evalscope 0.17.1__py3-none-any.whl → 1.0.0__py3-none-any.whl

evalscope/utils/model_utils.py

@@ -1,7 +1,5 @@
 import numpy as np
-import os
 import random
-import torch
 from enum import Enum
 from typing import TYPE_CHECKING, Any, Dict, Optional, Tuple, Union
 
@@ -69,8 +67,13 @@ def seed_everything(seed: int):
     """
     random.seed(seed)
     np.random.seed(seed)
-    torch.manual_seed(seed)
-    if torch.cuda.is_available():
-        torch.cuda.manual_seed_all(seed)
-        torch.backends.cudnn.deterministic = True
-        torch.backends.cudnn.benchmark = False
+    try:
+        import torch
+
+        torch.manual_seed(seed)
+        if torch.cuda.is_available():
+            torch.cuda.manual_seed_all(seed)
+            torch.backends.cudnn.deterministic = True
+            torch.backends.cudnn.benchmark = False
+    except ImportError:
+        pass

evalscope/utils/multi_choices.py

@@ -0,0 +1,271 @@
+# flake8: noqa: E501
+from __future__ import annotations
+
+import re
+from typing import TYPE_CHECKING, List, Optional
+
+if TYPE_CHECKING:
+    from evalscope.api.evaluator import Choices, Target, TaskState
+
+FEW_SHOT_TEMPLATE = r"""Here are some examples of how to answer similar questions:
+
+{fewshot}
+
+""".lstrip()
+
+CHINESE_FEW_SHOT_TEMPLATE = r"""以下是一些示例问题：
+
+{fewshot}
+
+""".lstrip()
+
+CHINESE_SINGLE_ANSWER_TEMPLATE = r"""回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：LETTER"（不带引号），其中 LETTER 是 {letters} 中的一个。
+
+问题：{question}
+选项：
+{choices}
+""".lstrip()
+
+CHINESE_SINGLE_ANSWER_TEMPLATE_COT = r"""回答下面的单项选择题，请选出其中的正确答案。你的回答的最后一行应该是这样的格式："答案：LETTER"（不带引号），其中 LETTER 是 {letters} 中的一个。请在回答前进行一步步思考。
+
+问题：{question}
+选项：
+{choices}
+""".lstrip()
+
+SINGLE_ANSWER_TEMPLATE = r"""
+Answer the following multiple choice question. The entire content of your response should be of the following format: 'ANSWER: $LETTER' (without quotes) where LETTER is one of {letters}.
+
+{question}
+
+{choices}
+""".strip()
+
+SINGLE_ANSWER_TEMPLATE_COT = r"""
+Answer the following multiple choice question. The last line of your response should be of the following format: 'ANSWER: $LETTER' (without quotes) where LETTER is one of {letters}. Think step by step before answering.
+
+{question}
+
+{choices}
+""".strip()
+
+MULTIPLE_ANSWER_TEMPLATE = r"""
+Answer the following multiple choice question where multiple answers may be correct. The entire content of your response should be of the following format: 'ANSWER: $LETTERS' (without quotes) where LETTERS is one or more of {letters}.
+
+{question}
+
+{choices}
+""".strip()
+
+MULTIPLE_ANSWER_TEMPLATE_COT = r"""
+Answer the following multiple choice question where multiple answers may be correct. The last line of your response should be of the following format: 'ANSWER: $LETTERS' (without quotes) where LETTERS is one or more of {letters}. Think step by step before answering.
+
+{question}
+
+{choices}
+""".strip()
+
+
+def unshuffle_choices(choices: Choices) -> Choices:
+    # `sorted` returns `list[Choice]`, but for consistency we wrap this back
+    # into a `Choices` object
+    return Choices(sorted(choices, key=lambda choice: choice.original_position))
+
+
+def answer_options(choices: Choices) -> str:
+    r"""
+    Returns the `choices` formatted as a multiple choice question, e.g.:
