Xing4.0-29B-A4B-Table_16k

Xing4.0-29B-A4B-Table_16k

Xing4.0-29B-A4B-Table_16k is a fine-tuned version of Xing4.0-29B-A4B, specifically optimized for spreadsheet understanding and data analysis.

The model is designed to analyze structured data stored in CSV files and spreadsheet workbooks. With a code execution environment, it can load files, inspect data structures, clean and transform data, perform statistical calculations, generate visualizations, and summarize analytical findings.

Xing4.0-29B-A4B-Table_16k is suitable for spreadsheet question answering, business data analysis, financial analysis, operational reporting, and data quality inspection.

Base Model

The base model is Xing4.0-29B-A4B (formerly the TeleChat series), developed by China Telecom AI Technology Co., Ltd.

It uses a Mixture-of-Experts (MoE) architecture with 29B total parameters, activating only 4B parameters per inference step, balancing generation quality and deployment efficiency. It natively supports a 256K context window, providing a strong foundation for processing long documents, large tables, and complex analytical instructions.

The base model was trained on Ascend NPUs using the MindSpore framework.

Training Details

• Fine-tuning Task: Spreadsheet understanding and data analysis
• Primary Data Formats: CSV, XLS, and XLSX
• Training Data: High-quality samples covering spreadsheet inspection, data cleaning, statistical calculation, comparative analysis, trend analysis, anomaly detection, visualization, and analytical report generation
• Training Length: 16K
• Training Method: Full-parameter fine-tuning
• Training Hardware: Ascend NPU cluster
• Training Framework: MindSpore

Key Capabilities

Quickstart

Inference

Xing4.0-29B-A4B-Table_16k can be accessed through an OpenAI-compatible API:

from openai import OpenAI

client = OpenAI(
    base_url="your-base-url",
    api_key="your-api-key",
)

table = """
| Month | Revenue (10k CNY) | Cost (10k CNY) | Customers |
|------|-------------------:|---------------:|----------:|
| Jan  | 120                | 78             | 860       |
| Feb  | 135                | 82             | 910       |
| Mar  | 128                | 85             | 940       |
| Apr  | 156                | 91             | 1020      |
"""

prompt = f"""
请分析以下经营数据：

{table}

请完成以下任务：
1. 计算各月利润和利润率；
2. 分析收入、成本和客户数的变化趋势；
3. 找出值得关注的异常或风险；
4. 给出三条有数据依据的经营建议。

请以结构化分析报告的形式输出，并明确说明计算过程和结论依据。
"""

completion = client.chat.completions.create(
    model="Xing4.0-29B-A4B-Table_16k",
    messages=[
        {
            "role": "user",
            "content": prompt,
        }
    ],
    temperature=1.1,
    top_p=0.95,
    extra_body={
        "repetition_penalty": 1.02,
        "skip_special_tokens": False,
        "spaces_between_special_tokens": False,
        "chat_template_kwargs": {
            "enable_thinking": True,
        },
    },
)

print(completion.choices[0].message.content)

Recommended Parameters

Recommended Analysis Tools

The model works best when integrated with a code execution environment containing common data-analysis libraries:

• pandas for tabular data loading, cleaning, transformation, and aggregation;
• NumPy for numerical calculations;
• openpyxl for reading and processing XLSX workbooks;
• xlrd or other compatible libraries for legacy XLS files;
• Matplotlib and Seaborn for data visualization;
• SciPy or statsmodels for advanced statistical analysis.