[Feature] add one script for subjective (#993)

* add one script for subjective

* add one script for subjective

* add one script for subjective

* add one script for subjective

---------

Co-authored-by: thebestannie <1290646445@qq.com>

configs/subjective/eval_subjective_alignbench.py

@@ -0,0 +1,21 @@
+from mmengine.config import read_base
+
+with read_base():
+    from ..datasets.subjective.alignbench.alignbench_judgeby_critiquellm import subjective_datasets
+    from .model_cfg import models, judge_model, given_pred, infer, gpt4, runner
+from opencompass.partitioners.sub_naive import SubjectiveNaivePartitioner
+from opencompass.partitioners.sub_size import SubjectiveSizePartitioner
+from opencompass.summarizers import AlignmentBenchSummarizer
+
+# -------------Inference Stage ----------------------------------------
+# For subjective evaluation, we often set do sample for models
+datasets = [*subjective_datasets]
+eval = dict(
+    partitioner=dict(
+        type=SubjectiveNaivePartitioner, mode='singlescore', models=models
+    ),
+    runner=runner,
+)
+
+summarizer = dict(type=AlignmentBenchSummarizer, judge_type='general')
+work_dir = 'outputs/alignment_bench/'

configs/subjective/eval_subjective_alpacaeval.py

@@ -0,0 +1,20 @@
+from mmengine.config import read_base
+
+with read_base():
+    from ..datasets.subjective.alpaca_eval.alpacav1_judgeby_gpt4 import subjective_datasets as alpacav1
+    from ..datasets.subjective.alpaca_eval.alpacav2_judgeby_gpt4 import subjective_datasets as alpacav2
+    from .model_cfg import models, judge_model, given_pred, infer, gpt4, runner
+from opencompass.partitioners.sub_naive import SubjectiveNaivePartitioner
+from opencompass.partitioners.sub_size import SubjectiveSizePartitioner
+from opencompass.summarizers import AlpacaSummarizer
+datasets = [*alpacav2]
+eval = dict(
+    partitioner=dict(
+        type=SubjectiveSizePartitioner, max_task_size=1000, mode='m2n', base_models=[gpt4], compare_models=models
+    ),
+runner=runner,
+given_pred=given_pred
+)
+work_dir = 'outputs/alpaca/'
+
+summarizer = dict(type=AlpacaSummarizer, judge_type='v2')

configs/subjective/eval_subjective_compassarena.py

@@ -0,0 +1,28 @@
+from os import getenv as gv
+from opencompass.models import HuggingFaceCausalLM
+from mmengine.config import read_base
+
+with read_base():
+    from ..datasets.subjective.compassarena.compassarena_compare import subjective_datasets
+    from .model_cfg import models, judge_model, given_pred, infer, gpt4, runner
+from opencompass.partitioners.sub_naive import SubjectiveNaivePartitioner
+from opencompass.partitioners.sub_size import SubjectiveSizePartitioner
+from opencompass.summarizers import CompassArenaSummarizer
+datasets = [*subjective_datasets]
+
+eval = dict(
+    partitioner=dict(
+        type=SubjectiveSizePartitioner,
+        strategy='split',
+        max_task_size=10000,
+        mode='m2n',
+        base_models=[gpt4],
+        compare_models=models,
+    ),
+runner=runner,
+given_pred=given_pred
+)
+
+work_dir = 'outputs/compass_arena/'
+
+summarizer = dict(type=CompassArenaSummarizer, summary_type='half_add')