+
+    ["choice 1", "choice 2", "choice 3"] ->
+        "A) choice 1\nB) choice 2\nC) choice 3"
+    """
+    indexes = list(range(len(choices)))
+
+    return '\n'.join([f'{answer_character(i)}) {choices[j].value}' for i, j in enumerate(indexes)])
+
+
+def prompt(question: str, choices: Choices, template: str, fewshot: Optional[str] = None) -> str:
+
+    choices_text = answer_options(choices)
+    letters = ','.join(answer_character(i) for i in range(len(choices)))
+    if not fewshot:
+        return template.format(
+            choices=choices_text,
+            letters=letters,
+            question=question,
+        )
+    else:
+        return template.format(
+            choices=choices_text,
+            letters=letters,
+            question=question,
+            fewshot=fewshot,
+        )
+
+
+def format_example(
+    question: str,
+    choices: Choices,
+    answer: Target,
+) -> str:
+    """Format a single example for few-shot learning.
+
+    Args:
+        question (str): The question text.
+        choices (list[str]): The list of choices.
+        answer (list[str]): The correct answers.
+
+    Returns:
+        str: Formatted example string.
+    """
+    choices_text = answer_options(choices)
+    return f'{question}\n{choices_text}\nANSWER: {answer.text}'
+
+
+def parse_answers(state: TaskState, multiple_correct: bool = False) -> set[str]:
+    """
+    Convenience function for extracting answers from the state output.
+
+    The generated response must be in the format 'ANSWER: <answers>',
+    otherwise we can't extract what the model thinks is "true". We can be a
+    bit flexible whether these are "AB" vs "A,B" vs "A B".
+
+    However, if the answer isn't in the expected format the model has
+    failed in the task so we'll ultimately just mark it as incorrect
+    """
+    # First check whether the string strictly ends with the expected answer
+    # In this case, we're looking for a single line which contains the expected
+    # ANSWER: <answer> string with only whitespace or a period/full stop at the end.
+    match = re.search(
+        r'(?i)^ANSWER\s*:\s*([A-Za-z\d ,]+)\s*(?:$|\n|\.)',
+        state.output.completion,
+        flags=re.MULTILINE,
+    )
+
+    # If we couldn't match the strict version, we can try the less strict
+    # version for backward compatibility
+    if match is None:
+        match = re.search(
+            r'(?i)ANSWER\s*:\s*([A-Za-z\d ,]+)(?:[^\w]|\n|$|\.)',
+            state.output.completion,
+        )
+
+    if match is None:
+        return set()
+
+    matched = match.group(1)
+
+    # Strip trailing period / full stop
+    matched = matched.strip()
+    matched = matched.rstrip('.')
+
+    allowed_options = set(answer_character(i) for i in range(len(state.choices)))
+
+    if multiple_correct:
+        # Match must contain only the allowed choices
+        # (may be separated by commas, spaces, the word 'and', or nothing at all)
+
+        matched = matched.replace(' and ', '')
+
+        matched = matched.replace(' ', '')
+
+        split_comma = set(matched.split(','))
+        if split_comma.issubset(allowed_options):
+            answers = split_comma
+            return answers
+
+        split_nothing = set(matched)
