59 lines
1.6 KiB
Markdown
59 lines
1.6 KiB
Markdown
---
|
||
last_reviewed: 2026-06-09
|
||
---
|
||
|
||
# LlamaIndex 多模态消息格式错误
|
||
|
||
## 错误现象
|
||
|
||
```
|
||
pydantic_core._pydantic_core.ValidationError: 2 validation errors for ChatMessage
|
||
blocks.0
|
||
Unable to extract tag using discriminator 'block_type' [type=union_tag_not_found, ...
|
||
blocks.1
|
||
Unable to extract tag using discriminator 'block_type' [type=union_tag_not_found, ...
|
||
```
|
||
|
||
## 触发条件
|
||
|
||
- llama-index-core >= 0.14.x
|
||
- 使用 `ChatMessage` 构造多模态消息（文本 + 图片）
|
||
- 传入 OpenAI 格式的 `content` 列表：`[{"type": "text", ...}, {"type": "image_url", ...}]`
|
||
|
||
## 原因
|
||
|
||
`llama-index-core 0.14.x` 重构了 `ChatMessage` 的内部结构：
|
||
|
||
| 版本 | 字段 | 多模态 content 格式 |
|
||
|------|------|-------------------|
|
||
| 0.14.x | `role`, `blocks`, `additional_kwargs` | `TextBlock` / `ImageBlock` 实例 |
|
||
| 旧版本 | `role`, `content` | OpenAI 风格字典列表 |
|
||
|
||
底层 Pydantic 模型使用 `block_type` 作为 union discriminator，OpenAI 格式的 `{"type": "text", ...}` 字典不包含该字段，导致验证失败。
|
||
|
||
## 修复方法
|
||
|
||
**正确写法（llama-index 原生 blocks 格式）：**
|
||
|
||
```python
|
||
from llama_index.core.llms import ChatMessage
|
||
from llama_index.core.base.llms.types import ImageBlock, TextBlock
|
||
|
||
messages = [
|
||
ChatMessage(role="system", content=system_prompt),
|
||
ChatMessage(
|
||
role="user",
|
||
blocks=[
|
||
TextBlock(text="请分析这张图片"),
|
||
ImageBlock(
|
||
url=f"data:image/jpeg;base64,{image_b64}",
|
||
detail="high",
|
||
),
|
||
],
|
||
),
|
||
]
|
||
```
|
||
## 适用版本
|
||
|
||
- llama-index-core: 0.14.22
|
||
- llama-index-llms-openai-like: 0.7.2 |