Skip to content

Commit bb15146

Browse files
Myhs-phzMyhs-phz
authored
[Update] Add evaluation example for Intern-S1-Pro (#2394)
1 parent 29ca947 commit bb15146

File tree

3 files changed

+239
-0
lines changed

3 files changed

+239
-0
lines changed

‎README.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -57,6 +57,7 @@ Just like a compass guides us on our journey, OpenCompass will guide you through
5757
5858
## 🚀 What's New <a><img width="35" height="20" src="https://user-images.githubusercontent.com/12782558/212848161-5e783dd6-11e8-4fe0-bbba-39ffb77730be.png"></a>
5959

60+
- **\[2026.02.05\]** OpenCompass now supports Intern-S1-Pro related general and scientific evaluation benchmarks. Please check [Example for Evaluating Intern-S1](examples/eval_intern_s1_pro.py) and [Model Card](https://huggingface.co/internlm/Intern-S1-Pro)for more details! 🔥🔥🔥
6061
- **\[2025.07.26\]** OpenCompass now supports Intern-S1 related general and scientific evaluation benchmarks. Please check [Tutorial for Evaluating Intern-S1](https://opencompass.readthedocs.io/en/latest/user_guides/interns1.html) for more details! 🔥🔥🔥
6162
- **\[2025.04.01\]** OpenCompass now supports `CascadeEvaluator`, a flexible evaluation mechanism that allows multiple evaluators to work in sequence. This enables creating customized evaluation pipelines for complex assessment scenarios. Check out the [documentation](docs/en/advanced_guides/llm_judge.md) for more details! 🔥🔥🔥
6263
- **\[2025.03.11\]** We have supported evaluation for `SuperGPQA` which is a great benchmark for measuring LLM knowledge ability 🔥🔥🔥

‎README_zh-CN.md‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -57,6 +57,7 @@
5757
5858
## 🚀 最新进展 <a><img width="35" height="20" src="https://user-images.githubusercontent.com/12782558/212848161-5e783dd6-11e8-4fe0-bbba-39ffb77730be.png"></a>
5959

60+
- **\[2026.02.05\]** OpenCompass 现已支持Intern-S1-Pro相关的通用及科学评测基准，请参阅[Intern-S1-Pro评测示例](examples/eval_intern_s1_pro.py)和[模型信息](https://huggingface.co/internlm/Intern-S1-Pro)了解详情！🔥🔥🔥
6061
- **\[2025.07.26\]** OpenCompass 现已支持Intern-S1相关的通用及科学评测基准，请参阅[Intern-S1评测教程](https://opencompass.readthedocs.io/zh-cn/latest/user_guides/interns1.html)了解详情！🔥🔥🔥
6162
- **\[2025.04.01\]** OpenCompass 现已支持 `CascadeEvaluator`，允许多个评估器按顺序工作，可以为更复杂的评估场景创建自定义评估流程，查看[文档](docs/zh_cn/advanced_guides/llm_judge.md)了解具体用法！🔥🔥🔥
6263
- **\[2025.03.11\]** 现已支持 `SuperGPQA` 覆盖285 个研究生学科的知识能力评测，欢迎尝试！🔥🔥🔥

