CLI update: subcommand must be specified

README.md

@@ -47,8 +47,9 @@ aha is a high-performance, cross-platform AI inference engine built with Rust an
 
 ## Changelog
 ### 2026-04-03
+- CLI update: subcommand must be specified
 - ChatCompletionParameters add repeat_penalty and repeat_last_n
-- generate add penalty repeat code
+- generate add penalty repeat
 
 ### 2026-04-02
 - refactor generate code 
@@ -113,7 +114,7 @@ aha list
 aha download -m Qwen/Qwen3-ASR-0.6B
 
 # Download model and start service
-aha -m Qwen/Qwen3-ASR-0.6B
+aha cli -m Qwen/Qwen3-ASR-0.6B
 
 # Run inference directly (without starting service)
 aha run -m Qwen/Qwen3-ASR-0.6B -i "audio.wav"

README.zh-CN.md

@@ -47,6 +47,7 @@ aha 是一款基于 Rust 和 Candle 框架构建的高性能跨平台 AI 推理
 ## 更新日志
 ## Changelog
 ### 2026-04-03
+- CLI 更新: 必须指定子命令
 - ChatCompletionParameters 新增 repeat_penalty 和 repeat_last_n 参数
 - 生成添加重复惩罚代码
 
@@ -73,9 +74,6 @@ aha 是一款基于 Rust 和 Candle 框架构建的高性能跨平台 AI 推理
 - 新增 LFM2.5-1.2B-Instruct
 - 新增 LFM2-1.2B
 
-### v0.2.3 (2026-03-18)
-- 新增 DeepSeek-OCR-2
-
 **[查看完整更新日志](docs/changelog.zh-CN.md)** →
 
 
@@ -116,7 +114,7 @@ aha list
 aha download -m Qwen/Qwen3-ASR-0.6B
 
 # 下载模型并启动服务
-aha -m Qwen/Qwen3-ASR-0.6B
+aha cli -m Qwen/Qwen3-ASR-0.6B
 
 # 直接运行推理（无需启动服务）
 aha run -m Qwen/Qwen3-ASR-0.6B -i "audio.wav"

docs/changelog.md

@@ -6,8 +6,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ### 2026-04-03
+- CLI update: subcommand must be specified
 - ChatCompletionParameters add repeat_penalty and repeat_last_n
-- generate add penalty repeat code
+- generate add penalty repeat
 
 ### 2026-04-02
 - refactor generate code 

docs/changelog.zh-CN.md

@@ -6,6 +6,7 @@
 本项目遵循 [语义化版本](https://semver.org/lang/zh-CN/spec/v2.0.0.html)。
 
 ### 2026-04-03
+- CLI 更新: 必须指定子命令
 - ChatCompletionParameters 新增 repeat_penalty 和 repeat_last_n 参数
 - 生成添加重复惩罚代码
 

docs/cli.md

@@ -18,16 +18,19 @@ aha [COMMAND] [OPTIONS]
 | `--weight-path <WEIGHT_PATH>` | Local model weight path | - |
 | `--save-dir <SAVE_DIR>` | Model download save directory | ~/.aha/ |
 | `--download-retries <DOWNLOAD_RETRIES>` | Download retry count | 3 |
-| `--gguf-path <GGUF_PATH>` | Local GGUF weight | - |
+| `--gguf-path <GGUF_PATH>` | Local GGUF weight（required when using GGUF models） | - |
 | `--mmproj-path <MMPROJ_PATH>` | Local mmproj GGUF weight | - |
+| `--onnx-path <ONNX_PATH>` | Local ONNX weight（required when using ONNX models） | - |
+| `--config-path <ONNX_PATH>` | extra config path for gguf/onnx | - |
 | `-h, --help` | Display help information | - |
 | `-V, --version` | Display version number | - |
 
 ## Commands
 
-### cli - Download model and start service (default)
+### cli - Download model and start service 
 
-Download the specified model and start an HTTP service. This command is used by default when no subcommand is specified.
+Download the specified model and start an HTTP service. 
+Download only supports models in safetensors format; for GGUF/ONNX models, you must specify a local file path.
 
 **Syntax:**
 ```bash
@@ -44,8 +47,10 @@ aha cli [OPTIONS] --model <MODEL>
 | `--weight-path <WEIGHT_PATH>` | Local model weight path (skip download if specified) | - |
 | `--save-dir <SAVE_DIR>` | Model download save directory | ~/.aha/ |
 | `--download-retries <DOWNLOAD_RETRIES>` | Download retry count | 3 |
-| `--gguf-path <GGUF_PATH>` | Local GGUF weight | - |
+| `--gguf-path <GGUF_PATH>` | Local GGUF weight（required when using GGUF models） | - |
 | `--mmproj-path <MMPROJ_PATH>` | Local mmproj GGUF weight | - |
+| `--onnx-path <ONNX_PATH>` | Local ONNX weight（required when using ONNX models） | - |
+| `--config-path <ONNX_PATH>` | extra config path for gguf/onnx | - |
 
 **Examples:**
 
@@ -59,9 +64,6 @@ aha cli -m Qwen/Qwen3-VL-2B-Instruct -p 8080 --save-dir /data/models
 # Use local model (skip download)
 aha cli -m Qwen/Qwen3-VL-2B-Instruct --weight-path /path/to/model
 
-# Backward compatible way (equivalent to cli subcommand)
-aha -m Qwen/Qwen3-VL-2B-Instruct
-
 # use gguf-path and mmproj-path
 aha cli -m qwen3.5-gguf --gguf-path /path/to/xxx.gguf --mmproj-path /path/to/mmproj-xxx.gguf
 ```
@@ -72,7 +74,7 @@ Run model inference directly without starting an HTTP service. Suitable for one-
 
