | 32.40 | 58.40 | 67.20 | 36.00 | 28.00 | 62.80 | 49.20 | 60.40 | 48.00 |
| qwen1.5-14b-hf | 7.20 | 78.40 | 75.20 | 41.20 | 27.60 | 74.40 | 46.00 | 81.60 | 8.00 |
| qwen1.5-32b-hf | 71.60 | 88.40 | 97.60 | 58.80 | 46.40 | 68.00 | 51.60 | 88.40 | 66.80 |
| qwen1.5-72b-hf | 61.20 | 88.40 | 96.00 | 60.40 | 49.20 | 86.40 | 34.80 | 86.80 | 53.60 |
| qwen1.5-moe-a2-7b-hf | 22.80 | 49.20 | 68.00 | 28.40 | 22.40 | 58.40 | 40.80 | 42.00 | 33.60 |
| mistral-7b-v0.1-hf | 30.40 | 79.60 | 70.80 | 54.40 | 42.80 | 77.60 | 47.20 | 70.00 | 30.40 |
| mistral-7b-v0.2-hf | 32.80 | 74.00 | 77.60 | 48.00 | 40.40 | 84.00 | 49.20 | 76.00 | 35.20 |
| mixtral-8x7b-v0.1-hf | 66.80 | 86.00 | 94.80 | 50.40 | 40.40 | 86.40 | 53.20 | 82.80 | 60.80 |
| mixtral-8x22b-v0.1-hf | 87.60 | 95.20 | 99.60 | 70.00 | 54.00 | 95.20 | 58.40 | 95.20 | 82.00 |
| yi-6b-hf | 17.20 | 49.20 | 72.40 | 34.40 | 28.00 | 76.80 | 32.40 | 56.80 | 9.20 |
| yi-34b-hf | 67.20 | 85.60 | 79.60 | 49.20 | 39.60 | 86.80 | 56.00 | 81.20 | 33.20 |
| deepseek-7b-base-hf | 17.60 | 51.20 | 72.40 | 28.80 | 20.00 | 78.40 | 28.80 | 46.80 | 1.60 |
| deepseek-67b-base-hf | 82.40 | 90.00 | 78.80 | 60.40 | 44.80 | 88.80 | 56.80 | 86.40 | 38.00 |
| model | navigate | dyck_languages | word_sorting | sports_understanding | boolean_expressions | object_counting | formal_fallacies | causal_judgement | web_of_lies |
|:------------------------:|-----------:|-----------------:|---------------:|-----------------------:|----------------------:|------------------:|-------------------:|-------------------:|--------------:|
| llama-7b-turbomind | 45.20 | 1.60 | 8.40 | 81.60 | 66.00 | 47.20 | 46.00 | 40.64 | 57.20 |
| llama-13b-turbomind | 59.20 | 0.80 | 14.40 | 76.40 | 69.20 | 46.40 | 47.20 | 53.48 | 66.80 |
| llama-30b-turbomind | 64.80 | 2.40 | 17.20 | 93.60 | 78.40 | 71.20 | 43.20 | 55.61 | 98.40 |
| llama-65b-turbomind | 72.40 | 6.80 | 21.60 | 98.80 | 81.60 | 70.00 | 40.80 | 55.61 | 99.60 |
| llama-2-7b-turbomind | 54.40 | 1.20 | 10.80 | 88.80 | 68.40 | 49.20 | 48.40 | 52.41 | 53.20 |
| llama-2-13b-turbomind | 74.40 | 2.80 | 18.80 | 97.60 | 74.40 | 52.80 | 46.40 | 54.55 | 96.00 |
| llama-2-70b-turbomind | 82.40 | 13.60 | 30.40 | 98.40 | 81.60 | 83.20 | 43.60 | 63.64 | 100.00 |
| llama-3-8b-turbomind | 90.00 | 9.20 | 38.80 | 95.20 | 87.60 | 84.80 | 51.20 | 50.27 | 100.00 |
| llama-3-70b-turbomind | 96.80 | 48.40 | 48.80 | 99.60 | 92.40 | 99.60 | 62.40 | 58.29 | 100.00 |
| internlm2-1.8b-turbomind | 64.40 | 0.40 | 3.20 | 66.40 | 54.00 | 50.00 | 49.20 | 48.13 | 46.80 |
| internlm2-7b-turbomind | 78.80 | 2.40 | 35.20 | 95.60 | 85.60 | 75.60 | 48.00 | 63.10 | 92.00 |
| internlm2-20b-turbomind | 88.80 | 15.60 | 36.00 | 96.80 | 88.80 | 76.00 | 50.40 | 56.68 | 100.00 |
| qwen-1.8b-turbomind | 50.00 | 0.00 | 0.80 | 62.80 | 29.20 | 2.40 | 6.00 | 12.83 | 1.60 |
| qwen-7b-turbomind | 62.80 | 1.60 | 18.00 | 81.60 | 75.20 | 68.80 | 50.00 | 63.64 | 66.80 |