+        if split_nothing.issubset(allowed_options):
+            answers = split_nothing
+            return answers
+
+    else:
+        # Match must contain a single letter in the allowed choices
+        if matched in allowed_options:
+            answers = {matched}
+            return answers
+
+    return set()
+
+
+def parse_answers_zh(state: TaskState, multiple_correct: bool = False) -> set[str]:
+    """
+    Convenience function for extracting answers from the state output in Chinese format.
+
+    The generated response must be in the format '答案：选项',
+    otherwise we can't extract what the model thinks is "true". We can be a
+    bit flexible whether these are "AB" vs "A,B" vs "A B".
+    """
+    # Simple pattern to capture answers with optional bold markdown
+    pattern = r'答案\s*[:：]\s*([A-Za-z0-9,，]+)'
+    match = re.search(pattern, state.output.completion, flags=re.MULTILINE)
+
+    if match is None:
+        return set()
+
+    matched = match.group(1).strip().rstrip('。.')
+    allowed_options = set(answer_character(i) for i in range(len(state.choices)))
+
+    if multiple_correct:
+        # Handle comma-separated or continuous letters
+        matched = matched.replace(' 和 ', '').replace(' ', '').replace('，', ',')
+        answers = set(matched.split(',')) if ',' in matched else set(matched)
+        return answers if answers.issubset(allowed_options) else set()
+    else:
+        # Single answer
+        return {matched} if matched in allowed_options else set()
+
+
+def set_choices_based_on_generated_response(state: TaskState, answers: set[str]) -> None:
+    true_answers = [answer_index(letter) for letter in answers]
+
+    for i in range(len(state.choices)):
+        if i in true_answers:
+            state.choices.mark_choice(i, True)
+        else:
+            state.choices.mark_choice(i, False)
+
+
+def valid_template(template: str) -> bool:
+    """Check if a template has the required capture groups for a multiple choice question"""
+    return bool(re.search(r'\{question\}', template) and re.search(r'\{choices\}', template))
+
+
+class MultipleChoiceTemplate:
+    """
+    Templates for multiple choice questions.
+    """
+
+    SINGLE_ANSWER = SINGLE_ANSWER_TEMPLATE
+    SINGLE_ANSWER_COT = SINGLE_ANSWER_TEMPLATE_COT
+    MULTIPLE_ANSWER = MULTIPLE_ANSWER_TEMPLATE
+    MULTIPLE_ANSWER_COT = MULTIPLE_ANSWER_TEMPLATE_COT
+    CHINESE_FEW_SHOT_TEMPLATE = CHINESE_FEW_SHOT_TEMPLATE
+    CHINESE_SINGLE_ANSWER_TEMPLATE = CHINESE_SINGLE_ANSWER_TEMPLATE
+    CHINESE_SINGLE_ANSWER_TEMPLATE_COT = CHINESE_SINGLE_ANSWER_TEMPLATE_COT
+
+
+def answer_character(index: int) -> str:
+    r"""
+    Helper to go from array index to char, for example:
+
+        0 -> 'A', 1 -> 'B', etc
+    """
+    if index < 26:
+        return chr(ord('A') + index)
+    else:
+        return str(index - 25)
+
+
+def answer_index(char: str) -> int:
+    r"""
+    Helper to go from char to array index, for example:
+
+        'A' -> 0, 'B' -> 1, etc
+    """
+    if char.isalpha() or char == ',' or char == ' ':
+        return ord(char.upper()) - ord('A')
+    elif char.isnumeric():
+        return 25 + int(char)
+    else:
+        raise ValueError(f'Unepxected multiple choice answer: {char} (must be a letter or number)')