 **Syntax:**
 ```bash
-aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>]
+aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>] [--onnx-path <ONNX_PATH>] [--config-path <CONFIG_PATH>]
 ```
 
 **Options:**
@@ -82,9 +84,12 @@ aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-p
 | `-m, --model <MODEL>` | Model type (required) | - |
 | `-i, --input <INPUT>` | Input text or file path (model-specific interpretation, supports 1-2 parameters: input1: prompt text, input2: file path) | - |
 | `-o, --output <OUTPUT>` | Output file path (optional, auto-generated if not specified) | - |
-| `--weight-path <WEIGHT_PATH>` | Local model weight path (required when using non-GGUF models) | - |
+| `--weight-path <WEIGHT_PATH>` | Local model weight path (required when using safetensors models) | - |
 | `--gguf-path <GGUF_PATH>` | Local GGUF model weight path（required when using GGUF models） | - |
 | `--mmproj-path <MMPROJ_PATH>` | Local mmproj GGUF weight path（optional，If not specified, the module will not be loaded） | - |
+| `--onnx-path <ONNX_PATH>` | Local ONNX weight（required when using ONNX models） | - |
+| `--config-path <ONNX_PATH>` | extra config path for gguf/onnx | - |
+
 **Examples:**
 
 ```bash
@@ -129,11 +134,13 @@ aha run -m qwen3.5-gguf -i 提取图片中的文本 -i https://ai.bdstatic.com/f
 
 ### serv - Start service
 
-Start HTTP service with a model. The `--weight-path` is optional - if not specified, it defaults to `~/.aha/{model_id}`.
+Start HTTP service with a model. 
+Safetensors model: The `--weight-path` is optional - if not specified, it defaults to `~/.aha/{model_id}`.
+GGUF/ONNX model: The `--gguf-path`/ `--onnx-path` must be specified
 
 **Syntax:**
 ```bash
-aha serv [OPTIONS] --model <MODEL> [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>]
+aha serv [OPTIONS] --model <MODEL> [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>] [--onnx-path <ONNX_PATH>] [--config-path <CONFIG_PATH>]
 ```
 
 **Options:**
@@ -336,7 +343,7 @@ Example:
 
 ```bash
 # One command to download and start service
-aha -m Qwen/Qwen3-VL-2B-Instruct
+aha cli -m Qwen/Qwen3-VL-2B-Instruct
 ```
 
 ### Scenario 2: Start service with existing model
@@ -360,7 +367,7 @@ aha serv -m Qwen/Qwen3-VL-2B-Instruct --weight-path /data/models/Qwen/Qwen3-VL-2
 
 ```bash
 # Start service on 0.0.0.0:8080, allow external access
-aha -m Qwen/Qwen3-VL-2B-Instruct -a 0.0.0.0 -p 8080
+aha cli -m Qwen/Qwen3-VL-2B-Instruct -a 0.0.0.0 -p 8080
 ```
 
 ## API Endpoints
@@ -395,18 +402,6 @@ After the service starts, the following API endpoints are available:
 - **Format**: JSON response
 
 
-## Backward Compatibility
-
-To maintain compatibility with older versions, the following two usage methods are equivalent:
-
-```bash
-# New way (recommended)
-aha cli -m Qwen/Qwen3-VL-2B-Instruct
-
-# Old way (backward compatible)
-aha -m Qwen/Qwen3-VL-2B-Instruct
-```
-
 ## Notes
 
 1. **serv subcommand requires `--weight-path`**: Since the `serv` subcommand does not download models, you must specify the path to an already downloaded model via `--weight-path`.

docs/cli.zh-CN.md

@@ -15,19 +15,22 @@ aha [COMMAND] [OPTIONS]
 | `-a, --address <ADDRESS>` | 服务监听地址 | 127.0.0.1 |
 | `-p, --port <PORT>` | 服务监听端口 | 10100 |
 | `-m, --model <MODEL>` | 模型类型（必选） | - |
-| `--weight-path <WEIGHT_PATH>` | 本地模型权重路径 | - |
+| `--weight-path <WEIGHT_PATH>` | 本地safetensors模型权重路径 | - |
 | `--save-dir <SAVE_DIR>` | 模型下载保存目录 | ~/.aha/ |
 | `--download-retries <DOWNLOAD_RETRIES>` | 下载重试次数 | 3 |
-| `--gguf-path <GGUF_PATH>` | 本地 GGUF 模型权重 | - |
+| `--gguf-path <GGUF_PATH>` | 本地 GGUF 模型权重（使用GGUF模型时必选） | - |
-| `--mmproj-path <MMPROJ_PATH>` | 本地 mmproj GGUF 模型权重 | - |
+| `--mmproj-path <MMPROJ_PATH>` | 本地 mmproj GGUF 模型权重（可选，未指定则不加载该模块） | - |
+| `--onnx-path <ONNX_PATH>` | 本地 ONNX 模型权重 （使用ONNX模型时必选） | - |
+| `--config-path <ONNX_PATH>` | GGUF/ONNX 需要的额外配置路径(可选) | - |
 | `-h, --help` | 显示帮助信息 | - |
 | `-V, --version` | 显示版本号 | - |
 
 ## 子命令
 
-### cli - 下载模型并启动服务（默认）
+### cli - 下载模型并启动服务
 
-下载指定的模型并启动 HTTP 服务。当不指定子命令时，默认使用此命令。
+下载指定的模型并启动 HTTP 服务。
+下载仅支持safetensors格式模型, GGUF/ONNX模型必须指定本地文件路径
 
 **语法：**
 ```bash
@@ -41,11 +44,13 @@ aha cli [OPTIONS] --model <MODEL>
 | `-a, --address <ADDRESS>` | 服务监听地址 | 127.0.0.1 |
 | `-p, --port <PORT>` | 服务监听端口 | 10100 |
 | `-m, --model <MODEL>` | 模型类型（必选） | - |
-| `--weight-path <WEIGHT_PATH>` | 本地模型权重路径（如指定则跳过下载） | - |
+| `--weight-path <WEIGHT_PATH>` | 本地safetensors模型权重路径（如指定则跳过下载） | - |
 | `--save-dir <SAVE_DIR>` | 模型下载保存目录 | ~/.aha/ |
 | `--download-retries <DOWNLOAD_RETRIES>` | 下载重试次数 | 3 |
-| `--gguf-path <GGUF_PATH>` | 本地 GGUF 模型权重 | - |
+| `--gguf-path <GGUF_PATH>` | 本地 GGUF 模型权重（使用GGUF模型时必选） | - |
-| `--mmproj-path <MMPROJ_PATH>` | 本地 mmproj GGUF 模型权重 | - |
+| `--mmproj-path <MMPROJ_PATH>` | 本地 mmproj GGUF 模型权重（可选，未指定则不加载该模块） | - |
+| `--onnx-path <ONNX_PATH>` | 本地 ONNX 模型权重 （使用ONNX模型时必选） | - |
+| `--config-path <ONNX_PATH>` | GGUF/ONNX 需要的额外配置路径(可选) | - |
 
