SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2502.09082 · ICML · 2025

CoSER: A Comprehensive Literary Dataset and Framework for Training and Evaluating LLM Role-Playing and Persona Simulation

Wei Wang, Rui Xu, Heng Wang, Yanghua Xiao, Shuchang Zhou, Xintao Wang, Jiangjie Chen, Jen-Tse Huang, Siyu Yuan, Yifei Zhang, Xinfeng Yuan, Haoran Guo, and 1 more

arXiv · PDF · Open in the Atlas

Code that ran

9 of 23 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 4 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 5 failures of our instrument; 14 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2502.09082.json · badge: /api/ran/2502.09082.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
neph0s/incharacter community
Papers with Code archive
8 ran · 14 no recorded run
Neph0s/CoSER marked official in Papers with Code
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
base64_to_float_array Ran
ran, honoured the contract we drafted
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("4b18614854f76d47")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64
import struct
import numpy as np

# Create a small array of floats
floats = [3.14, 2.71, 1.41, 0.5]

# Convert floats to bytes
byte_data = b''.join(struct.pack('!f', f) for f in floats)

# Encode to base64
base64_data = base64.b64encode(byte_data).decode('ascii')

ARGS = {
    'base64_data': base64_data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
base64_to_string Ran
ran; our contract draft was wrong, not the code
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("9083059a6053c01a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64

# Create a simple string to encode
test_string = "Hello, World!"

# Encode it to base64
base64_data = base64.b64encode(test_string.encode('utf-8')).decode('utf-8')

# Define arguments for the function
ARGS = {
    'base64_data': base64_data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_bge_embeddings Ran
ran, honoured the contract we drafted
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("8ac80353df541407")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned list, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from unittest.mock import MagicMock

torch.manual_seed(0)

# Create a minimal stand-in tokenizer
class MockTokenizer:
    def __call__(self, sentences, padding=True, truncation=True, return_tensors='pt', max_length=512):
        # Return a dict with 'input_ids' and 'attention_mask' tensors
        batch_size = len(sentences) if isinstance(sentences, list) else 1
        return {
            'input_ids': torch.randint(0, 1000, (batch_size, 10)),
            'attention_mask': torch.ones(batch_size, 10)
        }

# Create a minimal stand-in model
class MockBGEModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 384)
    
    def forward(self, input_ids, attention_mask, **kwargs):
        # Return a tuple where first element is embeddings of shape (batch_size, seq_len, hidden_dim)
        batch_size = input_ids.shape[0]
        seq_len = input_ids.shape[1]
        embeddings = self.linear(torch.randn(batch_size, seq_len, 10))
        return (embeddings,)
    
    def eval(self):
        return super().eval()

# Patch the global variables before calling the function
mock_tokenizer = MockTokenizer()
mock_model = MockBGEModel()

# Mock the AutoTokenizer and AutoModel to return our stand-ins
import transformers
original_auto_tokenizer = transformers.AutoTokenizer.from_pretrained
original_auto_model = transformers.AutoModel.from_pretrained

def mock_from_pretrained_tokenizer(model_name):
    return mock_tokenizer

def mock_from_pretrained_model(model_name):
    return mock_model

transformers.AutoTokenizer.from_pretrained = mock_from_pretrained_tokenizer
transformers.AutoModel.from_pretrained = mock_from_pretrained_model

# Define the arguments
ARGS = {
    'sentences': ['This is a test sentence.', 'This is another test sentence.']
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_bge_zh_embedding Ran
ran, honoured the contract we drafted
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("6087c37cde8b7d71")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from unittest.mock import Mock, MagicMock

torch.manual_seed(0)

# Create a minimal mock tokenizer
class MockTokenizer:
    def __call__(self, texts, padding=True, truncation=True, return_tensors='pt', max_length=512):
        if isinstance(texts, str):
            texts = [texts]
        batch_size = len(texts)
        # Return mock tokenizer output with input_ids and attention_mask
        return {
            'input_ids': torch.zeros((batch_size, 10), dtype=torch.long),
            'attention_mask': torch.ones((batch_size, 10), dtype=torch.long)
        }