| qwen-14b-turbomind | 75.60 | 1.20 | 26.80 | 88.80 | 80.40 | 74.40 | 50.00 | 53.48 | 96.80 |
| qwen-72b-turbomind | 56.00 | 14.40 | 35.20 | 87.60 | 91.60 | 81.60 | 5.60 | 31.55 | 62.40 |
| qwen1.5-0.5b-hf | 25.60 | 0.00 | 0.40 | 41.60 | 51.60 | 16.80 | 4.40 | 1.07 | 20.00 |
| qwen1.5-1.8b-hf | 55.60 | 0.00 | 1.60 | 63.60 | 55.20 | 47.60 | 4.40 | 28.88 | 11.20 |
| qwen1.5-4b-hf | 61.60 | 0.40 | 8.80 | 0.80 | 76.00 | 54.40 | 0.80 | 28.34 | 62.40 |
| qwen1.5-7b-hf | 63.60 | 2.40 | 20.80 | 72.40 | 69.60 | 26.80 | 0.00 | 40.64 | 0.00 |
| qwen1.5-14b-hf | 82.40 | 1.20 | 27.60 | 78.40 | 87.20 | 48.00 | 54.00 | 24.06 | 100.00 |
| qwen1.5-32b-hf | 86.80 | 5.60 | 36.80 | 90.00 | 86.40 | 66.40 | 35.60 | 62.57 | 95.60 |
| qwen1.5-72b-hf | 48.40 | 13.20 | 34.40 | 87.60 | 8.00 | 67.60 | 13.60 | 39.57 | 99.60 |
| qwen1.5-moe-a2-7b-hf | 56.80 | 2.00 | 8.80 | 79.60 | 73.60 | 66.80 | 4.00 | 53.48 | 50.40 |
| mistral-7b-v0.1-hf | 73.60 | 4.00 | 26.40 | 97.20 | 82.00 | 67.60 | 43.20 | 48.66 | 100.00 |
| mistral-7b-v0.2-hf | 72.80 | 4.00 | 30.40 | 97.20 | 81.20 | 66.80 | 46.00 | 52.41 | 100.00 |
| mixtral-8x7b-v0.1-hf | 85.60 | 18.80 | 33.60 | 98.00 | 90.80 | 85.20 | 49.60 | 55.61 | 90.80 |
| mixtral-8x22b-v0.1-hf | 92.80 | 51.60 | 40.00 | 98.40 | 91.60 | 95.60 | 54.80 | 56.15 | 100.00 |
| yi-6b-hf | 66.40 | 1.20 | 16.00 | 92.80 | 59.60 | 53.20 | 53.20 | 52.41 | 65.20 |
| yi-34b-hf | 81.20 | 18.80 | 36.40 | 97.60 | 85.60 | 84.00 | 51.20 | 59.89 | 99.60 |
| deepseek-7b-base-hf | 59.20 | 3.20 | 6.40 | 92.00 | 73.20 | 49.60 | 50.80 | 52.41 | 74.80 |
| deepseek-67b-base-hf | 85.20 | 30.00 | 33.20 | 99.60 | 84.80 | 82.40 | 46.80 | 56.68 | 99.60 |
## Chat Models
| model | bbh |
|:-----------------------------:|------:|
| qwen1.5-0.5b-chat-hf | 24.12 |
| qwen1.5-1.8b-chat-hf | 26.82 |
| qwen1.5-4b-chat-hf | 43.15 |
| qwen1.5-7b-chat-hf | 38.12 |
| qwen1.5-14b-chat-hf | 55.38 |
| qwen1.5-32b-chat-hf | 69.28 |
| qwen1.5-72b-chat-hf | 72.97 |
| qwen1.5-110b-chat-hf | 71.04 |
| internlm2-chat-1.8b-hf | 37.69 |
| internlm2-chat-1.8b-sft-hf | 37.12 |
| internlm2-chat-7b-hf | 57.83 |
| internlm2-chat-7b-sft-hf | 57.19 |
| internlm2-chat-20b-hf | 68.24 |
| internlm2-chat-20b-sft-hf | 69.38 |
| llama-3-8b-instruct-hf | 52.85 |
| llama-3-70b-instruct-hf | 82.42 |
| llama-3-8b-instruct-lmdeploy | 53.54 |
| llama-3-70b-instruct-lmdeploy | 82.58 |
| mistral-7b-instruct-v0.1-hf | 32.88 |
| mistral-7b-instruct-v0.2-hf | 48.84 |
| mixtral-8x7b-instruct-v0.1-hf | 59.64 |
### Details
| model | temporal_sequences | disambiguation_qa | date_understanding | tracking_shuffled_objects_three_objects | penguins_in_a_table | geometric_shapes | snarks | ruin_names | tracking_shuffled_objects_seven_objects |
|:-----------------------------:|---------------------:|--------------------:|---------------------:|------------------------------------------:|----------------------:|-------------------:|---------:|-------------:|------------------------------------------:|
