diff --git a/.translation-init b/.translation-init
index f8e3f7ac9b..7434727578 100644
--- a/.translation-init
+++ b/.translation-init
@@ -1 +1 @@
-Translation initialization: 2025-09-26T01:16:02.268240
+Translation initialization: 2025-09-26T04:21:21.696179
diff --git a/docs/cn/sql-reference/20-sql-functions/17-table-functions/01-infer-schema.md b/docs/cn/sql-reference/20-sql-functions/17-table-functions/01-infer-schema.md
index 7ab5fcf5e7..0db43ef033 100644
--- a/docs/cn/sql-reference/20-sql-functions/17-table-functions/01-infer-schema.md
+++ b/docs/cn/sql-reference/20-sql-functions/17-table-functions/01-infer-schema.md
@@ -4,11 +4,26 @@ title: INFER_SCHEMA
 
 自动检测文件元数据模式并检索列定义。
 
+`infer_schema` 目前支持以下文件格式：
+- **Parquet** - 原生支持模式推断
+- **CSV** - 支持自定义分隔符和表头检测
+- **NDJSON** - 换行分隔的 JSON 文件
 
-:::caution
+**压缩支持**：所有格式均支持扩展名为 `.zip`、`.xz`、`.zst` 的压缩文件。
 
-`infer_schema` 目前仅支持 parquet 文件格式。
+:::info 文件大小限制
+每个独立文件的模式推断最大大小限制为 **100MB**。
+:::
+
+:::info 模式合并
+处理多个文件时，`infer_schema` 会自动合并不同模式：
+
+- **兼容类型** 会被提升（例如，INT8 + INT16 → INT16）
+- **不兼容类型** 会回退到 **VARCHAR**（例如，INT + FLOAT → VARCHAR）
+- 某些文件中 **缺失的列** 会被标记为 **nullable**
+- 后续文件中的 **新列** 会被添加到最终模式
 
+这确保所有文件都能使用统一模式读取。
 :::
 
 ## 语法
@@ -17,81 +32,222 @@ title: INFER_SCHEMA
 INFER_SCHEMA(
   LOCATION => '{ internalStage | externalStage }'
   [ PATTERN => '<regex_pattern>']
+  [ FILE_FORMAT => '<format_name>' ]
+  [ MAX_RECORDS_PRE_FILE => <number> ]
+  [ MAX_FILE_COUNT => <number> ]
 )
 ```
 
-其中：
+## 参数
 
-### internalStage
+| 参数 | 描述 | 默认值 | 示例 |
+|-----------|-------------|---------|---------|
+| `LOCATION` | 暂存区位置：`@<stage_name>[/<path>]` | 必需 | `'@my_stage/data/'` |
+| `PATTERN` | 文件名匹配模式 | 所有文件 | `'*.csv'`, `'*.parquet'` |
+| `FILE_FORMAT` | 解析用的文件格式名称 | 暂存区格式 | `'csv_format'`, `'NDJSON'` |
+| `MAX_RECORDS_PRE_FILE` | 每文件采样的最大记录数 | 所有记录 | `100`, `1000` |
+| `MAX_FILE_COUNT` | 处理的最大文件数 | 所有文件 | `5`, `10` |
+
+## 示例
+
+### Parquet 文件
 
 ```sql
-internalStage ::= @<internal_stage_name>[/<path>]
+-- 创建暂存区并导出数据
+CREATE STAGE test_parquet;
+COPY INTO @test_parquet FROM (SELECT number FROM numbers(10)) FILE_FORMAT = (TYPE = 'PARQUET');
+
+-- 使用模式从 Parquet 文件推断模式
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_parquet',
+    pattern => '*.parquet'
+);
+```
+
+结果：
+```
++-------------+-----------------+----------+----------+----------+
+| column_name | type            | nullable | filenames| order_id |
++-------------+-----------------+----------+----------+----------+
+| number      | BIGINT UNSIGNED |    false | data_... |        0 |
++-------------+-----------------+----------+----------+----------+
 ```
 
-### externalStage
+### CSV 文件
 