 **示例：**
 
@@ -59,9 +64,6 @@ aha cli -m Qwen/Qwen3-VL-2B-Instruct -p 8080 --save-dir /data/models
 # 使用本地模型（不下载）
 aha cli -m Qwen/Qwen3-VL-2B-Instruct --weight-path /path/to/model
 
-# 向后兼容方式（等同于 cli 子命令）
-aha -m Qwen/Qwen3-VL-2B-Instruct
-
 # 指定gguf-path和mmproj-path
 aha cli -m qwen3.5-gguf --gguf-path /path/to/xxx.gguf --mmproj-path /path/to/mmproj-xxx.gguf
 ```
@@ -72,7 +74,7 @@ aha cli -m qwen3.5-gguf --gguf-path /path/to/xxx.gguf --mmproj-path /path/to/mmp
 
 **语法：**
 ```bash
-aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>]
+aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-path <WEIGHT_PATH>] [--gguf-path <GGUF_PATH>] [--mmproj-path <MMPROJ_PATH>] [--onnx-path <ONNX_PATH>] [--config-path <CONFIG_PATH>]
 ```
 
 **选项：**
@@ -82,9 +84,11 @@ aha run [OPTIONS] --model <MODEL> --input <INPUT> [--input <INPUT2>] [--weight-p
 | `-m, --model <MODEL>` | 模型类型（必选） | - |
 | `-i, --input <INPUT>` | 输入文本或文件路径（模型特定解释，支持1-2个参数, input1： 提示文本, input2: 文件地址） | - |
 | `-o, --output <OUTPUT>` | 输出文件路径（可选，未指定则自动生成） | - |
-| `--weight-path <WEIGHT_PATH>` | 本地模型权重路径（使用非GGUF模型时必选） | - |
+| `--weight-path <WEIGHT_PATH>` | 本地模型权重路径（使用Safetensors模型时必选） | - |
 | `--gguf-path <GGUF_PATH>` | 本地GGUF模型权重路径（使用GGUF模型时必选） | - |
 | `--mmproj-path <MMPROJ_PATH>` | 本地mmproj GGUF模型权重路径（可选，未指定则不加载该模块） | - |
+| `--onnx-path <ONNX_PATH>` | 本地 ONNX 模型权重 （使用ONNX模型时必选） | - |
+| `--config-path <ONNX_PATH>` | GGUF/ONNX 需要的额外配置路径(可选) | - |
 
 **示例：**
 
@@ -129,7 +133,9 @@ aha run -m qwen3.5-gguf -i 提取图片中的文本 -i https://ai.bdstatic.com/f
 
 ### serv - 启动服务
 
-使用指定模型启动 HTTP 服务。`--weight-path` 是可选的 - 如果不指定，默认使用 `~/.aha/{model_id}`。
+使用指定模型启动 HTTP 服务。
+safetensors模型`--weight-path` 是可选的 - 如果不指定，默认使用 `~/.aha/{model_id}`。
+GGUF/ONNX模型必须指定本地文件路径
 
 **语法：**
 ```bash
@@ -147,6 +153,8 @@ aha serv [OPTIONS] --model <MODEL> [--weight-path <WEIGHT_PATH>] [--gguf-path <G
 | `--allow-remote-shutdown` | 允许远程关机请求（不推荐） | false |
 | `--gguf-path <GGUF_PATH>` | 本地GGUF模型权重路径（使用GGUF模型时必选） | - |
 | `--mmproj-path <MMPROJ_PATH>` | 本地mmproj GGUF模型权重路径（可选，未指定则不加载该模块） | - |
+| `--onnx-path <ONNX_PATH>` | 本地 ONNX 模型权重 （使用ONNX模型时必选） | - |
+| `--config-path <ONNX_PATH>` | GGUF/ONNX 需要的额外配置路径(可选) | - |
 
 **示例：**
 
@@ -336,7 +344,7 @@ aha list -j
 
 ```bash
 # 一条命令下载并启动服务
-aha -m Qwen/Qwen3-VL-2B-Instruct
+aha cli -m Qwen/Qwen3-VL-2B-Instruct
 ```
 