| qwen1.5-0.5b-chat-hf | 25.60 | 42.00 | 20.00 | 31.20 | 15.07 | 14.40 | 46.07 | 24.80 | 13.20 |
| qwen1.5-1.8b-chat-hf | 28.80 | 36.00 | 30.40 | 35.20 | 19.18 | 7.60 | 46.63 | 24.00 | 9.60 |
| qwen1.5-4b-chat-hf | 8.00 | 56.00 | 64.80 | 28.40 | 48.63 | 19.60 | 60.67 | 34.00 | 14.40 |
| qwen1.5-7b-chat-hf | 39.60 | 37.60 | 62.40 | 36.80 | 60.96 | 30.80 | 54.49 | 38.00 | 20.00 |
| qwen1.5-14b-chat-hf | 61.60 | 63.60 | 70.00 | 54.00 | 74.66 | 33.60 | 67.42 | 61.20 | 35.60 |
| qwen1.5-32b-chat-hf | 94.40 | 77.60 | 78.00 | 66.00 | 93.84 | 46.00 | 82.58 | 73.60 | 61.60 |
| qwen1.5-72b-chat-hf | 70.40 | 72.40 | 84.40 | 67.20 | 89.73 | 52.00 | 79.21 | 86.40 | 68.80 |
| qwen1.5-110b-chat-hf | 74.80 | 71.20 | 82.80 | 74.80 | 89.04 | 48.00 | 90.45 | 87.60 | 73.60 |
| internlm2-chat-1.8b-hf | 35.60 | 52.40 | 48.80 | 29.60 | 39.73 | 24.40 | 51.69 | 27.20 | 13.20 |
| internlm2-chat-1.8b-sft-hf | 37.20 | 53.60 | 44.00 | 30.00 | 34.93 | 22.40 | 56.74 | 28.00 | 12.00 |
| internlm2-chat-7b-hf | 72.00 | 66.40 | 73.60 | 65.20 | 60.27 | 50.00 | 62.92 | 52.40 | 44.40 |
| internlm2-chat-7b-sft-hf | 67.20 | 66.80 | 58.00 | 63.20 | 48.63 | 45.60 | 64.04 | 59.60 | 42.80 |
| internlm2-chat-20b-hf | 80.40 | 76.00 | 77.60 | 88.80 | 78.08 | 36.40 | 71.91 | 71.60 | 77.20 |
| internlm2-chat-20b-sft-hf | 80.00 | 70.80 | 78.00 | 87.60 | 82.88 | 41.20 | 76.40 | 72.80 | 71.60 |
| llama-3-8b-instruct-hf | 70.40 | 42.80 | 28.40 | 81.20 | 13.01 | 49.20 | 44.94 | 73.20 | 42.40 |
| llama-3-70b-instruct-hf | 100.00 | 84.00 | 91.60 | 95.60 | 78.08 | 52.40 | 87.08 | 89.60 | 97.60 |
| llama-3-8b-instruct-lmdeploy | 73.20 | 45.60 | 34.00 | 79.60 | 31.51 | 48.40 | 47.75 | 76.80 | 47.60 |
| llama-3-70b-instruct-lmdeploy | 100.00 | 84.00 | 90.00 | 96.80 | 83.56 | 56.00 | 87.08 | 89.20 | 97.20 |
| mistral-7b-instruct-v0.1-hf | 32.00 | 22.40 | 52.40 | 35.20 | 30.82 | 23.20 | 38.76 | 46.00 | 18.40 |
| mistral-7b-instruct-v0.2-hf | 66.00 | 58.40 | 50.40 | 48.40 | 48.63 | 37.20 | 65.73 | 40.40 | 29.20 |
| mixtral-8x7b-instruct-v0.1-hf | 63.20 | 68.40 | 65.20 | 60.00 | 78.08 | 40.40 | 74.16 | 64.00 | 46.00 |
| model | tracking_shuffled_objects_five_objects | logical_deduction_three_objects | hyperbaton | logical_deduction_five_objects | logical_deduction_seven_objects | movie_recommendation | salient_translation_error_detection | reasoning_about_colored_objects | multistep_arithmetic_two |
|:-----------------------------:|-----------------------------------------:|----------------------------------:|-------------:|---------------------------------:|----------------------------------:|-----------------------:|--------------------------------------:|----------------------------------:|---------------------------:|
| qwen1.5-0.5b-chat-hf | 20.40 | 34.40 | 51.60 | 21.20 | 13.20 | 26.00 | 20.80 | 17.20 | 1.20 |
