Cache (5/5): Synchronize inference caches through Hugging Face - #127
Conversation
1cc01a3 to
b5ca140
Compare
37c82fe to
e5cd5a1
Compare
kargibora
left a comment
There was a problem hiding this comment.
Looks good to me, just a several comments
| @@ -0,0 +1,295 @@ | |||
| """Filtered Hugging Face synchronization for inference-cache folders.""" | |||
There was a problem hiding this comment.
I am not sure whether we shoudl expose this in judgearena/ perhaps it is better to put this to judgearena/utils or judgearena/cache/... (similarly other materials into this as well as we introduce more cache specific things, it makes more sense to create seperate package for it.
There was a problem hiding this comment.
(If we want to expose it as cache it makes sense to me that this stays under judgearena but still wanted to discuss)
There was a problem hiding this comment.
Agreed, moved the cache CLI to judgearena/cache/hf.py and updated the entry point to judgearena.cache.hf:cli. The other cache-specific modules now live under the same package.
…to cache-on-118/03-providers # Conflicts: # judgearena/inference.py # tests/test_inference_cache.py
…cache-on-118/04-runners
…che-on-118/05-hf-sync # Conflicts: # judgearena/inference.py # judgearena/models.py
…o cache-on-118/02-do-inference
…to cache-on-118/03-providers
…cache-on-118/04-runners
…che-on-118/05-hf-sync
Description
Adds an explicit CLI for sharing inference caches through a Hugging Face dataset repository.
pushed_at.metadata.jsonand the SQLite database in one commit.Synchronization remains separate from benchmark execution. The Parquet store, migration scripts, and cache browser from #71 are intentionally left out.
This is stacked on #124.