‎examples/eval_intern_s1_pro.py‎

Lines changed: 237 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,237 @@
1+
# flake8: noqa
2+
3+
from mmengine.config import read_base
4+
5+
from opencompass.partitioners import NaivePartitioner, NumWorkerPartitioner
6+
from opencompass.runners import LocalRunner
7+
from opencompass.tasks import OpenICLEvalTask, OpenICLInferTask
8+
from copy import deepcopy
9+
from opencompass.utils.text_postprocessors import extract_non_reasoning_content
10+
from opencompass.models import OpenAISDKStreaming
11+
12+
#######################################################################
13+
# PART 0 Essential Configs #
14+
#######################################################################
15+
with read_base():
16+
# Datasets
17+
18+
from opencompass.configs.datasets.mmlu_pro.mmlu_pro_0shot_nocot_genericllmeval_gen_08c1de import (
19+
mmlu_pro_datasets,
20+
)
21+
from opencompass.configs.datasets.gpqa.gpqa_cascade_eval_gen_772ea0 import (
22+
gpqa_datasets,
23+
)
24+
from opencompass.configs.datasets.aime2025.aime2025_cascade_eval_gen_5e9f4f import (
25+
aime2025_datasets,
26+
)
27+
from opencompass.configs.chatml_datasets.IMO_Bench_AnswerBench.IMO_Bench_AnswerBench_gen import (
28+
datasets as IMO_Bench_AnswerBench_chatml
29+
)
30+
from opencompass.configs.datasets.IFBench.IFBench_gen import (
31+
ifbench_datasets,
32+
)
33+
from opencompass.configs.datasets.livecodebench.livecodebench_gen_a4f90b import (
34+
LCBCodeGeneration_dataset,
35+
)
36+
from opencompass.configs.datasets.SmolInstruct.smolinstruct_0shot_instruct_gen import (
37+
smolinstruct_datasets_0shot_instruct as smolinstruct_datasets,
38+
)
39+
from opencompass.configs.datasets.matbench.matbench_llm_judge_gen_0e9276 import (
40+
matbench_datasets,
41+
)
42+
from opencompass.configs.datasets.biodata.biodata_task_gen import (
43+
biodata_task_datasets
44+
)
45+
from opencompass.configs.datasets.MolInstructions_chem.mol_instructions_chem_gen import (
46+
mol_gen_selfies_datasets
47+
)
48+
49+
# Summary Groups
50+
from opencompass.configs.summarizers.groups.mmlu_pro import \
51+
mmlu_pro_summary_groups
52+
from opencompass.configs.summarizers.groups.biodata import (
53+
biodata_summary_groups,
54+
)
55+
56+
LCBCodeGeneration_v6_datasets = deepcopy(LCBCodeGeneration_dataset)
57+
LCBCodeGeneration_v6_datasets['abbr'] = 'lcb_code_generation_v6'
58+
LCBCodeGeneration_v6_datasets['release_version'] = 'v6'
59+
LCBCodeGeneration_v6_datasets['eval_cfg']['evaluator'][
60+
'release_version'
61+
] = 'v6'
62+
LCBCodeGeneration_v6_datasets = [LCBCodeGeneration_v6_datasets]
63+
64+
#######################################################################
65+
# PART 1 Datasets List #
66+
#######################################################################
67+
# datasets list for evaluation
68+
69+
repeated_info = [
70+
(gpqa_datasets, 8),
71+
(aime2025_datasets, 32),
72+
]
73+
74+
for datasets_, num in repeated_info:
75+
for dataset_ in datasets_:
76+
dataset_['n'] = num
77+
dataset_['k'] = num
78+
79+
datasets = sum(
80+
(v for k, v in locals().items() if k.endswith('_datasets')),
81+
[],
82+
)
83+
84+
chatml_datasets = sum(
85+
(v for k, v in locals().items() if k.endswith('_chatml')),
86+
[],
87+
)
88+
89+
# LLM judge config: using LLM to evaluate predictions
90+
judge_cfg = dict(
91+
abbr='YOUR_JUDGE_MODEL',
92+
type=OpenAISDKStreaming,
93+
path='YOUR_JUDGE_MODEL',
94+
key='YOUR_JUDGE_KEY',
95+
openai_api_base='YOUR_JUDGE_URL',
96+
mode='mid',
97+
meta_template=dict(
98+
round=[
99+
dict(role='HUMAN', api_role='HUMAN'),
100+
dict(role='BOT', api_role='BOT', generate=True),
101+
]
102+
),
103+
query_per_second=16,
104+
batch_size=64,
105+
temperature=0.001,
106+
max_out_len=8192,
107+
max_seq_len=32768,
108+
)
109+
110+
for item in datasets:
111+
if 'judge_cfg' in item['eval_cfg']['evaluator']:
112+
item['eval_cfg']['evaluator']['judge_cfg'] = judge_cfg
113+
if 'llm_evaluator' in item['eval_cfg']['evaluator'].keys() and 'judge_cfg' in item['eval_cfg']['evaluator']['llm_evaluator']:
