aws / aws/amazon-q-developer-cli

bug: Q is too eager to compact the context window

Aperta
#2,787 3 commenti 0 reazioni 0 assegnatari Vedi su GitHub
Lingua principale
Rust
Stelle
2k
Fork
439
Metriche di merge delle PR
Nessuna PR unita negli ultimi 30g

Descrizione

### Checks

- [x] I have searched [github.com/aws/amazon-q-developer-cli/issues](https://github.com/aws/amazon-q-developer-cli/issues?q=) and there are no duplicates of my issue
- [x] I have run `q doctor` in the affected terminal session
- [x] I have run `q restart` and replicated the issue again

### Operating system

AL2

### Expected behaviour

Q doesn't compact the context window until it is necessary

### Actual behaviour

q tries to compact the history, even though 30% of the context window is still available

in the following example, I used ctrl+c to stop the summarization, and then checked the usage. I still have 30% space, even though it tried to compact the history. I expect Q to use the entire context window before compacting, doing this early makes Q lose it train of thought and I usually just have to completely start over from scratch.

There is no way this prompt caused Q to use 30% more tokens. Its just a follow up from its previous prompt.

if this was 90-95% I wouldn't think this was a bug, but still having 30% and getting a compaction makes no sense.

```

> it didn't there are check styles errors

The context window has overflowed, summarizing the history...

> /usage

Current context window (142280 of 200k tokens used)
████████████████████████████████████████████████████████████████████████████████ 71.14%

█ Context files: ~0 tokens (0.00%)
█ Tools: ~58850 tokens (29.43%)
█ Q responses: ~19160 tokens (9.58%)
█ Your prompts: ~64270 tokens (32.14%)

💡 Pro Tips:
Run /compact to replace the conversation history with its summary
Run /clear to erase the entire chat history
Run /context show to see tokens per context file

```

### Steps to reproduce

_No response_

### Environment

```yaml

```

Guida per i contributori

Apri la guida per i contributori

Direzione di ricerca

Inizia riproducendo il report nella sessione terminale interessata con q doctor, q restart e il prompt successivo, quindi ispeziona /usage per confrontare la percentuale di contesto quando inizia la compattazione. L'issue è completata quando Q non compatta a circa il 71% di utilizzo e compatta solo quando è effettivamente necessario riepilogare la finestra di contesto.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Valutazione

Stack tecnologico
rust
Ambito
ai, cli
Tipo di issue
Bug
Difficoltà
4/5
Tempo stimato
3-5 giorni
Stato di attività
Ferma
Chiarezza
Da chiarire
Idoneità per principianti
25/100

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.