 ### 场景 2：使用已有模型启动服务
@@ -360,7 +368,7 @@ aha serv -m Qwen/Qwen3-VL-2B-Instruct --weight-path /data/models/Qwen/Qwen3-VL-2
 
 ```bash
 # 在 0.0.0.0:8080 启动服务，允许外部访问
-aha -m Qwen/Qwen3-VL-2B-Instruct -a 0.0.0.0 -p 8080
+aha cli -m Qwen/Qwen3-VL-2B-Instruct -a 0.0.0.0 -p 8080
 ```
 
 ## API 接口
@@ -394,17 +402,6 @@ aha -m Qwen/Qwen3-VL-2B-Instruct -a 0.0.0.0 -p 8080
 - **安全性**: 默认仅允许本地访问，使用 `--allow-remote-shutdown` 标志启用远程访问（不推荐）
 - **格式**: JSON 响应
 
-## 向后兼容性
-
-为了保持与旧版本的兼容性，以下两种使用方式是等效的：
-
-```bash
-# 新方式（推荐）
-aha cli -m Qwen/Qwen3-VL-2B-Instruct
-
-# 旧方式（向后兼容）
-aha -m Qwen/Qwen3-VL-2B-Instruct
-```
 
 ## 注意事项
 

docs/installation.md

@@ -291,7 +291,8 @@ OpenBMB/MiniCPM4-0.5B                    OpenBMB              llm          ✔
 Qwen/Qwen2.5-VL-3B-Instruct              Qwen                 vlm          ✔       
 Qwen/Qwen2.5-VL-7B-Instruct              Qwen                 vlm                  
 Qwen/Qwen3-0.6B                          Qwen                 llm          ✔  
-Qwen/Qwen3-1.7B                          Qwen                 llm          ✔     
+Qwen/Qwen3-1.7B                          Qwen                 llm               
+Qwen/Qwen3-4B                            Qwen                 llm            
 Qwen/Qwen3.5-0.8B                        Qwen                 vlm          ✔       
 Qwen/Qwen3.5-2B                          Qwen                 vlm                  
 Qwen/Qwen3.5-4B                          Qwen                 vlm                  

docs/installation.zh-CN.md

@@ -291,7 +291,8 @@ OpenBMB/MiniCPM4-0.5B                    OpenBMB              llm          ✔
 Qwen/Qwen2.5-VL-3B-Instruct              Qwen                 vlm          ✔       
 Qwen/Qwen2.5-VL-7B-Instruct              Qwen                 vlm                  
 Qwen/Qwen3-0.6B                          Qwen                 llm          ✔ 
-Qwen/Qwen3-1.7B                          Qwen                 llm          ✔      
+Qwen/Qwen3-1.7B                          Qwen                 llm          
+Qwen/Qwen3-4B                            Qwen                 llm                
 Qwen/Qwen3.5-0.8B                        Qwen                 vlm          ✔       
 Qwen/Qwen3.5-2B                          Qwen                 vlm                  
 Qwen/Qwen3.5-4B                          Qwen                 vlm                  

docs/supported-models.md

@@ -15,7 +15,8 @@ OpenBMB/MiniCPM4-0.5B                    OpenBMB              llm          ✔
 Qwen/Qwen2.5-VL-3B-Instruct              Qwen                 vlm          ✔       
 Qwen/Qwen2.5-VL-7B-Instruct              Qwen                 vlm                  
 Qwen/Qwen3-0.6B                          Qwen                 llm          ✔
-Qwen/Qwen3-1.7B                          Qwen                 llm          ✔       
+Qwen/Qwen3-1.7B                          Qwen                 llm          
+Qwen/Qwen3-4B                            Qwen                 llm                 
 Qwen/Qwen3.5-0.8B                        Qwen                 vlm          ✔       
 Qwen/Qwen3.5-2B                          Qwen                 vlm                  
 Qwen/Qwen3.5-4B                          Qwen                 vlm                  
@@ -44,7 +45,7 @@ ZhipuAI/GLM-OCR                          ZhipuAI              ocr          ✔
 
 | Model | Parameters | Model Id | License |
 |-------|-----------|--------|---------|
-| **Qwen3-0.6B** | 0.6B | Qwen/Qwen3-0.6B <br> Qwen/Qwen3-1.7B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
+| **Qwen3-0.6B** | 0.6B | Qwen/Qwen3-0.6B <br> Qwen/Qwen3-1.7B <br> Qwen/Qwen3-4B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
 | **MiniCPM4-0.5B** | 0.5B | OpenBMB/MiniCPM4-0.5B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
 | **LFM2-1.2B** | 1.2B | LiquidAI/LFM2-1.2B | [lfm1.0](https://huggingface.co/LiquidAI/LFM2-1.2B/blob/main/LICENSE) |
 | **LFM2.5-1.2B-Instruct** | 1.2B | LiquidAI/LFM2.5-1.2B-Instruct | [lfm1.0](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/main/LICENSE) |