| qwen1.5-1.8b-chat-hf | 18.00 | 34.80 | 48.40 | 21.20 | 16.40 | 34.80 | 24.00 | 28.80 | 4.40 |
| qwen1.5-4b-chat-hf | 19.20 | 56.80 | 65.20 | 36.40 | 35.60 | 51.60 | 40.40 | 55.20 | 29.20 |
| qwen1.5-7b-chat-hf | 31.60 | 58.80 | 53.20 | 35.60 | 27.20 | 56.00 | 44.80 | 62.00 | 50.00 |
| qwen1.5-14b-chat-hf | 43.20 | 75.20 | 52.80 | 52.40 | 50.80 | 76.40 | 48.80 | 83.60 | 65.20 |
| qwen1.5-32b-chat-hf | 68.40 | 84.00 | 81.20 | 57.20 | 46.00 | 78.80 | 54.40 | 86.00 | 86.00 |
| qwen1.5-72b-chat-hf | 76.80 | 94.40 | 85.20 | 62.80 | 54.00 | 78.40 | 63.60 | 86.40 | 82.80 |
| qwen1.5-110b-chat-hf | 79.20 | 91.60 | 88.80 | 61.20 | 50.00 | 82.40 | 59.60 | 88.80 | 78.00 |
| internlm2-chat-1.8b-hf | 20.00 | 48.40 | 56.00 | 24.40 | 26.80 | 65.20 | 18.00 | 39.60 | 7.60 |
| internlm2-chat-1.8b-sft-hf | 18.40 | 48.00 | 51.20 | 20.40 | 25.20 | 63.20 | 22.00 | 38.80 | 6.00 |
| internlm2-chat-7b-hf | 48.40 | 75.20 | 84.80 | 42.00 | 36.80 | 79.60 | 53.20 | 65.60 | 26.40 |
| internlm2-chat-7b-sft-hf | 44.00 | 72.40 | 85.60 | 41.60 | 37.20 | 82.40 | 55.60 | 52.80 | 32.00 |
| internlm2-chat-20b-hf | 88.00 | 88.80 | 88.80 | 52.80 | 50.40 | 85.20 | 56.80 | 79.60 | 40.00 |
| internlm2-chat-20b-sft-hf | 83.20 | 90.00 | 90.40 | 55.60 | 48.80 | 84.40 | 57.60 | 79.20 | 38.40 |
| llama-3-8b-instruct-hf | 49.60 | 85.60 | 76.00 | 54.00 | 29.20 | 57.60 | 46.00 | 44.80 | 52.00 |
| llama-3-70b-instruct-hf | 99.20 | 96.80 | 95.20 | 77.20 | 65.20 | 80.00 | 69.60 | 94.80 | 84.00 |
| llama-3-8b-instruct-lmdeploy | 57.20 | 78.00 | 75.60 | 36.00 | 13.20 | 59.20 | 53.60 | 54.80 | 52.80 |
| llama-3-70b-instruct-lmdeploy | 98.80 | 96.40 | 96.80 | 75.20 | 68.80 | 79.60 | 67.60 | 94.00 | 84.80 |
| mistral-7b-instruct-v0.1-hf | 26.00 | 46.00 | 60.00 | 38.00 | 24.00 | 59.20 | 1.20 | 6.00 | 12.40 |
| mistral-7b-instruct-v0.2-hf | 39.60 | 63.60 | 64.00 | 44.00 | 33.20 | 56.00 | 42.40 | 68.40 | 14.00 |
| mixtral-8x7b-instruct-v0.1-hf | 46.40 | 71.60 | 88.80 | 48.00 | 36.80 | 60.00 | 50.00 | 81.20 | 59.20 |
| model | navigate | dyck_languages | word_sorting | sports_understanding | boolean_expressions | object_counting | formal_fallacies | causal_judgement | web_of_lies |
|:-----------------------------:|-----------:|-----------------:|---------------:|-----------------------:|----------------------:|------------------:|-------------------:|-------------------:|--------------:|
| qwen1.5-0.5b-chat-hf | 45.60 | 0.00 | 1.20 | 17.20 | 50.40 | 16.40 | 11.60 | 42.78 | 27.60 |
| qwen1.5-1.8b-chat-hf | 58.40 | 0.00 | 2.00 | 34.00 | 44.80 | 30.40 | 11.60 | 24.60 | 50.00 |
| qwen1.5-4b-chat-hf | 64.00 | 3.20 | 6.80 | 80.40 | 77.60 | 48.80 | 41.20 | 55.61 | 63.20 |
| qwen1.5-7b-chat-hf | 54.40 | 0.40 | 8.00 | 55.60 | 47.60 | 31.20 | 0.00 | 2.14 | 30.00 |
| qwen1.5-14b-chat-hf | 74.40 | 6.40 | 26.40 | 72.40 | 76.40 | 61.60 | 0.80 | 25.67 | 81.20 |
