Upload 2 files
Browse files- model_metadata.parquet +3 -0
- postprocess_jsonl_latest.py +1626 -0
model_metadata.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:0853f8b1a3286dbc3c4ca3da2f5a66c7922003a9f5b1d72036857de3c47188c6
|
| 3 |
+
size 628246962
|
postprocess_jsonl_latest.py
ADDED
|
@@ -0,0 +1,1626 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import json
|
| 3 |
+
import logging
|
| 4 |
+
import time
|
| 5 |
+
import traceback
|
| 6 |
+
from pathlib import Path
|
| 7 |
+
import shutil
|
| 8 |
+
import psutil
|
| 9 |
+
import glob
|
| 10 |
+
import gc
|
| 11 |
+
from datetime import datetime
|
| 12 |
+
from tqdm.auto import tqdm
|
| 13 |
+
from typing import Optional, Union, Set, Dict, List, Tuple
|
| 14 |
+
|
| 15 |
+
# Hugging Face related
|
| 16 |
+
from huggingface_hub import list_models, hf_hub_download, HfApi
|
| 17 |
+
from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError, HFValidationError
|
| 18 |
+
|
| 19 |
+
# Data handling and Parquet
|
| 20 |
+
import pandas as pd
|
| 21 |
+
import pyarrow as pa
|
| 22 |
+
import pyarrow.parquet as pq
|
| 23 |
+
|
| 24 |
+
# Embeddings
|
| 25 |
+
from sentence_transformers import SentenceTransformer
|
| 26 |
+
import torch
|
| 27 |
+
|
| 28 |
+
# --- IPFS CID Generation Code (from provided ipfs_multiformats.py) ---
|
| 29 |
+
import hashlib
|
| 30 |
+
from multiformats import CID, multihash
|
| 31 |
+
import tempfile
|
| 32 |
+
import sys
|
| 33 |
+
|
| 34 |
+
class ipfs_multiformats_py:
|
| 35 |
+
def __init__(self, resources=None, metadata=None):
|
| 36 |
+
self.multihash = multihash
|
| 37 |
+
# Added error handling for multihash version/import
|
| 38 |
+
if not hasattr(self.multihash, 'wrap') or not hasattr(self.multihash, 'decode'):
|
| 39 |
+
logging.warning("Multihash library structure might have changed. CID generation may fail.")
|
| 40 |
+
return None
|
| 41 |
+
|
| 42 |
+
def get_file_sha256(self, file_path):
|
| 43 |
+
hasher = hashlib.sha256()
|
| 44 |
+
try:
|
| 45 |
+
with open(file_path, 'rb') as f:
|
| 46 |
+
while chunk := f.read(8192):
|
| 47 |
+
hasher.update(chunk)
|
| 48 |
+
return hasher.digest()
|
| 49 |
+
except Exception as e:
|
| 50 |
+
logging.error(f"Error hashing file {file_path}: {e}")
|
| 51 |
+
return None
|
| 52 |
+
|
| 53 |
+
# Takes bytes input directly
|
| 54 |
+
def get_bytes_sha256(self, data_bytes: bytes):
|
| 55 |
+
hasher = hashlib.sha256()
|
| 56 |
+
hasher.update(data_bytes)
|
| 57 |
+
return hasher.digest()
|
| 58 |
+
|
| 59 |
+
def get_multihash_sha256(self, content_hash):
|
| 60 |
+
if content_hash is None:
|
| 61 |
+
return None
|
| 62 |
+
try:
|
| 63 |
+
# Try using multihash.digest instead of wrap
|
| 64 |
+
mh = self.multihash.digest(content_hash, 'sha2-256')
|
| 65 |
+
return mh
|
| 66 |
+
except Exception as e:
|
| 67 |
+
logging.error(f"Error creating multihash: {e}")
|
| 68 |
+
return None
|
| 69 |
+
|
| 70 |
+
def get_multihash_sha256_old(self, content_hash):
|
| 71 |
+
if content_hash is None:
|
| 72 |
+
return None
|
| 73 |
+
try:
|
| 74 |
+
# Use 'sha2-256' which corresponds to code 0x12
|
| 75 |
+
#mh = self.multihash.wrap(code='sha2-256', digest=content_hash)
|
| 76 |
+
mh = self.multihash.wrap('sha2-256', content_hash)
|
| 77 |
+
return mh
|
| 78 |
+
except Exception as e:
|
| 79 |
+
logging.error(f"Error wrapping hash in multihash: {e}")
|
| 80 |
+
return None
|
| 81 |
+
|
| 82 |
+
def get_cid_old(self, data):
|
| 83 |
+
"""Generates CID v1 base32 for bytes data or file path."""
|
| 84 |
+
mh = None
|
| 85 |
+
try:
|
| 86 |
+
if isinstance(data, (str, Path)) and os.path.isfile(data):
|
| 87 |
+
# logging.debug(f"Calculating CID for file: {data}")
|
| 88 |
+
file_content_hash = self.get_file_sha256(data)
|
| 89 |
+
mh = self.get_multihash_sha256(file_content_hash)
|
| 90 |
+
elif isinstance(data, bytes):
|
| 91 |
+
# logging.debug(f"Calculating CID for bytes (length: {len(data)})")
|
| 92 |
+
bytes_hash = self.get_bytes_sha256(data)
|
| 93 |
+
mh = self.get_multihash_sha256(bytes_hash)
|
| 94 |
+
elif isinstance(data, str):
|
| 95 |
+
# logging.debug(f"Calculating CID for string (length: {len(data)})")
|
| 96 |
+
# Treat string as UTF-8 bytes
|
| 97 |
+
bytes_hash = self.get_bytes_sha256(data.encode('utf-8'))
|
| 98 |
+
mh = self.get_multihash_sha256(bytes_hash)
|
| 99 |
+
else:
|
| 100 |
+
logging.warning(f"Unsupported data type for CID generation: {type(data)}. Skipping CID.")
|
| 101 |
+
return None
|
| 102 |
+
|
| 103 |
+
if mh:
|
| 104 |
+
# CIDv1, base32, raw codec (0x55)
|
| 105 |
+
cid = CID(base='base32', version=1, codec='raw', multihash=mh)
|
| 106 |
+
return str(cid)
|
| 107 |
+
else:
|
| 108 |
+
return None
|
| 109 |
+
except Exception as e:
|
| 110 |
+
logging.error(f"Error generating CID: {e}", exc_info=False)
|
| 111 |
+
return None
|
| 112 |
+
|
| 113 |
+
def get_cid(self, data):
|
| 114 |
+
"""Generates CID v1 base32 for bytes data or file path."""