docs/supported-models.zh-CN.md

@@ -15,7 +15,8 @@ OpenBMB/MiniCPM4-0.5B                    OpenBMB              llm          ✔
 Qwen/Qwen2.5-VL-3B-Instruct              Qwen                 vlm          ✔       
 Qwen/Qwen2.5-VL-7B-Instruct              Qwen                 vlm                  
 Qwen/Qwen3-0.6B                          Qwen                 llm          ✔
-Qwen/Qwen3-1.7B                          Qwen                 llm          ✔       
+Qwen/Qwen3-1.7B                          Qwen                 llm                 
+Qwen/Qwen3-4B                            Qwen                 llm          
 Qwen/Qwen3.5-0.8B                        Qwen                 vlm          ✔       
 Qwen/Qwen3.5-2B                          Qwen                 vlm                  
 Qwen/Qwen3.5-4B                          Qwen                 vlm                  
@@ -44,7 +45,7 @@ ZhipuAI/GLM-OCR                          ZhipuAI              ocr          ✔
 
 | 模型 | 参数量 | 模型id | 开源协议 |
 |------|--------|------|---------|
-| **Qwen3-0.6B** | 0.6B | Qwen/Qwen3-0.6B <br> Qwen/Qwen3-1.7B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
+| **Qwen3-0.6B** | 0.6B | Qwen/Qwen3-0.6B <br> Qwen/Qwen3-1.7B <br> Qwen/Qwen3-4B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
 | **MiniCPM4-0.5B** | 0.5B | OpenBMB/MiniCPM4-0.5B | [Apache 2.0](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/apache-2.0.md) |
 | **LFM2-1.2B** | 1.2B | LiquidAI/LFM2-1.2B | [lfm1.0](https://huggingface.co/LiquidAI/LFM2-1.2B/blob/main/LICENSE) |
 | **LFM2.5-1.2B-Instruct** | 1.2B | LiquidAI/LFM2.5-1.2B-Instruct | [lfm1.0](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/main/LICENSE) |

src/cli/args.rs

@@ -5,40 +5,8 @@ use clap::{Args, Parser, Subcommand};
 #[command(name = "aha")]
 #[command(version, about, long_about = None)]
 pub(crate) struct Cli {
-    /// Service listen address
-    #[arg(short, long, default_value = "127.0.0.1")]
-    pub address: Option<String>,
-
-    /// Service listen port
-    #[arg(short, long)]
-    pub port: Option<u16>,
-
-    /// Model type (required for backward compatibility)
-    #[arg(short, long)]
-    pub model: Option<WhichModel>,
-
-    /// Local model weight path
-    #[arg(long)]
-    pub weight_path: Option<String>,
-
-    /// Model download save directory
-    #[arg(long)]
-    pub save_dir: Option<String>,
-
-    /// Download retry count
-    #[arg(long)]
-    pub download_retries: Option<u32>,
-
-    /// Local GGUF model weight path (required for loading models with GGUF).
-    #[arg(long)]
-    pub gguf_path: Option<String>,
-
-    /// Local path for mmproj GGUF model weights (required for loading with multimodel GGUF)
-    #[arg(long)]
-    pub mmproj_path: Option<String>,
-
     #[command(subcommand)]
-    pub command: Option<Commands>,
+    pub command: Commands,
 }
 
 #[derive(Subcommand, Debug)]
@@ -61,7 +29,7 @@ pub(crate) enum Commands {
 
 /// Common/shared arguments for server operations
 #[derive(Args, Debug)]
-pub(crate) struct CommonArgs {
+pub(crate) struct ServerCommonArgs {
     /// Service listen address
     #[arg(short, long, default_value = "127.0.0.1")]
     pub address: String,
@@ -70,24 +38,42 @@ pub(crate) struct CommonArgs {
     #[arg(short, long, default_value_t = 10100)]
     pub port: u16,
 
-    /// Model type (required)
-    #[arg(short, long)]
-    pub model: WhichModel,
-
     /// Allow remote shutdown requests (default: local only, use with caution)
     #[arg(long)]
     pub allow_remote_shutdown: bool,
 }
 
-/// Arguments for the 'cli' subcommand (download + serve)
 #[derive(Args, Debug)]
-pub(crate) struct CliArgs {
+pub(crate) struct PathCommonArgs {
-    #[command(flatten)]
-    pub common: CommonArgs,
-
     /// Local model weight path (skip download if provided)
     #[arg(long)]
     pub weight_path: Option<String>,
+    /// Local GGUF model weight path (required for loading models with GGUF).
+    #[arg(long)]
+    pub gguf_path: Option<String>,
+
+    /// Local path for mmproj GGUF model weights (required for loading with multimodel GGUF)
+    #[arg(long)]
+    pub mmproj_path: Option<String>,
+
+    /// Local path for onnx model weights (required for loading with onnx)
+    #[arg(long)]
+    pub onnx_path: Option<String>,
+
+    /// config path for onnx/gguf model need extra config file
+    #[arg(long)]
+    pub config_path: Option<String>,
+}
+
+/// Arguments for the 'cli' subcommand (download + serve)
+#[derive(Args, Debug)]
+pub(crate) struct CliArgs {
+    /// Model type (required)
+    #[arg(short, long)]
+    pub model: WhichModel,
+
+    #[command(flatten)]
+    pub server_common: ServerCommonArgs,
 
     /// Model download save directory
     #[arg(long)]
@@ -97,32 +83,22 @@ pub(crate) struct CliArgs {
     #[arg(long)]
     pub download_retries: Option<u32>,
 
-    /// Local GGUF model weight path (required for loading models with GGUF).
+    #[command(flatten)]
-    #[arg(long)]
+    pub path_common: PathCommonArgs,
-    pub gguf_path: Option<String>,
-
-    /// Local path for mmproj GGUF model weights (required for loading with multimodel GGUF)
-    #[arg(long)]
-    pub mmproj_path: Option<String>,
 }
 