evalscope/utils/url_utils.py

@@ -0,0 +1,65 @@
+import base64
+import httpx
+import mimetypes
+import re
+
+
+def is_http_url(url: str) -> bool:
+    return url.startswith('http://') or url.startswith('https://')
+
+
+def is_data_uri(url: str) -> bool:
+    pattern = r'^data:([^;]+);base64,.*'
+    return re.match(pattern, url) is not None
+
+
+def data_uri_mime_type(data_url: str) -> str | None:
+    pattern = r'^data:([^;]+);.*'
+    match = re.match(pattern, data_url)
+    if match:
+        mime_type = match.group(1)
+        return mime_type
+    else:
+        return None
+
+
+def data_uri_to_base64(data_uri: str) -> str:
+    pattern = r'^data:[^,]+,'
+    stripped_uri = re.sub(pattern, '', data_uri)
+    return stripped_uri
+
+
+def file_as_data(file: str) -> tuple[bytes, str]:
+    if is_data_uri(file):
+        # resolve mime type and base64 content
+        mime_type = data_uri_mime_type(file) or 'image/png'
+        file_base64 = data_uri_to_base64(file)
+        file_bytes = base64.b64decode(file_base64)
+    else:
+        # guess mime type; need strict=False for webp images
+        type, _ = mimetypes.guess_type(file, strict=False)
+        if type:
+            mime_type = type
+        else:
+            mime_type = 'image/png'
+
+        # handle url or file
+        if is_http_url(file):
+            client = httpx.Client()
+            file_bytes = client.get(file).content
+        else:
+            with open(file, 'rb') as f:
+                file_bytes = f.read()
+
+    # return bytes and type
+    return file_bytes, mime_type
+
+
+def file_as_data_uri(file: str) -> str:
+    if is_data_uri(file):
+        return file
+    else:
+        bytes, mime_type = file_as_data(file)
+        base64_file = base64.b64encode(bytes).decode('utf-8')
+        file = f'data:{mime_type};base64,{base64_file}'
+        return file