|
| 115 |
+
try:
|
| 116 |
+
# Get the hash first
|
| 117 |
+
content_hash = None
|
| 118 |
+
if isinstance(data, (str, Path)) and os.path.isfile(data):
|
| 119 |
+
content_hash = self.get_file_sha256(data)
|
| 120 |
+
elif isinstance(data, bytes):
|
| 121 |
+
content_hash = self.get_bytes_sha256(data)
|
| 122 |
+
elif isinstance(data, str):
|
| 123 |
+
content_hash = self.get_bytes_sha256(data.encode('utf-8'))
|
| 124 |
+
else:
|
| 125 |
+
logging.warning(f"Unsupported data type for CID generation: {type(data)}. Skipping CID.")
|
| 126 |
+
return None
|
| 127 |
+
|
| 128 |
+
if not content_hash:
|
| 129 |
+
return None
|
| 130 |
+
|
| 131 |
+
# Try the new CID API format
|
| 132 |
+
try:
|
| 133 |
+
# Version 1 of multiformats may use from_digest or other method instead of passing multihash directly
|
| 134 |
+
from multiformats import multihash
|
| 135 |
+
digest = multihash.digest(content_hash, 'sha2-256')
|
| 136 |
+
cid = CID.from_digest(digest, 'raw') # Try this format first
|
| 137 |
+
return str(cid)
|
| 138 |
+
except (AttributeError, TypeError):
|
| 139 |
+
try:
|
| 140 |
+
# Try alternate creation method
|
| 141 |
+
mh = self.get_multihash_sha256(content_hash)
|
| 142 |
+
cid = CID(version=1, codec='raw', hash=mh) # Try with hash parameter
|
| 143 |
+
return str(cid)
|
| 144 |
+
except:
|
| 145 |
+
# Fallback to simple base64 encoding if CID creation fails
|
| 146 |
+
import base64
|
| 147 |
+
b64_hash = base64.b64encode(content_hash).decode('ascii')
|
| 148 |
+
return f"sha256:{b64_hash}"
|
| 149 |
+
|
| 150 |
+
except Exception as e:
|
| 151 |
+
logging.error(f"Error generating CID: {e}", exc_info=False)
|
| 152 |
+
# Fallback to a simple hash representation
|
| 153 |
+
try:
|
| 154 |
+
if isinstance(data, (str, Path)) and os.path.isfile(data):
|
| 155 |
+
content_hash = self.get_file_sha256(data)
|
| 156 |
+
elif isinstance(data, bytes):
|
| 157 |
+
content_hash = self.get_bytes_sha256(data)
|
| 158 |
+
elif isinstance(data, str):
|
| 159 |
+
content_hash = self.get_bytes_sha256(data.encode('utf-8'))
|
| 160 |
+
else:
|
| 161 |
+
return None
|
| 162 |
+
|
| 163 |
+
import base64
|
| 164 |
+
return f"sha256:{base64.b64encode(content_hash).decode('ascii')}"
|
| 165 |
+
except:
|
| 166 |
+
return None
|
| 167 |
+
# --- End IPFS CID Code ---
|
| 168 |
+
|
| 169 |
+
|
| 170 |
+
# --- Configuration ---
|
| 171 |
+
# --- Paths ---
|
| 172 |
+
GDRIVE_MOUNT_POINT = "/content/drive/MyDrive"
|
| 173 |
+
GDRIVE_FOLDER_NAME = "hf_metadata_dataset_collection"
|
| 174 |
+
LOCAL_FOLDER_NAME = "./hf_metadata_dataset_local_fallback"
|
| 175 |
+
LOCAL_WORK_DIR = Path(os.path.abspath("./hf_embedding_work"))
|
| 176 |
+
|
| 177 |
+
# Input JSONL File
|
| 178 |
+
INPUT_JSONL_FILENAME = "all_models_metadata.jsonl" # Assumed in final dir
|
| 179 |
+
|
| 180 |
+
# --- Output File Names ---
|
| 181 |
+
# Final Destination (Drive/Local Fallback)
|
| 182 |
+
FINAL_METADATA_PARQUET_FILENAME = "model_metadata.parquet" # Metadata + CIDs
|
| 183 |
+
FINAL_EMBEDDINGS_PARQUET_FILENAME = "model_embeddings.parquet" # CIDs + Embeddings
|
| 184 |
+
FINAL_LOG_FILENAME = "embedding_generator.log"
|
| 185 |
+
|
| 186 |
+
# Local Temporary Files (in LOCAL_WORK_DIR)
|
| 187 |
+
LOCAL_TEMP_METADATA_JSONL = "temp_model_metadata.jsonl"
|
| 188 |
+
LOCAL_TEMP_EMBEDDINGS_JSONL = "temp_model_embeddings.jsonl"
|
| 189 |
+
LOCAL_TEMP_LOG_FILENAME = "temp_embedding_generator.log"
|
| 190 |
+
|
| 191 |
+
# --- Batch Configuration ---
|
| 192 |
+
BATCH_SAVE_THRESHOLD = 1000 # Save after processing this many records
|
| 193 |
+
BATCH_SAVE_DIR_NAME = "batch_files" # Subdirectory for batch files
|
| 194 |
+
PERIODIC_MERGE_FREQUENCY = 5 # Merge to Google Drive every X batches (0 to disable)
|
| 195 |
+
CLEAN_AFTER_PERIODIC_MERGE = True # Whether to clean up batch files after periodic merge
|
| 196 |
+
|
| 197 |
+
# --- Memory Management Configuration ---
|
| 198 |
+
MEMORY_CLEANUP_THRESHOLD_MB = 1000 # Force extra cleanup if memory growth exceeds this
|
| 199 |
+
|
| 200 |
+
# --- Processing Config ---
|
| 201 |
+
MAX_RECORDS_TO_PROCESS = None # Limit records from JSONL (for testing), None for all
|
| 202 |
+
BATCH_SIZE = 1024
|
| 203 |
+
EMBEDDING_MODEL_NAME = 'all-MiniLM-L6-v2'
|
| 204 |
+
# Control what gets embedded and CID generated
|
| 205 |
+
PROCESS_CONFIG_JSON = True
|
| 206 |
+
PROCESS_README_CONTENT = True
|
| 207 |
+
|
| 208 |
+
# --- Hub Upload Config ---
|
| 209 |
+
UPLOAD_TO_HUB = True
|
| 210 |
+
TARGET_REPO_ID = "YourUsername/your-dataset-repo-name" # CHANGE THIS
|
| 211 |
+
TARGET_REPO_TYPE = "dataset"
|
| 212 |
+
METADATA_FILENAME_IN_REPO = "model_metadata.parquet"
|
| 213 |
+
EMBEDDINGS_FILENAME_IN_REPO = "model_embeddings.parquet"
|
| 214 |
+
|
| 215 |
+
# --- Setup Logging ---
|
| 216 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
| 217 |
+
|
| 218 |
+
|
| 219 |
+
# --- Helper Functions ---
|
| 220 |
+
def make_serializable(obj):
|
| 221 |
+
"""Converts common non-serializable types found in ModelInfo."""