 /// Arguments for the 'serv start' subcommand
 #[derive(Args, Debug)]
 pub(crate) struct ServArgs {
+    /// Model type (required)
+    #[arg(short, long)]
+    pub model: WhichModel,
+
     #[command(flatten)]
-    pub common: CommonArgs,
+    pub server_common: ServerCommonArgs,
 
-    /// Local model weight path (defaults to ~/.aha/{model_id} if not specified)
+    #[command(flatten)]
-    #[arg(long)]
+    pub path_common: PathCommonArgs,
-    pub weight_path: Option<String>,
-
-    /// Local GGUF model weight path (required for loading models with GGUF).
-    #[arg(long)]
-    pub gguf_path: Option<String>,
-
-    /// Local path for mmproj GGUF model weights (required for loading with multimodel GGUF)
-    #[arg(long)]
-    pub mmproj_path: Option<String>,
 }
 
 /// Arguments for the 'serv list' subcommand
@@ -164,17 +140,8 @@ pub(crate) struct RunArgs {
     #[arg(short, long)]
     pub output: Option<String>,
 
-    /// Local model weight path (defaults to ~/.aha/{model_id} if not specified)
+    #[command(flatten)]
-    #[arg(long)]
+    pub path_common: PathCommonArgs,
-    pub weight_path: Option<String>,
-
-    /// Local GGUF model weight path (required for loading models with GGUF).
-    #[arg(long)]
-    pub gguf_path: Option<String>,
-
-    /// Local path for mmproj GGUF model weights (required for loading with multimodel GGUF)
-    #[arg(long)]
-    pub mmproj_path: Option<String>,
 }
 
 /// Arguments for the 'delete' subcommand (delete model from default location)

src/cli/mod.rs

@@ -76,22 +76,27 @@ pub(crate) fn run_list(args: ListArgs) -> anyhow::Result<()> {
 /// Run the 'cli' subcommand: download model (if needed) and start service
 pub(crate) async fn run_cli(args: CliArgs) -> anyhow::Result<()> {
     let CliArgs {
-        common,
+        model,
-        weight_path,
+        server_common,
         save_dir,
         download_retries,
-        gguf_path,
+        path_common,
-        mmproj_path,
     } = args;
-    let model_id = common.model.as_string();
+    let model_id = model.as_string();
 
-    let (model_path, gguf, mmproj) = if model_id.contains("gguf") {
+    let (model_path, gguf, mmproj) = if model.is_gguf() {
-        if gguf_path.is_none() {
+        if path_common.gguf_path.is_none() {
             return Err(anyhow!("gguf model path is required"));
         }
-        ("GGUF".to_string(), gguf_path, mmproj_path)
+        (
+            "GGUF".to_string(),
+            path_common.gguf_path,
+            path_common.mmproj_path,
+        )
+    } else if model.is_onnx() {
+        return Err(anyhow!("onnx model not support now"));
     } else {
-        let model_path = match weight_path {
+        let model_path = match path_common.weight_path {
             Some(path) => path,
             None => {
                 let save_dir = match save_dir {
@@ -106,8 +111,13 @@ pub(crate) async fn run_cli(args: CliArgs) -> anyhow::Result<()> {
         (model_path, None, None)
     };
 
-    init(common.model, model_path, gguf, mmproj)?;
+    init(model, model_path, gguf, mmproj)?;
-    start_http_server(common.address, common.port, common.allow_remote_shutdown).await?;
+    start_http_server(
+        server_common.address,
+        server_common.port,
+        server_common.allow_remote_shutdown,
+    )
+    .await?;
 
     Ok(())
 }
@@ -115,27 +125,41 @@ pub(crate) async fn run_cli(args: CliArgs) -> anyhow::Result<()> {
 /// Run the 'serv' subcommand: start service only (no download)
 pub(crate) async fn run_serv(args: ServArgs) -> anyhow::Result<()> {
     let ServArgs {
-        common,
+        model,
-        weight_path,
+        server_common,
-        gguf_path,
+        path_common,
-        mmproj_path,
     } = args;
-    let model_id = common.model.as_string();
+    let (model_path, gguf, mmproj) = if model.is_gguf() {
-    let (model_path, gguf, mmproj) = if model_id.contains("gguf") {
+        if path_common.gguf_path.is_none() {
-        if gguf_path.is_none() {
             return Err(anyhow!("gguf model path is required"));
         }
-        ("GGUF".to_string(), gguf_path, mmproj_path)
+        (
+            "GGUF".to_string(),
+            path_common.gguf_path,
+            path_common.mmproj_path,
+        )
+    } else if model.is_onnx() {
+        return Err(anyhow!("onnx model not support now"));
     } else {
-        let model_path = match weight_path {
+        let model_path = match path_common.weight_path {
             Some(path) => path,
-            None => get_default_weight_path(common.model),
+            None => get_default_weight_path(model),
         };
+        if !std::path::Path::new(&model_path).exists() {
+            return Err(anyhow!(
+                "serv subcommand will not download model, use `weight-path` to pass the model path"
+            ));
+        }
         (model_path, None, None)
     };
 
-    init(common.model, model_path, gguf, mmproj)?;
+    init(model, model_path, gguf, mmproj)?;
-    start_http_server(common.address, common.port, common.allow_remote_shutdown).await?;
+    start_http_server(
+        server_common.address,
+        server_common.port,
+        server_common.allow_remote_shutdown,
+    )
+    .await?;
 
     Ok(())
 }
@@ -205,13 +229,11 @@ pub(crate) fn run_run(args: RunArgs) -> anyhow::Result<()> {
         model,
         input,
         output,
-        weight_path,
+        path_common,
-        gguf_path,
-        mmproj_path,
     } = args;
 