# Create a minimal mock model
class MockBGEModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 384)
    
    def forward(self, input_ids, attention_mask):
        # Return a tuple where first element is embeddings of shape (batch_size, seq_len, 384)
        batch_size = input_ids.shape[0]
        embeddings = self.linear(torch.randn(batch_size, 10, 10))
        return (embeddings,)
    
    def eval(self):
        return super().eval()

# Mock the AutoTokenizer and AutoModel to return our mocks
mock_tokenizer = MockTokenizer()
mock_model = MockBGEModel()

# Patch the global pools before calling the function
import sys
from unittest.mock import patch

# We need to set up the model pool and tokenizer pool
# Since the function uses globals, we'll prepare them
_model_pool = {'BAAI/bge-small-zh-v1.5': mock_model}
_tokenizer_pool = {'BAAI/bge-small-zh-v1.5': mock_tokenizer}

# Inject into the module
import transformers
original_auto_model = transformers.AutoModel.from_pretrained
original_auto_tokenizer = transformers.AutoTokenizer.from_pretrained

def mock_from_pretrained_model(model_name):
    return mock_model

def mock_from_pretrained_tokenizer(model_name):
    return mock_tokenizer

transformers.AutoModel.from_pretrained = mock_from_pretrained_model
transformers.AutoTokenizer.from_pretrained = mock_from_pretrained_tokenizer

# Test with a simple string input
ARGS = {
    'text_or_texts': '这是一个测试文本'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch, tqdm, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_character_prompt Ran
ran; our contract draft was wrong, not the code
Neph0s/CoSER/gca_evaluation/utils.py
pointer only (licence: NONE) · get_code("c7933743a582b312")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, Neph0s/CoSER when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_general_embedding Ran
ran, honoured the contract we drafted
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("ccd15a636f8315df")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import math

torch.manual_seed(0)

# Since the function tries to download from HuggingFace, we need to mock
# the model and tokenizer pools to avoid network calls
from unittest.mock import MagicMock
import transformers

# Create mock tokenizer
mock_tokenizer = MagicMock()
mock_tokenizer.return_value = {
    'input_ids': torch.tensor([[101, 2054, 102], [101, 1045, 102]]),
    'attention_mask': torch.tensor([[1, 1, 1], [1, 1, 1]]),
    'token_type_ids': torch.tensor([[0, 0, 0], [0, 0, 0]])
}

# Create mock model
mock_model = MagicMock()
mock_model.eval = MagicMock(return_value=None)

# Mock output: [batch_size, seq_len, hidden_dim]
mock_output = MagicMock()
mock_output.__getitem__ = MagicMock(side_effect=lambda x: torch.randn(2, 3, 4) if x == 0 else None)

mock_model.return_value = mock_output
mock_model.__call__ = MagicMock(return_value=mock_output)

# Populate the pools to avoid actual downloads
import sys
sys.modules['transformers'].AutoTokenizer.from_pretrained = MagicMock(return_value=mock_tokenizer)
sys.modules['transformers'].AutoModel.from_pretrained = MagicMock(return_value=mock_model)

# Test with a string input
ARGS = {
    'text_or_texts': 'This is a test sentence.',
    'model_name': 'BAAI/bge-small-zh-v1.5'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_text_from_data Ran
ran; our contract draft was wrong, not the code
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("a96df6daca0f00a5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import base64

def base64_to_string(base64_data):
    byte_array = base64.b64decode(base64_data)
    text = byte_array.decode('utf-8')
    return text

def get_text_from_data(data):
    if 'text' in data:
        return data['text']
    elif 'enc_text' in data:
        return base64_to_string(data['enc_text'])
    else:
        print('warning! failed to get text from data ', data)
        return ''

# Test case 1: data with 'text' key
test_data_with_text = {
    'text': 'Hello, World!'
}

# Test case 2: data with 'enc_text' key (base64 encoded)
test_string = 'This is encoded text'
encoded_text = base64.b64encode(test_string.encode('utf-8')).decode('utf-8')
test_data_with_enc_text = {
    'enc_text': encoded_text
}

# Use the first test case as the main argument
ARGS = {
    'data': test_data_with_text
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
is_chinese_or_english Ran
ran; our contract draft was wrong, not the code
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("f5012fac79a4d832")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

# Construct valid arguments for is_chinese_or_english
# The function takes a single parameter: text (a string)
# It checks if the text contains more Chinese characters or English characters