|
| 222 |
+
if hasattr(obj, 'isoformat'): return obj.isoformat()
|
| 223 |
+
if hasattr(obj, 'rfilename'): return obj.rfilename
|
| 224 |
+
try: return str(obj)
|
| 225 |
+
except Exception: return None
|
| 226 |
+
|
| 227 |
+
def safe_serialize_dict(data_dict):
|
| 228 |
+
"""Attempts to serialize a dictionary, handling non-serializable items."""
|
| 229 |
+
# This function might not be needed if we read directly from JSONL,
|
| 230 |
+
# but keep it for potential future use or if handling raw ModelInfo objects.
|
| 231 |
+
serializable_dict = {}
|
| 232 |
+
if not isinstance(data_dict, dict): logging.warning(f"safe_serialize_dict non-dict input: {type(data_dict)}"); return {}
|
| 233 |
+
for key, value in data_dict.items():
|
| 234 |
+
if isinstance(value, (list, tuple)): serializable_dict[key] = [make_serializable(item) for item in value]
|
| 235 |
+
elif isinstance(value, dict): serializable_dict[key] = safe_serialize_dict(value)
|
| 236 |
+
elif isinstance(value, (str, int, float, bool, type(None))): serializable_dict[key] = value
|
| 237 |
+
else: serializable_dict[key] = make_serializable(value)
|
| 238 |
+
return {k: v for k, v in serializable_dict.items() if v is not None or (k in data_dict and data_dict[k] is None)}
|
| 239 |
+
|
| 240 |
+
# --- NEW: Generate Record CID Function ---
|
| 241 |
+
def generate_record_cid(cid_generator, model_id: str, config_cid: Optional[str] = None, readme_cid: Optional[str] = None) -> str:
|
| 242 |
+
"""
|
| 243 |
+
Generate a primary record CID from model_id and available content CIDs.
|
| 244 |
+
This will be used as the primary key for both Parquet files.
|
| 245 |
+
"""
|
| 246 |
+
# Create a base string that combines all available IDs
|
| 247 |
+
cid_parts = [f"model:{model_id}"]
|
| 248 |
+
if config_cid:
|
| 249 |
+
cid_parts.append(f"config:{config_cid}")
|
| 250 |
+
if readme_cid:
|
| 251 |
+
cid_parts.append(f"readme:{readme_cid}")
|
| 252 |
+
|
| 253 |
+
# Join all parts and generate a CID from the combined string
|
| 254 |
+
combined_string = "|".join(cid_parts)
|
| 255 |
+
return cid_generator.get_cid(combined_string)
|
| 256 |
+
|
| 257 |
+
# --- Safe Parquet Saving Function ---
|
| 258 |
+
def save_dataframe_to_parquet_safely(df, filepath):
|
| 259 |
+
"""Saves DataFrame to Parquet with explicit schema handling for mixed types."""
|
| 260 |
+
try:
|
| 261 |
+
# First attempt: Convert known problematic columns to string
|
| 262 |
+
df_safe = df.copy()
|
| 263 |
+
|
| 264 |
+
# Handle the 'gated' column specifically which caused the original error
|
| 265 |
+
if 'gated' in df_safe.columns:
|
| 266 |
+
df_safe['gated'] = df_safe['gated'].astype(str)
|
| 267 |
+
|
| 268 |
+
# Convert all object columns except model_id and record_cid to string to be safe
|
| 269 |
+
for col in df_safe.select_dtypes(include=['object']).columns:
|
| 270 |
+
if col not in ['model_id', 'record_cid', 'config_cid', 'readme_cid']: # Keep IDs as is
|
| 271 |
+
df_safe[col] = df_safe[col].astype(str)
|
| 272 |
+
|
| 273 |
+
# Try saving with pandas
|
| 274 |
+
df_safe.to_parquet(filepath, index=False, compression='gzip')
|
| 275 |
+
return True
|
| 276 |
+
|
| 277 |
+
except Exception as e:
|
| 278 |
+
logging.warning(f"First attempt to save Parquet failed: {e}")
|
| 279 |
+
|
| 280 |
+
try:
|
| 281 |
+
# Second attempt: Use PyArrow with explicit schema
|
| 282 |
+
schema = pa.Schema.from_pandas(df)
|
| 283 |
+
fields = list(schema)
|
| 284 |
+
|
| 285 |
+
# Convert all string/binary fields to string type except IDs
|
| 286 |
+
for i, field in enumerate(fields):
|
| 287 |
+
if (pa.types.is_string(field.type) or pa.types.is_binary(field.type)) and \
|
| 288 |
+
field.name not in ['model_id', 'record_cid', 'config_cid', 'readme_cid']:
|
| 289 |
+
fields[i] = pa.field(field.name, pa.string())
|
| 290 |
+
|
| 291 |
+
new_schema = pa.schema(fields)
|
| 292 |
+
|
| 293 |
+
# Force conversion of problematic columns
|
| 294 |
+
df_safe = df.copy()
|
| 295 |
+
for col in df_safe.select_dtypes(include=['object']).columns:
|
| 296 |
+
if col not in ['model_id', 'record_cid', 'config_cid', 'readme_cid']:
|
| 297 |
+
df_safe[col] = df_safe[col].astype(str)
|
| 298 |
+
|
| 299 |
+
# Convert to table with schema and write
|
| 300 |
+
table = pa.Table.from_pandas(df_safe, schema=new_schema)
|
| 301 |
+
pq.write_table(table, filepath)
|
| 302 |
+
logging.info(f"Successfully saved to {filepath} using PyArrow with schema conversion")
|
| 303 |
+
return True
|
| 304 |
+
|
| 305 |
+
except Exception as e2:
|
| 306 |
+
logging.error(f"Both Parquet saving attempts failed for {filepath}: {e2}")
|
| 307 |
+
|
| 308 |
+
# Last resort - save to CSV instead
|
| 309 |
+
try:
|
| 310 |
+
csv_filepath = filepath.with_suffix('.csv')
|
| 311 |
+
logging.warning(f"Falling back to CSV format: {csv_filepath}")
|
| 312 |
+
df.to_csv(csv_filepath, index=False)
|
| 313 |
+
logging.info(f"Saved as CSV instead: {csv_filepath}")
|
| 314 |
+
return False
|
| 315 |
+
except Exception as e3:
|
| 316 |
+
logging.error(f"Even CSV fallback failed: {e3}")
|
| 317 |
+
return False
|
| 318 |
+
|
| 319 |
+
# --- UPDATED: Load Processed CIDs from EMBEDDINGS Parquet and Batch Files ---
|
| 320 |
+
def load_processed_cids_from_parquet(filepath: Path, batch_dir: Optional[Path] = None) -> set:
|
| 321 |
+
"""
|
| 322 |
+
Reads the record_cid column from:
|
| 323 |
+
1. The final EMBEDDINGS Parquet file
|
| 324 |
+
2. Any batch files in the batch_dir, if provided
|
| 325 |
+
3. Also checks for CSV fallback files
|
| 326 |
+
|
| 327 |
+
Returns a set of processed record_cids.