     // Use default weight path if not specified
-    let weight_path = match weight_path {
+    let weight_path = match path_common.weight_path {
         Some(path) => path,
         None => get_default_weight_path(model),
     };
@@ -243,6 +265,9 @@ pub(crate) fn run_run(args: RunArgs) -> anyhow::Result<()> {
         WhichModel::Qwen3_1_7B => {
             qwen3::Qwen3Exec::run(&input, output.as_deref(), &weight_path)?;
         }
+        WhichModel::Qwen3_4B => {
+            qwen3::Qwen3Exec::run(&input, output.as_deref(), &weight_path)?;
+        }
         WhichModel::Qwen3_5_0_8B => {
             qwen3_5::Qwen3_5Exec::run(&input, output.as_deref(), &weight_path)?;
         }
@@ -256,7 +281,12 @@ pub(crate) fn run_run(args: RunArgs) -> anyhow::Result<()> {
             qwen3_5::Qwen3_5Exec::run(&input, output.as_deref(), &weight_path)?;
         }
         WhichModel::Qwen3_5Gguf => {
-            qwen3_5::Qwen3_5Exec::run_gguf(&input, output.as_deref(), gguf_path, mmproj_path)?;
+            qwen3_5::Qwen3_5Exec::run_gguf(
+                &input,
+                output.as_deref(),
+                path_common.gguf_path,
+                path_common.mmproj_path,
+            )?;
         }
         WhichModel::Qwen3ASR0_6B => {
             qwen3_asr::Qwen3ASRExec::run(&input, output.as_deref(), &weight_path)?;

src/main.rs

@@ -1,7 +1,7 @@
 use clap::Parser;
 
 use crate::cli::{
-    args::{Cli, CliArgs, Commands, CommonArgs},
+    args::{Cli, Commands},
     run_cli, run_delete, run_download, run_list, run_ps, run_run, run_serv,
 };
 
@@ -11,32 +11,13 @@ mod server;
 #[tokio::main]
 async fn main() -> anyhow::Result<()> {
     let cli = Cli::parse();
-
     match cli.command {
-        Some(Commands::Cli(args)) => run_cli(args).await,
+        Commands::Cli(args) => run_cli(args).await,
-        Some(Commands::Serv(args)) => run_serv(args).await,
+        Commands::Serv(args) => run_serv(args).await,
-        Some(Commands::Ps(args)) => run_ps(args),
+        Commands::Ps(args) => run_ps(args),
-        Some(Commands::Delete(args)) => run_delete(args),
+        Commands::Delete(args) => run_delete(args),
-        Some(Commands::Download(args)) => run_download(args).await,
+        Commands::Download(args) => run_download(args).await,
-        Some(Commands::Run(args)) => run_run(args),
+        Commands::Run(args) => run_run(args),
-        Some(Commands::List(args)) => run_list(args),
+        Commands::List(args) => run_list(args),
-        None => {
-            // Backward compatibility: when no subcommand is provided, use 'cli' behavior
-            let model = cli.model.expect("Model is required (use -m or --model)");
-            let args = CliArgs {
-                common: CommonArgs {
-                    address: cli.address.unwrap_or_else(|| "127.0.0.1".to_string()),
-                    port: cli.port.unwrap_or(10100),
-                    model,
-                    allow_remote_shutdown: false,
-                },
-                weight_path: cli.weight_path,
-                save_dir: cli.save_dir,
-                download_retries: cli.download_retries,
-                gguf_path: cli.gguf_path,
-                mmproj_path: cli.mmproj_path,
-            };
-            run_cli(args).await
-        }
     }
 }

src/models/common/model_mapping.rs

@@ -20,6 +20,8 @@ pub enum WhichModel {
     Qwen3_0_6B,
     #[value(name = "Qwen/Qwen3-1.7B")]
     Qwen3_1_7B,
+    #[value(name = "Qwen/Qwen3-4B")]
+    Qwen3_4B,
     #[value(name = "Qwen/Qwen3.5-0.8B")]
     Qwen3_5_0_8B,
     #[value(name = "Qwen/Qwen3.5-2B")]
@@ -67,18 +69,40 @@ pub enum WhichModel {
 }
 
 impl WhichModel {
-    /// Get the ModelScope model ID for this model variant
+    /// Get the model ID for this model variant
     pub fn as_string(&self) -> String {
         self.to_possible_value()
             .expect("not exists")
             .get_name()
             .to_string()
     }
+
+    /// Checks if the model is in GGUF format
+    ///
+    /// Returns true if the model ID contains "gguf", false otherwise
+    pub fn is_gguf(&self) -> bool {
+        let model_id = self.as_string();
+        model_id.to_lowercase().contains("gguf")
+    }
+
+    /// Checks if the model is in ONNX format
+    ///
+    /// Returns true if the model ID contains "onnx", false otherwise
+    pub fn is_onnx(&self) -> bool {
+        let model_id = self.as_string();
+        model_id.to_lowercase().contains("onnx")
+    }
+
     /// Get the WhichModel enum list
     pub fn model_list() -> Vec<Self> {
         WhichModel::value_variants().to_vec()
     }
 