evalscope/version.py

@@ -1,4 +1,4 @@
 # Copyright (c) Alibaba, Inc. and its affiliates.
 
-__version__ = '0.17.1'
-__release_datetime__ = '2025-07-18 17:00:00'
+__version__ = '1.0.0'
+__release_datetime__ = '2025-08-25 12:00:00'

{evalscope-0.17.1.dist-info → evalscope-1.0.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: evalscope
-Version: 0.17.1
+Version: 1.0.0
 Summary: EvalScope: Lightweight LLMs Evaluation Framework
 Home-page: https://github.com/modelscope/evalscope
 Author: ModelScope team
@@ -18,7 +18,10 @@ Requires-Python: >=3.9
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: accelerate
-Requires-Dist: datasets==3.2.0
+Requires-Dist: colorlog
+Requires-Dist: datasets==3.6.0
+Requires-Dist: docstring-parser
+Requires-Dist: dotenv
 Requires-Dist: immutabledict
 Requires-Dist: jieba
 Requires-Dist: jsonlines
@@ -28,11 +31,14 @@ Requires-Dist: matplotlib
 Requires-Dist: modelscope[framework]>=1.27
 Requires-Dist: nltk>=3.9
 Requires-Dist: openai
+Requires-Dist: overrides
 Requires-Dist: pandas
 Requires-Dist: pillow
 Requires-Dist: pyarrow
+Requires-Dist: pydantic
 Requires-Dist: pyyaml>=5.1
 Requires-Dist: requests
+Requires-Dist: rich
 Requires-Dist: rouge-chinese
 Requires-Dist: rouge-score>=0.1.0
 Requires-Dist: sacrebleu
@@ -50,10 +56,14 @@ Requires-Dist: iopath; extra == "aigc"
 Requires-Dist: omegaconf; extra == "aigc"
 Requires-Dist: open-clip-torch; extra == "aigc"
 Requires-Dist: opencv-python; extra == "aigc"
+Requires-Dist: peft>=0.17; extra == "aigc"
 Requires-Dist: torchvision; extra == "aigc"
 Provides-Extra: all
 Requires-Dist: accelerate; extra == "all"
-Requires-Dist: datasets==3.2.0; extra == "all"
+Requires-Dist: colorlog; extra == "all"
+Requires-Dist: datasets==3.6.0; extra == "all"
+Requires-Dist: docstring-parser; extra == "all"
+Requires-Dist: dotenv; extra == "all"
 Requires-Dist: immutabledict; extra == "all"
 Requires-Dist: jieba; extra == "all"
 Requires-Dist: jsonlines; extra == "all"
@@ -63,11 +73,14 @@ Requires-Dist: matplotlib; extra == "all"
 Requires-Dist: modelscope[framework]>=1.27; extra == "all"
 Requires-Dist: nltk>=3.9; extra == "all"
 Requires-Dist: openai; extra == "all"
+Requires-Dist: overrides; extra == "all"
 Requires-Dist: pandas; extra == "all"
 Requires-Dist: pillow; extra == "all"
 Requires-Dist: pyarrow; extra == "all"
+Requires-Dist: pydantic; extra == "all"
 Requires-Dist: pyyaml>=5.1; extra == "all"
 Requires-Dist: requests; extra == "all"
+Requires-Dist: rich; extra == "all"
 Requires-Dist: rouge-chinese; extra == "all"
 Requires-Dist: rouge-score>=0.1.0; extra == "all"
 Requires-Dist: sacrebleu; extra == "all"
@@ -91,7 +104,6 @@ Requires-Dist: webdataset>0.2.0; extra == "all"
 Requires-Dist: aiohttp; extra == "all"
 Requires-Dist: fastapi; extra == "all"
 Requires-Dist: numpy; extra == "all"
-Requires-Dist: rich; extra == "all"
 Requires-Dist: sse-starlette; extra == "all"
 Requires-Dist: transformers; extra == "all"
 Requires-Dist: uvicorn; extra == "all"
@@ -102,8 +114,9 @@ Requires-Dist: iopath; extra == "all"
 Requires-Dist: omegaconf; extra == "all"
 Requires-Dist: open-clip-torch; extra == "all"
 Requires-Dist: opencv-python; extra == "all"
+Requires-Dist: peft>=0.17; extra == "all"
 Requires-Dist: torchvision; extra == "all"
-Requires-Dist: bfcl-eval; extra == "all"
+Requires-Dist: bfcl-eval==2025.6.16; extra == "all"
 Requires-Dist: human-eval; extra == "all"
 Requires-Dist: pytest; extra == "all"
 Requires-Dist: pytest-cov; extra == "all"
@@ -112,7 +125,7 @@ Provides-Extra: app
 Requires-Dist: gradio==5.4.0; extra == "app"
 Requires-Dist: plotly<6.0.0,>=5.23.0; extra == "app"
 Provides-Extra: dev
-Requires-Dist: bfcl-eval; extra == "dev"
+Requires-Dist: bfcl-eval==2025.6.16; extra == "dev"
 Requires-Dist: human-eval; extra == "dev"
 Requires-Dist: pytest; extra == "dev"
 Requires-Dist: pytest-cov; extra == "dev"
@@ -175,9 +188,9 @@ Requires-Dist: ms-vlmeval>=0.0.17; extra == "vlmeval"
 - [📝 Introduction](#-introduction)
 - [☎ User Groups](#-user-groups)
 - [🎉 News](#-news)
-- [🛠️ Installation](#️-installation)
-  - [Method 1: Install Using pip](#method-1-install-using-pip)
-  - [Method 2: Install from Source](#method-2-install-from-source)
+- [🛠️ Environment Setup](#️-environment-setup)
+  - [Method 1. Install via pip](#method-1-install-via-pip)
+  - [Method 2. Install from source](#method-2-install-from-source)
 - [🚀 Quick Start](#-quick-start)
   - [Method 1. Using Command Line](#method-1-using-command-line)
   - [Method 2. Using Python Code](#method-2-using-python-code)
@@ -258,6 +271,13 @@ Please scan the QR code below to join our community groups:
 