|
| 328 |
+
"""
|
| 329 |
+
processed_cids = set()
|
| 330 |
+
|
| 331 |
+
# 1. Load from final Parquet if it exists
|
| 332 |
+
if filepath.is_file():
|
| 333 |
+
logging.info(f"Found existing EMBEDDINGS Parquet: {filepath}. Loading processed CIDs...")
|
| 334 |
+
try:
|
| 335 |
+
# Only load the record_cid column for efficiency
|
| 336 |
+
df_existing = pd.read_parquet(filepath, columns=['record_cid'])
|
| 337 |
+
file_cids = set(df_existing['record_cid'].tolist())
|
| 338 |
+
processed_cids.update(file_cids)
|
| 339 |
+
logging.info(f"Loaded {len(file_cids)} CIDs from existing Embeddings Parquet.")
|
| 340 |
+
except Exception as e:
|
| 341 |
+
logging.warning(f"Could not load 'record_cid' from '{filepath}': {e}. Will check for CSV fallback.")
|
| 342 |
+
# Check for CSV fallback
|
| 343 |
+
csv_filepath = filepath.with_suffix('.csv')
|
| 344 |
+
if csv_filepath.is_file():
|
| 345 |
+
try:
|
| 346 |
+
df_csv = pd.read_csv(csv_filepath, usecols=['record_cid'])
|
| 347 |
+
csv_cids = set(df_csv['record_cid'].tolist())
|
| 348 |
+
processed_cids.update(csv_cids)
|
| 349 |
+
logging.info(f"Loaded {len(csv_cids)} CIDs from CSV fallback: {csv_filepath}")
|
| 350 |
+
except Exception as csv_e:
|
| 351 |
+
logging.warning(f"Could not load CIDs from CSV fallback: {csv_e}")
|
| 352 |
+
|
| 353 |
+
# 2. Load from batch files if provided
|
| 354 |
+
if batch_dir and batch_dir.is_dir():
|
| 355 |
+
# Check both Parquet and CSV batch files
|
| 356 |
+
batch_files_parquet = list(batch_dir.glob("embeddings_batch_*.parquet"))
|
| 357 |
+
batch_files_csv = list(batch_dir.glob("embeddings_batch_*.csv"))
|
| 358 |
+
|
| 359 |
+
if batch_files_parquet:
|
| 360 |
+
logging.info(f"Found {len(batch_files_parquet)} embedding batch Parquet files.")
|
| 361 |
+
batch_cids_count = 0
|
| 362 |
+
|
| 363 |
+
for batch_file in batch_files_parquet:
|
| 364 |
+
try:
|
| 365 |
+
df_batch = pd.read_parquet(batch_file, columns=['record_cid'])
|
| 366 |
+
batch_cids = set(df_batch['record_cid'].tolist())
|
| 367 |
+
batch_cids_count += len(batch_cids)
|
| 368 |
+
processed_cids.update(batch_cids)
|
| 369 |
+
except Exception as e:
|
| 370 |
+
logging.warning(f"Error loading CIDs from batch file {batch_file}: {e}")
|
| 371 |
+
|
| 372 |
+
logging.info(f"Loaded {batch_cids_count} additional CIDs from Parquet batch files.")
|
| 373 |
+
|
| 374 |
+
if batch_files_csv:
|
| 375 |
+
logging.info(f"Found {len(batch_files_csv)} embedding batch CSV files.")
|
| 376 |
+
csv_batch_cids_count = 0
|
| 377 |
+
|
| 378 |
+
for batch_file in batch_files_csv:
|
| 379 |
+
try:
|
| 380 |
+
df_batch = pd.read_csv(batch_file, usecols=['record_cid'])
|
| 381 |
+
batch_cids = set(df_batch['record_cid'].tolist())
|
| 382 |
+
csv_batch_cids_count += len(batch_cids)
|
| 383 |
+
processed_cids.update(batch_cids)
|
| 384 |
+
except Exception as e:
|
| 385 |
+
logging.warning(f"Error loading CIDs from CSV batch file {batch_file}: {e}")
|
| 386 |
+
|
| 387 |
+
logging.info(f"Loaded {csv_batch_cids_count} additional CIDs from CSV batch files.")
|
| 388 |
+
|
| 389 |
+
total_cids = len(processed_cids)
|
| 390 |
+
if total_cids > 0:
|
| 391 |
+
logging.info(f"Total of {total_cids} unique record CIDs loaded for resume.")
|
| 392 |
+
else:
|
| 393 |
+
logging.info(f"No existing processed CIDs found. Will process all records.")
|
| 394 |
+
|
| 395 |
+
return processed_cids
|
| 396 |
+
|
| 397 |
+
|
| 398 |
+
# final conversion from JSONL to Parquet would happen only once at the end of all processing.
|
| 399 |
+
def convert_jsonl_to_parquet(
|
| 400 |
+
meta_jsonl_path: Path,
|
| 401 |
+
embed_jsonl_path: Path,
|
| 402 |
+
local_temp_metadata_path: Path,
|
| 403 |
+
local_temp_embeddings_path: Path,
|
| 404 |
+
chunk_size: int = 50000,
|
| 405 |
+
max_memory_mb: int = 2000 # Memory threshold for adaptive processing
|
| 406 |
+
):
|
| 407 |
+
"""
|
| 408 |
+
Convert very large JSONL files to Parquet using a streaming approach with minimal memory usage.