| qwen1.5-32b-chat-hf | 90.00 | 10.40 | 28.40 | 82.40 | 92.80 | 76.80 | 32.40 | 41.71 | 100.00 |
| qwen1.5-72b-chat-hf | 81.20 | 18.40 | 37.60 | 95.20 | 92.80 | 76.00 | 50.40 | 63.64 | 100.00 |
| qwen1.5-110b-chat-hf | 91.60 | 18.00 | 39.60 | 82.80 | 80.80 | 75.20 | 22.40 | 35.83 | 100.00 |
| internlm2-chat-1.8b-hf | 63.20 | 0.00 | 6.00 | 58.00 | 56.80 | 48.80 | 54.80 | 52.94 | 48.40 |
| internlm2-chat-1.8b-sft-hf | 63.20 | 0.00 | 5.60 | 58.00 | 56.80 | 50.00 | 52.40 | 56.68 | 47.60 |
| internlm2-chat-7b-hf | 73.60 | 3.60 | 18.00 | 55.20 | 83.60 | 62.80 | 50.00 | 58.29 | 97.20 |
| internlm2-chat-7b-sft-hf | 71.60 | 4.40 | 20.00 | 82.00 | 84.00 | 60.00 | 51.60 | 52.94 | 98.00 |
| internlm2-chat-20b-hf | 82.40 | 8.00 | 36.00 | 55.60 | 84.40 | 78.00 | 50.40 | 59.36 | 100.00 |
| internlm2-chat-20b-sft-hf | 81.60 | 10.40 | 36.40 | 89.20 | 82.40 | 80.40 | 48.40 | 55.61 | 100.00 |
| llama-3-8b-instruct-hf | 82.80 | 8.80 | 37.20 | 94.40 | 78.80 | 89.60 | 45.20 | 24.06 | 25.60 |
| llama-3-70b-instruct-hf | 95.20 | 18.80 | 49.20 | 98.00 | 94.00 | 90.00 | 73.20 | 68.98 | 100.00 |
| llama-3-8b-instruct-lmdeploy | 83.60 | 10.00 | 40.40 | 96.00 | 77.20 | 89.20 | 43.60 | 37.43 | 3.20 |
| llama-3-70b-instruct-lmdeploy | 95.60 | 22.40 | 48.80 | 96.80 | 91.60 | 87.20 | 72.00 | 69.52 | 100.00 |
| mistral-7b-instruct-v0.1-hf | 70.80 | 0.80 | 5.20 | 68.80 | 69.60 | 51.60 | 3.20 | 12.30 | 33.60 |
| mistral-7b-instruct-v0.2-hf | 62.40 | 4.00 | 15.60 | 81.20 | 70.40 | 50.40 | 32.00 | 34.76 | 98.40 |
| mixtral-8x7b-instruct-v0.1-hf | 76.40 | 12.80 | 23.20 | 55.20 | 85.60 | 83.60 | 40.00 | 43.32 | 88.80 |
opencompass/configs/datasets/bbh/bbh_0shot_nocot_academic_gen.py 0 → 100644
View file @ be3dfa50
# flake8: noqa
import os
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.evaluator import GenericLLMEvaluator
from opencompass.datasets import BBHDataset
from opencompass.datasets.generic import generic_llmjudge_academic_postprocess
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_multiple_choice_sets = [
'temporal_sequences',
'disambiguation_qa',
'date_understanding',
'tracking_shuffled_objects_three_objects',
'penguins_in_a_table',
'geometric_shapes',
'snarks',
'ruin_names',
'tracking_shuffled_objects_seven_objects',
'tracking_shuffled_objects_five_objects',
'logical_deduction_three_objects',
'hyperbaton',
'logical_deduction_five_objects',
'logical_deduction_seven_objects',
'movie_recommendation',
'salient_translation_error_detection',
'reasoning_about_colored_objects',
]
bbh_free_form_sets = [
'multistep_arithmetic_two',
'navigate',
'dyck_languages',
'word_sorting',
'sports_understanding',
'boolean_expressions',
'object_counting',
'formal_fallacies',
'causal_judgement',
'web_of_lies',
]
GRADER_TEMPLATE = """
Please as a grading expert, judge whether the final answers given by the candidates below are consistent with the standard answers, that is, whether the candidates answered correctly.