configs/subjective/eval_subjective_mtbench.py

@@ -0,0 +1,25 @@
+from mmengine.config import read_base
+
+with read_base():
+    from ..datasets.subjective.multiround.mtbench_single_judge_diff_temp import subjective_datasets
+    # from .datasets.subjective.multiround.mtbench_pair_judge import subjective_datasets
+    from .model_cfg import models, judge_model, given_pred, infer, gpt4, runner
+from opencompass.partitioners.sub_naive import SubjectiveNaivePartitioner
+from opencompass.partitioners.sub_size import SubjectiveSizePartitioner
+from opencompass.summarizers import MTBenchSummarizer
+
+datasets = [*subjective_datasets]
+
+for model in models:
+    if 'generation_kwargs' in model:
+        if 'do_sample' in model['generation_kwargs']:
+            del model['generation_kwargs']['do_sample']
+            
+eval = dict(
+    partitioner=dict(type=SubjectiveSizePartitioner, strategy='split', max_task_size=10000, mode='singlescore', models=models),
+    runner=runner
+)
+
+summarizer = dict(type=MTBenchSummarizer, judge_type='single')
+
+work_dir = 'outputs/mtbench/'

configs/subjective/model_cfg.py

@@ -0,0 +1,85 @@
+from opencompass.models import HuggingFaceCausalLM, HuggingFace, HuggingFaceChatGLM3
+from opencompass.models.openai_api import OpenAIAllesAPIN
+from opencompass.partitioners import NaivePartitioner, SizePartitioner
+from opencompass.partitioners.sub_naive import SubjectiveNaivePartitioner
+from opencompass.partitioners.sub_size import SubjectiveSizePartitioner
+from opencompass.runners import LocalRunner
+from opencompass.runners import SlurmSequentialRunner
+from opencompass.tasks import OpenICLInferTask
+from opencompass.tasks.subjective_eval import SubjectiveEvalTask
+api_meta_template = dict(
+    round=[
+        dict(role='HUMAN', api_role='HUMAN'),
+        dict(role='BOT', api_role='BOT', generate=True),
+    ]
+)
+# -------------Inference Stage ----------------------------------------
+# For subjective evaluation, we often set do sample for models
+models = [
+    dict(
+        type=HuggingFaceChatGLM3,
+        abbr='chatglm3-6b-hf',
+        path='THUDM/chatglm3-6b',
+        tokenizer_path='THUDM/chatglm3-6b',
+        model_kwargs=dict(
+            device_map='auto',
+            trust_remote_code=True,
+        ),
+        tokenizer_kwargs=dict(
+            padding_side='left',
+            truncation_side='left',
+            trust_remote_code=True,
+        ),
+        generation_kwargs=dict(
+            do_sample=True,
+        ),
+        meta_template=api_meta_template,
+        max_out_len=2048,
+        max_seq_len=4096,
+        batch_size=1,
+        run_cfg=dict(num_gpus=1, num_procs=1),
+    )
+]
+
+
+judge_model = dict(
+        abbr='GPT4-Turbo',
+        type=OpenAIAllesAPIN, path='gpt-4-1106-preview',
+        key='',  # The key will be obtained from $OPENAI_API_KEY, but you can write down your key here as well
+        url='',
+        meta_template=api_meta_template,
+        query_per_second=1,
+        max_out_len=1024,
+        max_seq_len=4096,
+        batch_size=1,
+        retry=30,
+        temperature = 0
+)
+
+infer = dict(
+    partitioner=dict(type=SizePartitioner, strategy='split', max_task_size=10000),
+    runner=dict(
+        type=SlurmSequentialRunner,
+        partition='llmeval',
+        quotatype='auto',
+        max_num_workers=256,
+        task=dict(type=OpenICLInferTask),
+    ),
+)
+runner=dict(type=LocalRunner, max_num_workers=12, task=dict(type=SubjectiveEvalTask, judge_cfg=judge_model))
+
+gpt4 = dict(
+    abbr='gpt4-turbo',
+    type=OpenAIAllesAPIN,
+    path='gpt-4-1106-preview',
+    key='',  # The key will be obtained from $OPENAI_API_KEY, but you can write down your key here as well
+    meta_template=api_meta_template,
+    query_per_second=1,
+    max_out_len=2048,
+    max_seq_len=4096,
+    batch_size=4,
+    retry=20,
+    temperature=1,
+) 
+given_pred = [{'abbr':'gpt4-turbo', 'path':'your path'}]
+