 
 ## 🎉 News
+
+> [!IMPORTANT]
+> **Version 1.0 Refactoring**
+>
+> Version 1.0 introduces a major overhaul of the evaluation framework, establishing a new, more modular and extensible API layer under `evalscope/api`. Key improvements include standardized data models for benchmarks, samples, and results; a registry-based design for components such as benchmarks and metrics; and a rewritten core evaluator that orchestrates the new architecture. Existing benchmark adapters have been migrated to this API, resulting in cleaner, more consistent, and easier-to-maintain implementations.
+
+- 🔥 **[2025.08.22]** Version 1.0 Refactoring.
 - 🔥 **[2025.07.18]** The model stress testing now supports randomly generating image-text data for multimodal model evaluation. For usage instructions, refer to the [documentation](https://evalscope.readthedocs.io/en/latest/user_guides/stress_test/examples.html#id4).
 - 🔥 **[2025.07.16]** Support for [τ-bench](https://github.com/sierra-research/tau-bench) has been added, enabling the evaluation of AI Agent performance and reliability in real-world scenarios involving dynamic user and tool interactions. For usage instructions, please refer to the [documentation](https://evalscope.readthedocs.io/zh-cn/latest/get_started/supported_dataset/llm.html#bench).
 - 🔥 **[2025.07.14]** Support for "Humanity's Last Exam" ([Humanity's-Last-Exam](https://modelscope.cn/datasets/cais/hle)), a highly challenging evaluation benchmark. For usage instructions, refer to the [documentation](https://evalscope.readthedocs.io/en/latest/get_started/supported_dataset/llm.html#humanity-s-last-exam).
@@ -268,12 +288,12 @@ Please scan the QR code below to join our community groups:
 - 🔥 **[2025.05.29]** Added support for two long document evaluation benchmarks: [DocMath](https://modelscope.cn/datasets/yale-nlp/DocMath-Eval/summary) and [FRAMES](https://modelscope.cn/datasets/iic/frames/summary). For usage guidelines, please refer to the [documentation](https://evalscope.readthedocs.io/en/latest/get_started/supported_dataset.html).
 - 🔥 **[2025.05.16]** Model service performance stress testing now supports setting various levels of concurrency and outputs a performance test report. [Reference example](https://evalscope.readthedocs.io/en/latest/user_guides/stress_test/quick_start.html#id3).
 - 🔥 **[2025.05.13]** Added support for the [ToolBench-Static](https://modelscope.cn/datasets/AI-ModelScope/ToolBench-Static) dataset to evaluate model's tool-calling capabilities. Refer to the [documentation](https://evalscope.readthedocs.io/en/latest/third_party/toolbench.html) for usage instructions. Also added support for the [DROP](https://modelscope.cn/datasets/AI-ModelScope/DROP/dataPeview) and [Winogrande](https://modelscope.cn/datasets/AI-ModelScope/winogrande_val) benchmarks to assess the reasoning capabilities of models.
+<details><summary>More</summary>
+
 - 🔥 **[2025.04.29]** Added Qwen3 Evaluation Best Practices, [welcome to read 📖](https://evalscope.readthedocs.io/en/latest/best_practice/qwen3.html)
 - 🔥 **[2025.04.27]** Support for text-to-image evaluation: Supports 8 metrics including MPS, HPSv2.1Score, etc., and evaluation benchmarks such as EvalMuse, GenAI-Bench. Refer to the [user documentation](https://evalscope.readthedocs.io/en/latest/user_guides/aigc/t2i.html) for more details.
 - 🔥 **[2025.04.10]** Model service stress testing tool now supports the `/v1/completions` endpoint (the default endpoint for vLLM benchmarking)
 - 🔥 **[2025.04.08]** Support for evaluating embedding model services compatible with the OpenAI API has been added. For more details, check the [user guide](https://evalscope.readthedocs.io/en/latest/user_guides/backend/rageval_backend/mteb.html#configure-evaluation-parameters).
-<details><summary>More</summary>
-
 - 🔥 **[2025.03.27]** Added support for [AlpacaEval](https://www.modelscope.cn/datasets/AI-ModelScope/alpaca_eval/dataPeview) and [ArenaHard](https://modelscope.cn/datasets/AI-ModelScope/arena-hard-auto-v0.1/summary) evaluation benchmarks. For usage notes, please refer to the [documentation](https://evalscope.readthedocs.io/en/latest/get_started/supported_dataset.html)
 - 🔥 **[2025.03.20]** The model inference service stress testing now supports generating prompts of specified length using random values. Refer to the [user guide](https://evalscope.readthedocs.io/en/latest/user_guides/stress_test/examples.html#using-the-random-dataset) for more details.
 - 🔥 **[2025.03.13]** Added support for the [LiveCodeBench](https://www.modelscope.cn/datasets/AI-ModelScope/code_generation_lite/summary) code evaluation benchmark, which can be used by specifying `live_code_bench`. Supports evaluating QwQ-32B on LiveCodeBench, refer to the [best practices](https://evalscope.readthedocs.io/en/latest/best_practice/eval_qwq.html).
@@ -306,58 +326,87 @@ Please scan the QR code below to join our community groups:
 