Here are some evaluation criteria:
1. Please refer to the given standard answer. You don't need to re-generate the answer to the question because the standard answer has been given. You only need to judge whether the candidate's answer is consistent with the standard answer according to the form of the question. Don't try to answer the original question. You can assume that the standard answer is definitely correct.
2. Because the candidate's answer may be different from the standard answer in the form of expression, before making a judgment, please understand the question and the standard answer first, and then judge whether the candidate's answer is correct, but be careful not to try to answer the original question.
3. Some answers may contain multiple items, such as multiple-choice questions, multiple-select questions, fill-in-the-blank questions, etc. As long as the answer is the same as the standard answer, it is enough. For multiple-select questions and multiple-blank fill-in-the-blank questions, the candidate needs to answer all the corresponding options or blanks correctly to be considered correct.
4. Some answers may be expressed in different ways, such as some answers may be a mathematical expression, some answers may be a textual description, as long as the meaning expressed is the same. And some formulas are expressed in different ways, but they are equivalent and correct.
5. If the prediction is given with \\boxed{}, please ignore the \\boxed{} and only judge whether the candidate's answer is consistent with the standard answer.
Please judge whether the following answers are consistent with the standard answer based on the above criteria. Grade the predicted answer of this new question as one of:
A: CORRECT
B: INCORRECT
Just return the letters "A" or "B", with no text around it.
Here is your task. Simply reply with either CORRECT, INCORRECT. Don't apologize or correct yourself if there was a mistake; we are just trying to grade the answer.
<Original Question Begin>: \n{input}\n<Original Question End>\n\n
<Gold Target Begin>: \n{target}\n<Gold Target End>\n\n
<Predicted Answer Begin>: \n{prediction}\n<Predicted End>\n\n
Judging the correctness of candidates' answers:
""".strip()
bbh_sets = bbh_multiple_choice_sets + bbh_free_form_sets
# For zero shot inference in bbh
bbh_datasets = []
for _name in bbh_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=f"Question: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(
type=GenericLLMEvaluator,
prompt_template=dict(
type=PromptTemplate,
template=dict(
begin=[
dict(
role='SYSTEM',
fallback_role='HUMAN',
prompt="You are a helpful assistant who evaluates the correctness and quality of models' outputs.")
],
round=[
dict(
role='HUMAN',
prompt=GRADER_TEMPLATE
),
]),
),
dataset_cfg=dict(
type=BBHDataset,
name=_name,
path='opencompass/bbh',
reader_cfg=bbh_reader_cfg,
),
judge_cfg=dict(),
dict_postprocessor=dict(type=generic_llmjudge_academic_postprocess, metric_name='score'),
),
pred_role='BOT',
)
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy())
)
# For original 3 shot inference in bbh
bbh_3_shot_datasets = []
for _name in bbh_sets:
with open(os.path.join(os.path.dirname(__file__), 'lib_prompt', f'{_name}.txt'), 'r') as f:
_hint = f.read()
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=f"Follow the given examples and answer the question.\n{_hint}\n\nQ: {{input}}\nA: Let's think step by step."
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(
type=GenericLLMEvaluator,
prompt_template=dict(
type=PromptTemplate,
template=dict(
begin=[
dict(
role='SYSTEM',
fallback_role='HUMAN',
prompt="You are a helpful assistant who evaluates the correctness and quality of models' outputs.")