|
| 409 |
+
|
| 410 |
+
Args:
|
| 411 |
+
meta_jsonl_path: Path to metadata JSONL file
|
| 412 |
+
embed_jsonl_path: Path to embeddings JSONL file
|
| 413 |
+
local_temp_metadata_path: Output path for metadata Parquet file
|
| 414 |
+
local_temp_embeddings_path: Output path for embeddings Parquet file
|
| 415 |
+
chunk_size: Initial number of records to process at once (will adapt based on memory usage)
|
| 416 |
+
max_memory_mb: Maximum memory usage threshold in MB
|
| 417 |
+
"""
|
| 418 |
+
import json
|
| 419 |
+
import os
|
| 420 |
+
import gc
|
| 421 |
+
import time
|
| 422 |
+
import pyarrow as pa
|
| 423 |
+
import pyarrow.parquet as pq
|
| 424 |
+
from tqdm import tqdm
|
| 425 |
+
import psutil
|
| 426 |
+
|
| 427 |
+
logging.info("Starting optimized streaming conversion from JSONL to Parquet")
|
| 428 |
+
|
| 429 |
+
def get_memory_usage_mb():
|
| 430 |
+
"""Get current memory usage in MB"""
|
| 431 |
+
process = psutil.Process(os.getpid())
|
| 432 |
+
return process.memory_info().rss / (1024 * 1024)
|
| 433 |
+
|
| 434 |
+
def estimate_total_lines(file_path, sample_size=1000000):
|
| 435 |
+
"""Estimate total lines in file without reading entire file"""
|
| 436 |
+
try:
|
| 437 |
+
# Get file size
|
| 438 |
+
file_size = os.path.getsize(file_path)
|
| 439 |
+
|
| 440 |
+
# If file is small enough, just count lines directly
|
| 441 |
+
if file_size < 100 * 1024 * 1024: # 100 MB
|
| 442 |
+
with open(file_path, 'r') as f:
|
| 443 |
+
return sum(1 for _ in f)
|
| 444 |
+
|
| 445 |
+
# Sample beginning of file to estimate line size
|
| 446 |
+
line_count = 0
|
| 447 |
+
bytes_read = 0
|
| 448 |
+
with open(file_path, 'r') as f:
|
| 449 |
+
for _ in range(sample_size):
|
| 450 |
+
line = f.readline()
|
| 451 |
+
if not line:
|
| 452 |
+
break
|
| 453 |
+
bytes_read += len(line.encode('utf-8'))
|
| 454 |
+
line_count += 1
|
| 455 |
+
|
| 456 |
+
if line_count == 0:
|
| 457 |
+
return 0
|
| 458 |
+
|
| 459 |
+
# Calculate average line size and estimate total
|
| 460 |
+
avg_line_size = bytes_read / line_count
|
| 461 |
+
estimated_lines = int(file_size / avg_line_size)
|
| 462 |
+
|
| 463 |
+
logging.info(f"Estimated lines in {file_path.name}: {estimated_lines:,} (based on avg line size: {avg_line_size:.1f} bytes)")
|
| 464 |
+
return estimated_lines
|
| 465 |
+
|
| 466 |
+
except Exception as e:
|
| 467 |
+
logging.error(f"Error estimating lines in file: {e}")
|
| 468 |
+
return 0
|
| 469 |
+
|
| 470 |
+
def infer_schema_from_samples(file_path, num_samples=1000):
|
| 471 |
+
"""Infer schema by sampling from beginning, middle, and end of file"""
|
| 472 |
+
try:
|
| 473 |
+
file_size = os.path.getsize(file_path)
|
| 474 |
+
if file_size == 0:
|
| 475 |
+
return None
|
| 476 |
+
|
| 477 |
+
samples = []
|
| 478 |
+
with open(file_path, 'r') as f:
|
| 479 |
+
# Read samples from beginning
|
| 480 |
+
for _ in range(num_samples // 3):
|
| 481 |
+
line = f.readline()
|
| 482 |
+
if not line:
|
| 483 |
+
break
|
| 484 |
+
try:
|
| 485 |
+
samples.append(json.loads(line))
|
| 486 |
+
except json.JSONDecodeError:
|
| 487 |
+
continue
|
| 488 |
+
|
| 489 |
+
# Read samples from middle
|
| 490 |
+
middle_pos = file_size // 2
|
| 491 |
+
f.seek(middle_pos)
|
| 492 |
+
f.readline() # Skip partial line
|
| 493 |
+
for _ in range(num_samples // 3):
|
| 494 |
+
line = f.readline()
|
| 495 |
+
if not line:
|
| 496 |
+
break
|
| 497 |
+
try:
|
| 498 |
+
samples.append(json.loads(line))
|
| 499 |
+
except json.JSONDecodeError:
|
| 500 |
+
continue
|
| 501 |
+
|
| 502 |
+
# Read samples from end
|
| 503 |
+
end_pos = max(0, file_size - 100000) # 100 KB from end
|
| 504 |
+
f.seek(end_pos)
|
| 505 |
+
f.readline() # Skip partial line
|
| 506 |
+
for _ in range(num_samples // 3):
|
| 507 |
+
line = f.readline()
|
| 508 |
+
if not line:
|
| 509 |
+
break
|
| 510 |
+
try:
|
| 511 |
+
samples.append(json.loads(line))
|
| 512 |
+
except json.JSONDecodeError:
|
| 513 |
+
continue
|
| 514 |
+
|
| 515 |
+
if not samples:
|
| 516 |
+
logging.error(f"No valid JSON samples found in {file_path}")
|
| 517 |
+
return None
|
| 518 |
+
|
| 519 |
+
# Convert samples to pyarrow schema
|
| 520 |
+
import pandas as pd
|
| 521 |
+
sample_df = pd.DataFrame(samples)
|
| 522 |
+
|
| 523 |
+
# Convert all columns to string type to avoid type mismatches
|
| 524 |
+
for col in sample_df.columns:
|
| 525 |
+
if col != 'embedding': # Keep embedding as is since it's numeric
|
| 526 |
+
sample_df[col] = sample_df[col].astype(str)
|
| 527 |
+
|
| 528 |
+
# Handle embedding field specially if it exists
|
| 529 |
+
if 'embedding' in sample_df.columns:
|
| 530 |
+
# Ensure embedding is a list of float
|
| 531 |
+
if sample_df['embedding'].dtype != 'object':
|
| 532 |
+
# If not already a list, convert to string
|
| 533 |
+
sample_df['embedding'] = sample_df['embedding'].astype(str)
|
| 534 |
+
|
| 535 |
+
# Convert to PyArrow Table and extract schema
|
| 536 |
+
table = pa.Table.from_pandas(sample_df)
|
| 537 |
+
logging.info(f"Inferred schema with {len(table.schema.names)} fields")
|
| 538 |
+
return table.schema
|
| 539 |
+
|
| 540 |
+