ARGS = {
    'text': '你好world'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
response_postprocess Ran
ran; our contract draft was wrong, not the code
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("e0af9feb46733253")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import re

def response_postprocess(text, dialogue_bra_token='「', dialogue_ket_token='」'):
    lines = text.split('\n')
    new_lines = ''
    first_name = None
    for line in lines:
        line = line.strip(' ')
        match = re.match('^(.*?)[::]' + dialogue_bra_token + '(.*?)' + dialogue_ket_token + '$', line)
        if match:
            curr_name = match.group(1)
            if first_name is None:
                first_name = curr_name
                new_lines += match.group(2)
            elif curr_name != first_name:
                return first_name + ':' + dialogue_bra_token + new_lines + dialogue_ket_token
            else:
                new_lines += match.group(2)
        elif first_name == None:
            return text
        else:
            return first_name + ':' + dialogue_bra_token + new_lines + dialogue_ket_token
    return first_name + ':' + dialogue_bra_token + new_lines + dialogue_ket_token

ARGS = {
    'text': 'Alice:「Hello there」\nAlice:「How are you」',
    'dialogue_bra_token': '「',
    'dialogue_ket_token': '」'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ChatHaruhi Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("72ca06e73065b5f0")
How it ran

no invocation recorded.

cached Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("ded92e88e0b003ab")
How it ran

no invocation recorded.

download_models Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("3886f44e69594927")
How it ran

no invocation recorded.

get_bge_embedding Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("4a6302a49807a5e7")
How it ran

no invocation recorded.

get_bge_embeddings_safe Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("fb71211b0752f685")
How it ran

no invocation recorded.

get_db_from_type Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("d39ee52667a78ef8")
How it ran

no invocation recorded.

get_embedding_for_chinese Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("338cf403e8136171")
How it ran

no invocation recorded.

get_embedding_for_english Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("1acb7e8163ea94f9")
How it ran

no invocation recorded.

get_general_embeddings Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("b6444230426a2754")
How it ran

no invocation recorded.

get_general_embeddings_safe Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("9d6500eb148b11cd")
How it ran

no invocation recorded.

get_luotuo_model Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("936046d800c3e83d")
How it ran

no invocation recorded.

luotuo_embedding Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("cd9ff808f0197b70")
How it ran

no invocation recorded.

luotuo_openai_embedding Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("0ef004806ceb47bb")
How it ran

no invocation recorded.

tiktokenizer Not yet run
unverified (harvested, no recorded run)
neph0s/incharacter/code/ChatHaruhi/ChatHaruhi.py
code served (permissive licence) · get_code("55e9d4ef95279fed")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 23 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 22 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
get_text_from_data
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2308.09597 (2023-08)
ChatHaruhi: Reviving Anime Character in Reality via Large Language Model

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Role-playing language agents (RPLAs) have emerged as promising applications of large language models (LLMs). However, simulating established characters presents a challenging task for RPLAs, due to the lack of authentic character datasets and nuanced evaluation methods using such data. In this paper, we present CoSER, a collection of a high-quality dataset, open models, and an evaluation protocol towards effective RPLAs of established characters. The CoSER dataset covers 17,966 characters from 771 renowned books. It provides authentic dialogues with real-world intricacies, as well as diverse data types such as conversation setups, character experiences and internal thoughts. Drawing from acting methodology, we introduce givencircumstance acting for training and evaluating role-playing LLMs, where LLMs sequentially portray multiple characters in book scenes. Using our dataset, we develop CoSER 8B and CoSER 70B, i.e., advanced open role-playing LLMs built on LLaMA-3.1 models. Extensive experiments demonstrate the value of the CoSER dataset for RPLA training, evaluation and retrieval. Moreover, CoSER 70B exhibits state-of-the-art performance surpassing or matching GPT-4o on our evaluation and three existing benchmarks, i.e., achieving 75.80% and 93.47% accuracy on the InCharacter and LifeChoice benchmarks respectively. Our code, dataset and models are available at: https://github.com/Neph0s/CoSER.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2502.09082")
get_code_for_paper("2502.09082")
have("2502.09082")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2502.09082.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2502.09082.svg)](https://syntology.ai/paper/2502.09082)

Connect an agent — have() is free.