 ```sql
-externalStage ::= @<external_stage_name>[/<path>]
+-- 创建暂存区并导出 CSV 数据
+CREATE STAGE test_csv;
+COPY INTO @test_csv FROM (SELECT number FROM numbers(10)) FILE_FORMAT = (TYPE = 'CSV');
+
+-- 创建 CSV 文件格式
+CREATE FILE FORMAT csv_format TYPE = 'CSV';
+
+-- 使用模式和文件格式推断模式
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_csv',
+    pattern => '*.csv',
+    file_format => 'csv_format'
+);
 ```
 
-### PATTERN = 'regex_pattern'
+结果：
+```
++-------------+---------+----------+----------+----------+
+| column_name | type    | nullable | filenames| order_id |
++-------------+---------+----------+----------+----------+
+| column_1    | BIGINT  |     true | data_... |        0 |
++-------------+---------+----------+----------+----------+
+```
 
-一个基于 [PCRE2](https://www.pcre.org/current/doc/html/) 的正则表达式模式字符串，用单引号括起来，指定要匹配的文件名。点击[这里](#loading-data-with-pattern-matching)查看示例。有关 PCRE2 语法，请参见 http://www.pcre.org/current/doc/html/pcre2syntax.html。
+带表头的 CSV 文件：
 
-## 示例
+```sql
+-- 创建支持表头的 CSV 文件格式
+CREATE FILE FORMAT csv_headers_format
+TYPE = 'CSV'
+field_delimiter = ','
+skip_header = 1;
+
+-- 导出带表头的数据
+CREATE STAGE test_csv_headers;
+COPY INTO @test_csv_headers FROM (
+  SELECT number as user_id, 'user_' || number::string as user_name
+  FROM numbers(5)
+) FILE_FORMAT = (TYPE = 'CSV', output_header = true);
+
+-- 推断带表头的模式
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_csv_headers',
+    file_format => 'csv_headers_format'
+);
+```
 
-在 stage 中生成一个 parquet 文件：
+限制记录数以加快推断：
 
 ```sql
-CREATE STAGE infer_parquet FILE_FORMAT = (TYPE = PARQUET);
-COPY INTO @infer_parquet FROM (SELECT * FROM numbers(10)) FILE_FORMAT = (TYPE = PARQUET);
+-- 仅采样前 5 条记录进行模式推断
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_csv',
+    pattern => '*.csv',
+    file_format => 'csv_format',
+    max_records_pre_file => 5
+);
 ```
 
+### NDJSON 文件
+
 ```sql
-LIST @infer_parquet;
-+-------------------------------------------------------+------+------------------------------------+-------------------------------+---------+
-| name                                                  | size | md5                                | last_modified                 | creator |
-+-------------------------------------------------------+------+------------------------------------+-------------------------------+---------+
-| data_e0fd9cba-f45c-4c43-aa07-d6d87d134378_0_0.parquet |  258 | "7DCC9FFE04EA1F6882AED2CF9640D3D4" | 2023-02-09 05:21:52.000 +0000 | NULL    |
-+-------------------------------------------------------+------+------------------------------------+-------------------------------+---------+
+-- 创建暂存区并导出 NDJSON 数据
+CREATE STAGE test_ndjson;
+COPY INTO @test_ndjson FROM (SELECT number FROM numbers(10)) FILE_FORMAT = (TYPE = 'NDJSON');
+
+-- 使用模式和 NDJSON 格式推断模式
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_ndjson',
+    pattern => '*.ndjson',
+    file_format => 'NDJSON'
+);
 ```
 
-### `infer_schema`
+结果：
+```
++-------------+---------+----------+----------+----------+
+| column_name | type    | nullable | filenames| order_id |
++-------------+---------+----------+----------+----------+
+| number      | BIGINT  |     true | data_... |        0 |
++-------------+---------+----------+----------+----------+
+```
 
+限制记录数以加快推断：
 
