fix(opencode): add input limit for compaction #8465
Conversation
|
Hey! Your PR title Please update it to start with one of:
Where See CONTRIBUTING.md for details. |
|
The following comment was made by an LLM, it may be inaccurate: No duplicate PRs found |
brandon93s
changed the title
There was a problem hiding this comment.
Pull request overview
This PR attempts to address the ChatGPT Pro plan's lower context limit compared to the API limit by adding an input field to the model's limit configuration for the gpt-5.2-codex model.
Changes:
- Added
input: 272000to the limit object for the gpt-5.2-codex model configuration
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
| const count = input.tokens.input + input.tokens.cache.read + input.tokens.output | ||
| const output = Math.min(input.model.limit.output, SessionPrompt.OUTPUT_TOKEN_MAX) || SessionPrompt.OUTPUT_TOKEN_MAX | ||
| const usable = context - output | ||
| const usable = input.model.limit.input || context - output |
There was a problem hiding this comment.
If an input limit is provided, it's always the usable limit. This can be less than context - output.
|
Auto-compaction now triggers based on the optional input limit. There is a separate issue (I believe with all models) where the next outgoing message can overrun context/input limits. In this scenario, the error should be caught with an auto-compaction running on thread history, and then a retry of the latest message occurring. Perhaps integrated into |
| delete provider.models[modelId] | ||
| } | ||
| } | ||
|
|
There was a problem hiding this comment.
See anomalyco/models.dev#646, this is now upstream including limit.input
brandon93s
changed the title
1 participant
What does this PR do?
inputlimit from models.devHow did you verify your code works?
bun devto confirm compaction triggers