 </details>
 
-## 🛠️ Installation
-### Method 1: Install Using pip
-We recommend using conda to manage your environment and installing dependencies with pip:
+## 🛠️ Environment Setup
+
+### Method 1. Install via pip
+
+We recommend using conda to manage your environment and pip to install dependencies. This allows you to use the latest evalscope PyPI package.
 
 1. Create a conda environment (optional)
+```shell
+# Python 3.10 is recommended
+conda create -n evalscope python=3.10
+
+# Activate the conda environment
+conda activate evalscope
+```
+2. Install dependencies via pip
+```shell
+pip install evalscope
+```
+3. Install additional dependencies (optional)
+  - To use model service inference benchmarking features, install the perf dependency:
     ```shell
-    # It is recommended to use Python 3.10
-    conda create -n evalscope python=3.10
-    # Activate the conda environment
-    conda activate evalscope
+    pip install 'evalscope[perf]'
     ```
-
-2. Install dependencies using pip
+  - To use visualization features, install the app dependency:
+    ```shell
+    pip install 'evalscope[app]'
+    ```
+  - If you need to use other evaluation backends, you can install OpenCompass, VLMEvalKit, or RAGEval as needed:
+    ```shell
+    pip install 'evalscope[opencompass]'
+    pip install 'evalscope[vlmeval]'
+    pip install 'evalscope[rag]'
+    ```
+  - To install all dependencies:
     ```shell
-    pip install evalscope                # Install Native backend (default)
-    # Additional options
-    pip install 'evalscope[opencompass]'   # Install OpenCompass backend
-    pip install 'evalscope[vlmeval]'       # Install VLMEvalKit backend
-    pip install 'evalscope[rag]'           # Install RAGEval backend
-    pip install 'evalscope[perf]'          # Install dependencies for the model performance testing module
-    pip install 'evalscope[app]'           # Install dependencies for visualization
-    pip install 'evalscope[all]'           # Install all backends (Native, OpenCompass, VLMEvalKit, RAGEval)
+    pip install 'evalscope[all]'
     ```
 
-> [!WARNING]
-> As the project has been renamed to `evalscope`, for versions `v0.4.3` or earlier, you can install using the following command:
+> [!NOTE]
+> The project has been renamed to `evalscope`. For version `v0.4.3` or earlier, you can install it with:
 > ```shell
-> pip install llmuses<=0.4.3
+>  pip install llmuses<=0.4.3
 > ```
-> To import relevant dependencies using `llmuses`:
-> ``` python
+> Then, import related dependencies using `llmuses`:
+> ```python
 > from llmuses import ...
 > ```
 
-### Method 2: Install from Source
-1. Download the source code
-    ```shell
-    git clone https://github.com/modelscope/evalscope.git
-    ```
+### Method 2. Install from source
+
+Installing from source allows you to use the latest code and makes it easier for further development and debugging.
 
+1. Clone the source code
+```shell
+git clone https://github.com/modelscope/evalscope.git
+```
 2. Install dependencies
-    ```shell
-    cd evalscope/
-    pip install -e .                  # Install Native backend
-    # Additional options
-    pip install -e '.[opencompass]'   # Install OpenCompass backend
-    pip install -e '.[vlmeval]'       # Install VLMEvalKit backend
-    pip install -e '.[rag]'           # Install RAGEval backend
-    pip install -e '.[perf]'          # Install Perf dependencies
-    pip install -e '.[app]'           # Install visualization dependencies
-    pip install -e '.[all]'           # Install all backends (Native, OpenCompass, VLMEvalKit, RAGEval)
-    ```
+```shell
+cd evalscope/
+
+pip install -e .
+```
+3. Install additional dependencies
+ - To use model service inference benchmarking features, install the perf dependency:
+   ```shell
+   pip install '.[perf]'
+   ```
+ - To use visualization features, install the app dependency:
+   ```shell
+   pip install '.[app]'
+   ```
+ - If you need to use other evaluation backends, you can install OpenCompass, VLMEvalKit, or RAGEval as needed:
+   ```shell
+   pip install '.[opencompass]'
+   pip install '.[vlmeval]'
+   pip install '.[rag]'
+   ```
+ - To install all dependencies:
+   ```shell
+   pip install '.[all]'
+   ```
 
 
 ## 🚀 Quick Start