except Exception as e:
|
| 541 |
+
logging.error(f"Error inferring schema: {e}", exc_info=True)
|
| 542 |
+
return None
|
| 543 |
+
|
| 544 |
+
def stream_jsonl_to_parquet(jsonl_path, parquet_path, file_type, initial_chunk_size):
|
| 545 |
+
"""Process a JSONL file in a streaming fashion with adaptive chunk sizing"""
|
| 546 |
+
if not jsonl_path.exists():
|
| 547 |
+
logging.warning(f"{file_type} JSONL file not found: {jsonl_path}")
|
| 548 |
+
return False
|
| 549 |
+
|
| 550 |
+
logging.info(f"Starting streaming conversion of {file_type} JSONL: {jsonl_path} -> {parquet_path}")
|
| 551 |
+
start_time = time.time()
|
| 552 |
+
|
| 553 |
+
# Get schema by sampling
|
| 554 |
+
schema = infer_schema_from_samples(jsonl_path)
|
| 555 |
+
if schema is None:
|
| 556 |
+
logging.error(f"Failed to infer schema for {file_type}")
|
| 557 |
+
return False
|
| 558 |
+
|
| 559 |
+
# Estimate total for progress reporting
|
| 560 |
+
estimated_total = estimate_total_lines(jsonl_path)
|
| 561 |
+
|
| 562 |
+
# Track current chunk size - will adapt based on memory usage
|
| 563 |
+
current_chunk_size = initial_chunk_size
|
| 564 |
+
records_processed = 0
|
| 565 |
+
chunk_count = 0
|
| 566 |
+
|
| 567 |
+
try:
|
| 568 |
+
# Create parquet writer with inferred schema
|
| 569 |
+
with pq.ParquetWriter(parquet_path, schema) as writer:
|
| 570 |
+
# Process in chunks to limit memory usage
|
| 571 |
+
buffer = []
|
| 572 |
+
|
| 573 |
+
with tqdm(total=estimated_total, desc=f"Converting {file_type}") as pbar:
|
| 574 |
+
with open(jsonl_path, 'r') as f:
|
| 575 |
+
for line_num, line in enumerate(f, 1):
|
| 576 |
+
try:
|
| 577 |
+
record = json.loads(line)
|
| 578 |
+
|
| 579 |
+
# Convert all string fields to ensure type consistency
|
| 580 |
+
for key, value in record.items():
|
| 581 |
+
if key != 'embedding' and value is not None and not isinstance(value, (list, dict)):
|
| 582 |
+
record[key] = str(value)
|
| 583 |
+
|
| 584 |
+
buffer.append(record)
|
| 585 |
+
|
| 586 |
+
# When buffer reaches chunk size, write to parquet
|
| 587 |
+
if len(buffer) >= current_chunk_size:
|
| 588 |
+
# Convert buffer to PyArrow table
|
| 589 |
+
import pandas as pd
|
| 590 |
+
chunk_df = pd.DataFrame(buffer)
|
| 591 |
+
|
| 592 |
+
# Handle embedding field specially if it exists
|
| 593 |
+
if 'embedding' in chunk_df.columns:
|
| 594 |
+
# Ensure embedding is a list of float
|
| 595 |
+
if chunk_df['embedding'].dtype != 'object':
|
| 596 |
+
# If not already a list, convert to string
|
| 597 |
+
chunk_df['embedding'] = chunk_df['embedding'].astype(str)
|
| 598 |
+
|
| 599 |
+
# Convert non-embedding fields to string
|
| 600 |
+
for col in chunk_df.columns:
|
| 601 |
+
if col != 'embedding':
|
| 602 |
+
chunk_df[col] = chunk_df[col].astype(str)
|
| 603 |
+
|
| 604 |
+
# Write chunk
|
| 605 |
+
table = pa.Table.from_pandas(chunk_df, schema=schema)
|
| 606 |
+
writer.write_table(table)
|
| 607 |
+
|
| 608 |
+
# Update progress
|
| 609 |
+
records_processed += len(buffer)
|
| 610 |
+
pbar.update(len(buffer))
|
| 611 |
+
chunk_count += 1
|
| 612 |
+
|
| 613 |
+
# Clear buffer and force garbage collection
|
| 614 |
+
buffer = []
|
| 615 |
+
del chunk_df, table
|
| 616 |
+
gc.collect()
|
| 617 |
+
|
| 618 |
+
# Adaptive chunk sizing based on memory usage
|
| 619 |
+
current_memory = get_memory_usage_mb()
|
| 620 |
+
if current_memory > max_memory_mb:
|
| 621 |
+
# Reduce chunk size if memory usage is too high
|
| 622 |
+
new_chunk_size = max(1000, int(current_chunk_size * 0.8))
|
| 623 |
+
logging.info(f"Memory usage high ({current_memory:.1f} MB). Reducing chunk size from {current_chunk_size} to {new_chunk_size}")
|
| 624 |
+
current_chunk_size = new_chunk_size
|
| 625 |
+
elif current_memory < max_memory_mb * 0.5 and current_chunk_size < initial_chunk_size:
|
| 626 |
+
# Increase chunk size if memory usage is low
|
| 627 |
+
new_chunk_size = min(initial_chunk_size, int(current_chunk_size * 1.2))
|
| 628 |
+
logging.info(f"Memory usage low ({current_memory:.1f} MB). Increasing chunk size from {current_chunk_size} to {new_chunk_size}")
|
| 629 |
+
current_chunk_size = new_chunk_size
|
| 630 |
+
|
| 631 |
+
# Log progress periodically
|
| 632 |
+
if chunk_count % 10 == 0:
|
| 633 |
+
elapsed = time.time() - start_time
|
| 634 |
+
rate = records_processed / elapsed if elapsed > 0 else 0
|
| 635 |
+
logging.info(f"Processed {records_processed:,} records ({rate:.1f} records/sec), memory: {current_memory:.1f} MB")
|
| 636 |
+
|
| 637 |
+
except json.JSONDecodeError:
|
| 638 |
+
logging.warning(f"Invalid JSON at line {line_num}")
|
| 639 |
+
continue
|
| 640 |
+
except Exception as e:
|
| 641 |
+
logging.warning(f"Error processing line {line_num}: {e}")
|
| 642 |
+
continue
|
| 643 |
+
|
| 644 |
+
# Write any remaining records
|
| 645 |
+
if buffer:
|
| 646 |
+
try:
|
| 647 |
+
import pandas as pd
|
| 648 |
+
chunk_df = pd.DataFrame(buffer)
|
| 649 |
+
|
| 650 |
+
# Handle embedding field specially if it exists
|
| 651 |
+
if 'embedding' in chunk_df.columns:
|
| 652 |
+
# Ensure embedding is a list of float
|
| 653 |
+