],
round=[
dict(
role='HUMAN',
prompt=GRADER_TEMPLATE
),
]),
),
dataset_cfg=dict(
type=BBHDataset,
name=_name,
path='opencompass/bbh',
reader_cfg=bbh_reader_cfg,
),
judge_cfg=dict(),
dict_postprocessor=dict(type=generic_llmjudge_academic_postprocess, metric_name='score'),
),
pred_role='BOT',
)
bbh_3_shot_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
opencompass/configs/datasets/bbh/bbh_0shot_nocot_gen_925fc4.py 0 → 100644
View file @ be3dfa50
import os
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.openicl.icl_evaluator import AccEvaluator
from opencompass.datasets import BBHDataset, BBHEvaluator, bbh_mcq_postprocess, BBHEvaluator_mcq
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_multiple_choice_sets = [
'temporal_sequences',
'disambiguation_qa',
'date_understanding',
'tracking_shuffled_objects_three_objects',
'penguins_in_a_table',
'geometric_shapes',
'snarks',
'ruin_names',
'tracking_shuffled_objects_seven_objects',
'tracking_shuffled_objects_five_objects',
'logical_deduction_three_objects',
'hyperbaton',
'logical_deduction_five_objects',
'logical_deduction_seven_objects',
'movie_recommendation',
'salient_translation_error_detection',
'reasoning_about_colored_objects',
]
bbh_free_form_sets = [
'multistep_arithmetic_two',
'navigate',
'dyck_languages',
'word_sorting',
'sports_understanding',
'boolean_expressions',
'object_counting',
'formal_fallacies',
'causal_judgement',
'web_of_lies',
]
bbh_datasets = []
for _name in bbh_multiple_choice_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Question: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator_mcq),
pred_role='BOT',
pred_postprocessor=dict(type=bbh_mcq_postprocess),
dataset_postprocessor=dict(type=bbh_mcq_postprocess))
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
for _name in bbh_free_form_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Question: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(evaluator=dict(type=BBHEvaluator), pred_role='BOT')
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
opencompass/configs/datasets/bbh/bbh_0shot_nocot_gen_9c32f6.py 0 → 100644
View file @ be3dfa50
import os
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.openicl.icl_evaluator import AccEvaluator
from opencompass.datasets import BBHDataset, BBHEvaluator, bbh_mcq_postprocess, BBHEvaluator_mcq
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_multiple_choice_sets = [
'temporal_sequences',
'disambiguation_qa',
'date_understanding',
'tracking_shuffled_objects_three_objects',
'penguins_in_a_table',
'geometric_shapes',
'snarks',
'ruin_names',
'tracking_shuffled_objects_seven_objects',
'tracking_shuffled_objects_five_objects',
'logical_deduction_three_objects',
'hyperbaton',
'logical_deduction_five_objects',
'logical_deduction_seven_objects',
'movie_recommendation',
'salient_translation_error_detection',
'reasoning_about_colored_objects',
]
bbh_free_form_sets = [
'multistep_arithmetic_two',
'navigate',
'dyck_languages',
'word_sorting',
'sports_understanding',
'boolean_expressions',
'object_counting',
'formal_fallacies',
'causal_judgement',
'web_of_lies',
]
bbh_datasets = []
for _name in bbh_multiple_choice_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Follow the given examples and answer the question.\n\nQuestion: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator_mcq),
pred_role='BOT',
pred_postprocessor=dict(type=bbh_mcq_postprocess),
dataset_postprocessor=dict(type=bbh_mcq_postprocess))
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
for _name in bbh_free_form_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Follow the given examples and answer the question.\n\nQuestion: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(evaluator=dict(type=BBHEvaluator), pred_role='BOT')
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
opencompass/configs/datasets/bbh/bbh_0shot_nocot_gen_ea7952.py 0 → 100644
View file @ be3dfa50
import os
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.openicl.icl_evaluator import AccEvaluator
from opencompass.datasets import BBHDataset, BBHEvaluator, bbh_mcq_postprocess, BBHEvaluator_mcq
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_multiple_choice_sets = [
'temporal_sequences',
'disambiguation_qa',
'date_understanding',
'tracking_shuffled_objects_three_objects',
'penguins_in_a_table',
'geometric_shapes',
'snarks',
'ruin_names',
'tracking_shuffled_objects_seven_objects',
'tracking_shuffled_objects_five_objects',
'logical_deduction_three_objects',
'hyperbaton',
'logical_deduction_five_objects',
'logical_deduction_seven_objects',
'movie_recommendation',
'salient_translation_error_detection',
'reasoning_about_colored_objects',
]
bbh_free_form_sets = [
'multistep_arithmetic_two',
'navigate',
'dyck_languages',
'word_sorting',
'sports_understanding',
'boolean_expressions',
'object_counting',
'formal_fallacies',
'causal_judgement',
'web_of_lies',
]
bbh_datasets = []
for _name in bbh_multiple_choice_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Question: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator_mcq),