 ```sql
-SELECT * FROM INFER_SCHEMA(location => '@infer_parquet/data_e0fd9cba-f45c-4c43-aa07-d6d87d134378_0_0.parquet');
-+-------------+-----------------+----------+----------+
-| column_name | type            | nullable | order_id |
-+-------------+-----------------+----------+----------+
-| number      | BIGINT UNSIGNED |        0 |        0 |
-+-------------+-----------------+----------+----------+
+-- 仅采样前 5 条记录进行模式推断
+SELECT * FROM INFER_SCHEMA(
+    location => '@test_ndjson',
+    pattern => '*.ndjson',
+    file_format => 'NDJSON',
+    max_records_pre_file => 5
+);
 ```
 
-### 使用模式匹配的 `infer_schema`
+### 多文件模式合并
+
+当文件模式不同时，`infer_schema` 会智能合并：
 
 ```sql
-SELECT * FROM infer_schema(location => '@infer_parquet/', pattern => '.*parquet');
-+-------------+-----------------+----------+----------+
-| column_name | type            | nullable | order_id |
-+-------------+-----------------+----------+----------+
-| number      | BIGINT UNSIGNED |        0 |        0 |
-+-------------+-----------------+----------+----------+
+-- 假设有多个不同模式的 CSV 文件：
+-- file1.csv: id(INT), name(VARCHAR)
+-- file2.csv: id(INT), name(VARCHAR), age(INT)
+-- file3.csv: id(FLOAT), name(VARCHAR), age(INT)
+
+SELECT * FROM INFER_SCHEMA(
+    location => '@my_stage/',
+    pattern => '*.csv',
+    file_format => 'csv_format'
+);
 ```
 
-### 从 Parquet 文件创建表
+结果显示合并后的模式：
+```
++-------------+---------+----------+-----------+----------+
+| column_name | type    | nullable | filenames | order_id |
++-------------+---------+----------+-----------+----------+
+| id          | VARCHAR |     true | file1,... |        0 |  -- INT+FLOAT→VARCHAR
+| name        | VARCHAR |     true | file1,... |        1 |
+| age         | BIGINT  |     true | file1,... |        2 |  -- file1 缺失→nullable
++-------------+---------+----------+-----------+----------+
+```
 
-`infer_schema` 只能显示 parquet 文件的模式，无法从中创建表。
+### 模式匹配与文件限制
 
-要从 parquet 文件创建表：
+使用模式匹配从多个文件推断模式：
 
 ```sql
-CREATE TABLE mytable AS SELECT * FROM @infer_parquet/ (pattern=>'.*parquet') LIMIT 0;
-
-DESC mytable;
-+--------+-----------------+------+---------+-------+
-| Field  | Type            | Null | Default | Extra |
-+--------+-----------------+------+---------+-------+
-| number | BIGINT UNSIGNED | NO   | 0       |       |
-+--------+-----------------+------+---------+-------+
+-- 从目录中所有 CSV 文件推断模式
+SELECT * FROM INFER_SCHEMA(
+    location => '@my_stage/',
+    pattern => '*.csv'
+);
+```
+
+限制处理文件数以提升性能：
+
+```sql
+-- 仅处理前 5 个匹配文件
+SELECT * FROM INFER_SCHEMA(
+    location => '@my_stage/',
+    pattern => '*.csv',
+    max_file_count => 5
+);
+```
+
+### 压缩文件
+
+`infer_schema` 自动处理压缩文件：
+
+```sql
+-- 适用于压缩 CSV 文件
+SELECT * FROM INFER_SCHEMA(location => '@my_stage/data.csv.zip');
+
+-- 适用于压缩 NDJSON 文件
+SELECT * FROM INFER_SCHEMA(
+    location => '@my_stage/data.ndjson.xz',
+    file_format => 'NDJSON',
+    max_records_pre_file => 50
+);
+```
+
+### 从推断模式创建表
+
+`infer_schema` 函数显示模式但不创建表。要从推断模式创建表：
+
+```sql
+-- 从文件模式创建表结构
+CREATE TABLE my_table AS
+SELECT * FROM @my_stage/ (pattern=>'*.parquet')
+LIMIT 0;
+
+-- 验证表结构
+DESC my_table;
 ```
\ No newline at end of file