if chunk_df['embedding'].dtype != 'object':
|
| 654 |
+
# If not already a list, convert to string
|
| 655 |
+
chunk_df['embedding'] = chunk_df['embedding'].astype(str)
|
| 656 |
+
|
| 657 |
+
# Convert non-embedding fields to string
|
| 658 |
+
for col in chunk_df.columns:
|
| 659 |
+
if col != 'embedding':
|
| 660 |
+
chunk_df[col] = chunk_df[col].astype(str)
|
| 661 |
+
|
| 662 |
+
# Write final chunk
|
| 663 |
+
table = pa.Table.from_pandas(chunk_df, schema=schema)
|
| 664 |
+
writer.write_table(table)
|
| 665 |
+
|
| 666 |
+
# Update progress
|
| 667 |
+
records_processed += len(buffer)
|
| 668 |
+
pbar.update(len(buffer))
|
| 669 |
+
|
| 670 |
+
except Exception as e:
|
| 671 |
+
logging.error(f"Error writing final chunk: {e}")
|
| 672 |
+
|
| 673 |
+
# Report final stats
|
| 674 |
+
elapsed = time.time() - start_time
|
| 675 |
+
rate = records_processed / elapsed if elapsed > 0 else 0
|
| 676 |
+
logging.info(f"Successfully converted {records_processed:,} {file_type} records in {elapsed:.1f} seconds ({rate:.1f} records/sec)")
|
| 677 |
+
logging.info(f"Created {file_type} Parquet file: {parquet_path} ({os.path.getsize(parquet_path) / (1024*1024):.1f} MB)")
|
| 678 |
+
return True
|
| 679 |
+
|
| 680 |
+
except Exception as e:
|
| 681 |
+
logging.error(f"Error during {file_type} conversion: {e}", exc_info=True)
|
| 682 |
+
return False
|
| 683 |
+
|
| 684 |
+
# Convert metadata file
|
| 685 |
+
meta_success = stream_jsonl_to_parquet(meta_jsonl_path, local_temp_metadata_path, "metadata", chunk_size)
|
| 686 |
+
|
| 687 |
+
# Force garbage collection before processing embeddings
|
| 688 |
+
gc.collect()
|
| 689 |
+
|
| 690 |
+
# Convert embeddings file
|
| 691 |
+
embed_success = stream_jsonl_to_parquet(embed_jsonl_path, local_temp_embeddings_path, "embeddings", chunk_size)
|
| 692 |
+
|
| 693 |
+
if meta_success and embed_success:
|
| 694 |
+
logging.info("JSONL to Parquet conversion completed successfully")
|
| 695 |
+
return True
|
| 696 |
+
else:
|
| 697 |
+
logging.error("JSONL to Parquet conversion encountered errors")
|
| 698 |
+
return False
|
| 699 |
+
|
| 700 |
+
|
| 701 |
+
|
| 702 |
+
# --- Sync Local Files to Final Destination ---
|
| 703 |
+
def sync_local_files_to_final(
|
| 704 |
+
local_metadata_path: Path,
|
| 705 |
+
local_embeddings_path: Path,
|
| 706 |
+
local_log_path: Path,
|
| 707 |
+
final_metadata_path: Path,
|
| 708 |
+
final_embeddings_path: Path,
|
| 709 |
+
final_log_path: Path
|
| 710 |
+
):
|
| 711 |
+
"""
|
| 712 |
+
Copies local Parquet/log files to overwrite final destination files.
|
| 713 |
+
Returns True if all necessary copies succeeded.
|
| 714 |
+
"""
|
| 715 |
+
success = True # Assume success initially
|
| 716 |
+
|
| 717 |
+
# Copy Metadata Parquet or CSV
|
| 718 |
+
if local_metadata_path.is_file():
|
| 719 |
+
try:
|
| 720 |
+
logging.info(f"Copying local Metadata '{local_metadata_path}' to '{final_metadata_path}'...")
|
| 721 |
+
final_metadata_path.parent.mkdir(parents=True, exist_ok=True)
|
| 722 |
+
shutil.copyfile(local_metadata_path, final_metadata_path)
|
| 723 |
+
logging.info("Metadata file copy successful.")
|
| 724 |
+
except Exception as e:
|
| 725 |
+
logging.error(f"Failed to copy Metadata file: {e}", exc_info=True)
|
| 726 |
+
success = False
|
| 727 |
+
|
| 728 |
+
# Also check for CSV fallback
|
| 729 |
+
csv_path = local_metadata_path.with_suffix('.csv')
|
| 730 |
+
if csv_path.is_file():
|
| 731 |
+
try:
|
| 732 |
+
csv_dest = final_metadata_path.with_suffix('.csv')
|
| 733 |
+
logging.info(f"Copying CSV fallback: {csv_path} to {csv_dest}")
|
| 734 |
+
shutil.copyfile(csv_path, csv_dest)
|
| 735 |
+
except Exception as e:
|
| 736 |
+
logging.error(f"Failed to copy CSV fallback: {e}")
|
| 737 |
+
# Don't affect overall success status for CSV fallback
|
| 738 |
+
else:
|
| 739 |
+
logging.debug("Local Metadata file non-existent. Skipping copy.")
|
| 740 |
+
|
| 741 |
+
# Copy Embeddings Parquet or CSV
|
| 742 |
+
if local_embeddings_path.is_file():
|
| 743 |
+
try:
|
| 744 |
+
logging.info(f"Copying local Embeddings '{local_embeddings_path}' to '{final_embeddings_path}'...")
|
| 745 |
+
final_embeddings_path.parent.mkdir(parents=True, exist_ok=True)
|
| 746 |
+
shutil.copyfile(local_embeddings_path, final_embeddings_path)
|
| 747 |
+
logging.info("Embeddings file copy successful.")
|
| 748 |
+
except Exception as e:
|
| 749 |
+
logging.error(f"Failed to copy Embeddings file: {e}", exc_info=True)
|
| 750 |
+
success = False
|
| 751 |
+
|
| 752 |
+
# Also check for CSV fallback
|
| 753 |
+
csv_path = local_embeddings_path.with_suffix('.csv')
|
| 754 |
+
if csv_path.is_file():
|
| 755 |
+
try:
|
| 756 |
+
csv_dest = final_embeddings_path.with_suffix('.csv')
|
| 757 |
+
logging.info(f"Copying CSV fallback: {csv_path} to {csv_dest}")
|
| 758 |
+
shutil.copyfile(csv_path, csv_dest)
|
| 759 |
+
except Exception as e:
|
| 760 |
+
logging.error(f"Failed to copy CSV fallback: {e}")
|
| 761 |
+
# Don't affect overall success status for CSV fallback
|
| 762 |
+
else:
|
| 763 |
+
logging.debug("Local Embeddings file non-existent. Skipping copy.")