pred_role='BOT',
pred_postprocessor=dict(type=bbh_mcq_postprocess),
dataset_postprocessor=dict(type=bbh_mcq_postprocess))
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
for _name in bbh_free_form_sets:
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Follow the given examples and answer the question.\n\nQuestion: {{input}}\n You must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(evaluator=dict(type=BBHEvaluator), pred_role='BOT')
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
opencompass/configs/datasets/bbh/bbh_gen.py 0 → 100644
View file @ be3dfa50
from mmengine.config import read_base
with read_base():
from .bbh_gen_5b92b0 import bbh_datasets # noqa: F401, F403
opencompass/configs/datasets/bbh/bbh_gen_2879b0.py 0 → 100644
View file @ be3dfa50
import os
from mmengine.config import read_base
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.datasets import BBHDataset, bbh_mcq_postprocess, BBHEvaluator, BBHEvaluator_mcq
with read_base():
from .bbh_subset_settings import settings
bbh_datasets = []
for name, test_type in settings:
with open(os.path.join(os.path.dirname(__file__), 'lib_prompt', f'{name}.txt'), 'r') as f:
hint = f.read()
task_prompt, body = hint.split('\n\nQ:', 1)
sections = ('Q:' + body).split('\n\n')
prompt_rounds = []
for index, section in enumerate(sections):
question, answer = section.split('\nA:')
answer = 'A:' + answer
if index == 0:
desc = task_prompt.strip() + '\n'
else:
desc = ''
prompt_rounds.append(dict(role='HUMAN', prompt=f'{desc}{question.strip()}'))
prompt_rounds.append(dict(role='BOT', prompt=answer.strip()))
prompt_rounds.append(dict(role='HUMAN', prompt='Q: {input}'))
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_infer_cfg = dict(
prompt_template=dict(type=PromptTemplate, template=dict(round=prompt_rounds)),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
if test_type == 'mcq':
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator_mcq),
pred_role='BOT',
pred_postprocessor=dict(type=bbh_mcq_postprocess),
dataset_postprocessor=dict(type=bbh_mcq_postprocess))
else:
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator),
pred_role='BOT')
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=name,
abbr='bbh-' + name,
reader_cfg=bbh_reader_cfg.copy(),
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
opencompass/configs/datasets/bbh/bbh_gen_4a31fa.py 0 → 100644
View file @ be3dfa50
import os
from opencompass.openicl.icl_prompt_template import PromptTemplate
from opencompass.openicl.icl_retriever import ZeroRetriever
from opencompass.openicl.icl_inferencer import GenInferencer
from opencompass.openicl.icl_evaluator import AccEvaluator
from opencompass.datasets import BBHDataset, BBHEvaluator, bbh_mcq_postprocess, BBHEvaluator_mcq
bbh_reader_cfg = dict(input_columns=['input'], output_column='target')
bbh_multiple_choice_sets = [
'temporal_sequences',
'disambiguation_qa',
'date_understanding',
'tracking_shuffled_objects_three_objects',
'penguins_in_a_table',
'geometric_shapes',
'snarks',
'ruin_names',
'tracking_shuffled_objects_seven_objects',
'tracking_shuffled_objects_five_objects',
'logical_deduction_three_objects',
'hyperbaton',
'logical_deduction_five_objects',
'logical_deduction_seven_objects',
'movie_recommendation',
'salient_translation_error_detection',
'reasoning_about_colored_objects',
]
bbh_free_form_sets = [
'multistep_arithmetic_two',
'navigate',
'dyck_languages',
'word_sorting',
'sports_understanding',
'boolean_expressions',
'object_counting',
'formal_fallacies',
'causal_judgement',
'web_of_lies',
]
bbh_datasets = []
for _name in bbh_multiple_choice_sets:
with open(os.path.join(os.path.dirname(__file__), 'lib_prompt', f'{_name}.txt'), 'r') as f:
_hint = f.read()
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Follow the given examples and answer the question.\n{_hint}\n\nQ: {{input}}\nA: Let's think step by step. And you must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(
evaluator=dict(type=BBHEvaluator_mcq),
pred_role='BOT',
pred_postprocessor=dict(type=bbh_mcq_postprocess),
dataset_postprocessor=dict(type=bbh_mcq_postprocess))
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
for _name in bbh_free_form_sets:
with open(os.path.join(os.path.dirname(__file__), 'lib_prompt', f'{_name}.txt'), 'r') as f:
_hint = f.read()
bbh_infer_cfg = dict(
prompt_template=dict(
type=PromptTemplate,
template=dict(round=[
dict(
role='HUMAN',
prompt=
f"Follow the given examples and answer the question.\n{_hint}\n\nQ: {{input}}\nA: Let's think step by step. And you must give your final answer by starting with 'So the answer is' "
)
])),
retriever=dict(type=ZeroRetriever),
inferencer=dict(type=GenInferencer, max_out_len=512))
bbh_eval_cfg = dict(evaluator=dict(type=BBHEvaluator), pred_role='BOT')
bbh_datasets.append(
dict(
type=BBHDataset,
path='opencompass/bbh',
name=_name,
abbr='bbh-' + _name,
reader_cfg=bbh_reader_cfg,
infer_cfg=bbh_infer_cfg.copy(),
eval_cfg=bbh_eval_cfg.copy()))
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment