google-deepmind / google-deepmind/gemma
Dynamic cache resize
Open
enhancement
- Dominant language
- Python
- Stars
- 5.7k
- Forks
- 1k
- Avg merge
- 10h 33m
- Merged PRs (30d)
- 2
Description
In the ChatSampler, it would be good if users could resize the cache, like:
```python
sampler = gm.text.ChatSampler(cache_length=1024)
sampler.resize_cache(2048)
```
On Colab when reaching the full cache length, it would allow to still run more prompts without having to restart from scratch
Contributor guide
Assessment
This issue has not been assessed yet.