+    /// Extracts the model owner/organization from the model ID
+    ///
+    /// Splits the model ID string on '/' and returns the first part which typically represents
+    /// the organization or user who owns the model in Hugging Face format (e.g., "Qwen" from "Qwen/Qwen3-0.6B")
+    /// Returns "none" if the model ID doesn't contain a '/' separator
     pub fn model_owner(&self) -> String {
         let name = self.as_string();
         let names: Vec<&str> = name.split("/").collect();
@@ -96,6 +120,7 @@ impl WhichModel {
             WhichModel::MiniCPM4_0_5B
             | WhichModel::Qwen3_0_6B
             | WhichModel::Qwen3_1_7B
+            | WhichModel::Qwen3_4B
             | WhichModel::LFM2_1_2B
             | WhichModel::LFM2_5_1_2BInstruct => "llm",
             // VLM models

src/models/mod.rs

@@ -129,12 +129,27 @@ impl<'a> GenerateModel for ModelInstance<'a> {
     }
 }
 
-pub fn load_model<'a>(
+#[allow(unused)]
+pub fn load_gguf_model<'a>(
     model_type: WhichModel,
-    path: &str,
+    config_path: Option<&str>, // 有些gguf未包含模型其他配置，需额外指定
-    gguf: Option<&str>,
+    gguf_path: &str,
-    mmproj: Option<&str>,
+    mmproj_path: Option<&str>,
 ) -> Result<ModelInstance<'a>> {
+    let model = match model_type {
+        WhichModel::Qwen3_5Gguf => {
+            let model = Qwen3_5GenerateModel::init_from_gguf(gguf_path, mmproj_path, None)?;
+            ModelInstance::Qwen3_5(model)
+        }
+        _ => {
+            let model_id = model_type.as_string();
+            return Err(anyhow!("model id {model_id} is not gguf model"));
+        }
+    };
+    Ok(model)
+}
+
+pub fn load_model<'a>(model_type: WhichModel, path: &str) -> Result<ModelInstance<'a>> {
     let model = match model_type {
         WhichModel::MiniCPM4_0_5B => {
             let model = MiniCPMGenerateModel::init(path, None, None)?;
@@ -172,6 +187,10 @@ pub fn load_model<'a>(
             let model = Qwen3GenerateModel::init(path, None, None)?;
             ModelInstance::Qwen3(model)
         }
+        WhichModel::Qwen3_4B => {
+            let model = Qwen3GenerateModel::init(path, None, None)?;
+            ModelInstance::Qwen3(model)
+        }
         WhichModel::Qwen3_5_0_8B => {
             let model = Qwen3_5GenerateModel::init(path, None, None)?;
             ModelInstance::Qwen3_5(model)
@@ -188,14 +207,6 @@ pub fn load_model<'a>(
             let model = Qwen3_5GenerateModel::init(path, None, None)?;
             ModelInstance::Qwen3_5(model)
         }
-        WhichModel::Qwen3_5Gguf => {
-            if gguf.is_none() {
-                return Err(anyhow!("Qwen3_5Gguf gguf model path is required"));
-            }
-            let gguf = gguf.unwrap();
-            let model = Qwen3_5GenerateModel::init_from_gguf(gguf, mmproj, None)?;
-            ModelInstance::Qwen3_5(model)
-        }
         WhichModel::Qwen3ASR0_6B => {
             let model = Qwen3AsrGenerateModel::init(path, None, None)?;
             ModelInstance::Qwen3ASR(model)
@@ -264,6 +275,10 @@ pub fn load_model<'a>(
             let model = GlmOcrGenerateModel::init(path, None, None)?;
             ModelInstance::GlmOCR(model)
         }
+        _ => {
+            let model_id = model_type.as_string();
+            return Err(anyhow!("model id {model_id} is not safetensor model"));
+        }
     };
     Ok(model)
 }

src/models/qwen3_5/generate.rs

@@ -59,7 +59,7 @@ impl<'a> Qwen3_5GenerateModel<'a> {
             qwen3_5,
             device,
             model_name: model_name.to_string(),
-            repeat_penalty: 1.01,
+            repeat_penalty: 1.0,
             repeat_last_n: 64,
         })
     }
@@ -116,7 +116,7 @@ impl<'a> Qwen3_5GenerateModel<'a> {
             qwen3_5,
             device,
             model_name: stem.to_string(),
-            repeat_penalty: 1.1,
+            repeat_penalty: 1.2,
             repeat_last_n: 64,
         })
     }

src/server/api.rs

@@ -2,9 +2,11 @@ use std::pin::pin;
 use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::{Arc, OnceLock};
 
+use aha::models::load_gguf_model;
 use aha::models::{GenerateModel, ModelInstance, common::model_mapping::WhichModel, load_model};
 use aha::params::chat::ChatCompletionParameters;
 use aha::utils::string_to_static_str;
+use anyhow::anyhow;
 use rocket::futures::StreamExt;
 use rocket::serde::{Serialize, json::Json};
 use rocket::{
@@ -37,10 +39,21 @@ pub fn init(
     gguf: Option<String>,
     mmproj: Option<String>,
 ) -> anyhow::Result<()> {
-    let model_path = string_to_static_str(path);
+    let model = if model_type.is_gguf() {
-    let gguf = gguf.map(string_to_static_str);
+        if let Some(gguf_path) = gguf {
-    let mmproj = mmproj.map(string_to_static_str);
+            let gguf_path = string_to_static_str(gguf_path);
-    let model = load_model(model_type, model_path, gguf, mmproj)?;
+            let mmproj_path = mmproj.map(string_to_static_str);
+            load_gguf_model(model_type, None, gguf_path, mmproj_path)?
+        } else {
+            return Err(anyhow!("gguf model need gguf model path"));
+        }
+    } else if model_type.is_onnx() {
+        return Err(anyhow!("onnx comming soon but now not support"));
+    } else {
+        let model_path = string_to_static_str(path);
+        load_model(model_type, model_path)?
+    };
+
     MODEL.get_or_init(|| {
         Arc::new(RwLock::new(StoredModel {
             which_model: model_type,