114+
item['eval_cfg']['evaluator']['llm_evaluator']['judge_cfg'] = judge_cfg
115+
116+
for item in chatml_datasets:
117+
if item['evaluator']['type'] == 'llm_evaluator':
118+
item['evaluator']['judge_cfg'] = judge_cfg
119+
if item['evaluator']['type'] == 'cascade_evaluator':
120+
item['evaluator']['llm_evaluator']['judge_cfg'] = judge_cfg
121+
122+
123+
#######################################################################
124+
# PART 2 Datset Summarizer #
125+
#######################################################################
126+
127+
summary_groups = sum(
128+
[v for k, v in locals().items() if k.endswith('_summary_groups')], []
129+
)
130+
131+
summarizer = dict(
132+
dataset_abbrs=[
133+
['mmlu_pro', 'accuracy'],
134+
['IFBench', 'score'],
135+
['GPQA_diamond', 'accuracy (8 runs average)'],
136+
['aime2025', 'accuracy (32 runs average)'],
137+
['lcb_code_generation_v6', 'pass@1'],
138+
['bio_data', 'naive_average'],
139+
['IMO-Bench-AnswerBench', 'accuracy'],
140+
'',
141+
'Mol_Instruct',
142+
['FS-selfies', 'score'],
143+
['MC-selfies', 'score'],
144+
['MG-selfies', 'score'],
145+
['PP-selfies', 'score'],
146+
['RP-selfies', 'score'],
147+
['RS-selfies', 'score'],
148+
'',
149+
'SmolInstruct',
150+
['NC-I2F-0shot-instruct', 'score'],
151+
['NC-I2S-0shot-instruct', 'score'],
152+
['NC-S2F-0shot-instruct', 'score'],
153+
['NC-S2I-0shot-instruct', 'score'],
154+
['PP-ESOL-0shot-instruct', 'score'],
155+
['PP-Lipo-0shot-instruct', 'score'],
156+
['PP-BBBP-0shot-instruct', 'accuracy'],
157+
['PP-ClinTox-0shot-instruct', 'accuracy'],
158+
['PP-HIV-0shot-instruct', 'accuracy'],
159+
['PP-SIDER-0shot-instruct', 'accuracy'],
160+
['MC-0shot-instruct', 'score'],
161+
['MG-0shot-instruct', 'score'],
162+
['FS-0shot-instruct', 'score'],
163+
['RS-0shot-instruct', 'score'],
164+
'',
165+
['matbench_expt_gap', 'mae'],
166+
['matbench_steels', 'mae'],
167+
['matbench_expt_is_metal', 'accuracy'],
168+
['matbench_glass', 'accuracy'],
169+
],
170+
summary_groups=summary_groups,
171+
)
172+
173+
174+
#######################################################################
175+
# PART 3 Models #
176+
#######################################################################
177+
178+
api_meta_template = dict(
179+
round=[
180+
dict(role='SYSTEM', api_role='SYSTEM'), # System prompt is only needed when evaluating Bio_data and Mol_instructions
181+
dict(role='HUMAN', api_role='HUMAN'),
182+
dict(role='BOT', api_role='BOT', generate=True),
183+
]
184+
)
185+
186+
models = [
187+
dict(
188+
abbr='intern-s1-pro',
189+
type=OpenAISDKStreaming,
190+
path='intern-s1-pro',
191+
key='YOUR_API_KEY',
192+
openai_api_base='YOUR_API_BASE',
193+
meta_template=api_meta_template,
194+
query_per_second=16,
195+
batch_size=8,
196+
temperature=0.8,
197+
retry=10,
198+
max_out_len=65536,
199+
max_seq_len=65536,
200+
extra_body={
201+
'chat_template_kwargs': {'enable_thinking': True} # Disable thinking when evaluating scientific benchmarks
202+
},
203+
pred_postprocessor=dict(
204+
type=extract_non_reasoning_content,
205+
),
206+
),
207+
]
208+
209+
#######################################################################
210+
# PART 4 Inference/Evaluation Configuaration #
211+
#######################################################################
212+
213+
# infer with local runner
214+
infer = dict(
215+
partitioner=dict(type=NumWorkerPartitioner, num_worker=8),
216+
runner=dict(
217+
type=LocalRunner,
218+
max_num_workers=16,
219+
task=dict(type=OpenICLInferTask),
220+
),
221+
)
222+
223+
# eval with local runner
224+
eval = dict(
225+
partitioner=dict(type=NaivePartitioner, n=10),
226+
runner=dict(
227+
type=LocalRunner,
228+
max_num_workers=16,
229+
task=dict(type=OpenICLEvalTask)
230+
),
231+
)
232+
233+
#######################################################################
234+
# PART 5 Utils Configuaration #
235+
#######################################################################
236+
237+
work_dir = './outputs/oc_intern_s1_pro_eval'

0 commit comments

Comments
 (0)