Spaces:
Running
on
Zero
Running
on
Zero
File size: 2,737 Bytes
886d8e9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
---
title: LMC Messages
---
To support the incoming `L`anguage `M`odel `C`omputer architecture, we extend OpenAI's messages format to include additional information, and a new role called `computer`:
```python
# The user sends a message.
{"role": "user", "type": "message", "content": "What's 2380*3875?"}
# The assistant runs some code.
{"role": "assistant", "type": "code", "format": "python", "content": "2380*3875"}
# The computer responds with the result of the code.
{"role": "computer", "type": "console", "format": "output", "content": "9222500"}
# The assistant sends a message.
{"role": "assistant", "type": "message", "content": "The result of multiplying 2380 by 3875 is 9222500."}
```
## Anatomy
Each message in the LMC architecture has the following parameters (`format` is only present for some types):
```
{
"role": "<role>", # Who is sending the message.
"type": "<type>", # What kind of message is being sent.
"format": "<format>" # Some types need to be further specified, so they optionally use this parameter.
"content": "<content>", # What the message says.
}
```
Parameter|Description|
---|---|
`role`|The sender of the message.|
`type`|The kind of message being sent.|
`content`|The actual content of the message.|
`format`|The format of the content (optional).|
## Roles
Role|Description|
---|---|
`user`|The individual interacting with the system.|
`assistant`|The language model.|
`computer`|The system that executes the language model's commands.|
## Possible Message Types / Formats
Any role can produce any of the following formats, but we've included a `Common Roles` column to give you a sense of the message type's usage.
Type|Format|Content Description|Common Roles
---|---|---|---|
message|None|A text-only message.|`user`, `assistant`|
console|active_line|The active line of code (from the most recent code block) that's executing.|`computer`|
console|output|Text output resulting from `print()` statements in Python, `console.log()` statements in Javascript, etc. **This includes errors.**|`computer`|
image|base64|A `base64` image in PNG format (default)|`user`, `computer`|
image|base64.png|A `base64` image in PNG format|`user`, `computer`|
image|base64.jpeg|A `base64` image in JPEG format|`user`, `computer`|
image|path|A path to an image.|`user`, `computer`|
code|html|HTML code that should be executed.|`assistant`, `computer`|
code|javascript|JavaScript code that should be executed.|`assistant`, `computer`|
code|python|Python code that should be executed.|`assistant`|
code|r|R code that should be executed.|`assistant`|
code|applescript|AppleScript code that should be executed.|`assistant`|
code|shell|Shell code that should be executed.|`assistant`|
|