|
| 764 |
+
|
| 765 |
+
# Copy Log File
|
| 766 |
+
if local_log_path.is_file() and local_log_path.stat().st_size > 0:
|
| 767 |
+
try:
|
| 768 |
+
logging.info(f"Copying local log '{local_log_path}' to overwrite '{final_log_path}'...")
|
| 769 |
+
final_log_path.parent.mkdir(parents=True, exist_ok=True)
|
| 770 |
+
shutil.copyfile(local_log_path, final_log_path)
|
| 771 |
+
logging.info("Log file copy successful.")
|
| 772 |
+
except Exception as e:
|
| 773 |
+
logging.error(f"Failed to copy log file: {e}", exc_info=True)
|
| 774 |
+
success = False # Log copy fail is less critical but still indicate
|
| 775 |
+
else:
|
| 776 |
+
logging.debug("Local temp log empty/non-existent. Skipping log copy.")
|
| 777 |
+
|
| 778 |
+
return success
|
| 779 |
+
|
| 780 |
+
|
| 781 |
+
# Track memory across perform_periodic_merge() function calls
|
| 782 |
+
last_merge_memory_usage = 0
|
| 783 |
+
|
| 784 |
+
def perform_periodic_merge(
|
| 785 |
+
batch_save_dir: Path,
|
| 786 |
+
merged_batch_tracker: Set[str],
|
| 787 |
+
local_temp_metadata_path: Path,
|
| 788 |
+
local_temp_embeddings_path: Path,
|
| 789 |
+
final_log_path: Path,
|
| 790 |
+
local_temp_log_path: Path
|
| 791 |
+
):
|
| 792 |
+
"""
|
| 793 |
+
100% JSONL-only periodic merge with NO Parquet operations whatsoever.
|
| 794 |
+
Only merges to JSONL files, conversion to Parquet happens separately at the end.
|
| 795 |
+
"""
|
| 796 |
+
global last_merge_memory_usage
|
| 797 |
+
|
| 798 |
+
# Track memory at function start
|
| 799 |
+
process = psutil.Process()
|
| 800 |
+
start_memory = process.memory_info().rss / (1024 * 1024)
|
| 801 |
+
logging.info(f"Starting JSONL-only periodic merge. Current memory: {start_memory:.2f} MB")
|
| 802 |
+
|
| 803 |
+
# Define paths for working JSONL files (strip .parquet suffix if present)
|
| 804 |
+
meta_jsonl_path = Path(str(local_temp_metadata_path).replace('.parquet', '.jsonl'))
|
| 805 |
+
embed_jsonl_path = Path(str(local_temp_embeddings_path).replace('.parquet', '.jsonl'))
|
| 806 |
+
|
| 807 |
+
# Find all JSONL batch files that haven't been merged yet
|
| 808 |
+
meta_batch_files = []
|
| 809 |
+
embed_batch_files = []
|
| 810 |
+
|
| 811 |
+
# Only look for JSONL batch files
|
| 812 |
+
for batch_file in batch_save_dir.glob("metadata_batch_*.jsonl"):
|
| 813 |
+
if batch_file.name not in merged_batch_tracker:
|
| 814 |
+
meta_batch_files.append(batch_file)
|
| 815 |
+
|
| 816 |
+
for batch_file in batch_save_dir.glob("embeddings_batch_*.jsonl"):
|
| 817 |
+
if batch_file.name not in merged_batch_tracker:
|
| 818 |
+
embed_batch_files.append(batch_file)
|
| 819 |
+
|
| 820 |
+
if not meta_batch_files and not embed_batch_files:
|
| 821 |
+
logging.info("No new JSONL batches to merge periodically.")
|
| 822 |
+
return 0
|
| 823 |
+
|
| 824 |
+
logging.info(f"Performing JSONL-only merge of {len(meta_batch_files)} metadata files and {len(embed_batch_files)} embedding files")
|
| 825 |
+
|
| 826 |
+
# --- Process metadata files ---
|
| 827 |
+
if meta_batch_files:
|
| 828 |
+
try:
|
| 829 |
+
# Load existing record CIDs from JSONL to avoid duplicates
|
| 830 |
+
existing_cids = set()
|
| 831 |
+
|
| 832 |
+
# Check if JSONL exists from previous run and load CIDs
|
| 833 |
+
if meta_jsonl_path.exists():
|
| 834 |
+
logging.info(f"Scanning existing JSONL for CIDs: {meta_jsonl_path}")
|
| 835 |
+
with open(meta_jsonl_path, 'r') as f:
|
| 836 |
+
for line in f:
|
| 837 |
+
try:
|
| 838 |
+
record = json.loads(line)
|
| 839 |
+
if 'record_cid' in record:
|
| 840 |
+
existing_cids.add(record['record_cid'])
|
| 841 |
+
except:
|
| 842 |
+
pass
|
| 843 |
+
logging.info(f"Found {len(existing_cids)} existing CIDs in metadata JSONL")
|
| 844 |
+
|
| 845 |
+
# Open JSONL in append mode
|
| 846 |
+
with open(meta_jsonl_path, 'a') as jsonl_out:
|
| 847 |
+
# Process each batch file
|
| 848 |
+
for batch_file in meta_batch_files:
|
| 849 |
+
try:
|
| 850 |
+
logging.info(f"Processing metadata batch: {batch_file.name}")
|
| 851 |
+
|
| 852 |
+
# Process the JSONL batch file line by line
|
| 853 |
+
new_records_count = 0
|
| 854 |
+
total_records_count = 0
|
| 855 |
+
|
| 856 |
+
with open(batch_file, 'r') as batch_in:
|
| 857 |
+
for line in batch_in:
|
| 858 |
+
total_records_count += 1
|
| 859 |
+
try:
|
| 860 |
+
record = json.loads(line)
|
| 861 |
+
# Filter out records with CIDs that already exist
|
| 862 |
+
if 'record_cid' in record and record['record_cid'] not in existing_cids:
|
| 863 |
+
# Write new record to output JSONL
|
| 864 |
+
jsonl_out.write(line)
|
| 865 |
+
# Add to existing CIDs to avoid future duplicates
|
| 866 |
+
existing_cids.add(record['record_cid'])
|
| 867 |
+
new_records_count += 1
|
| 868 |
+
except json.JSONDecodeError:
|
| 869 |
+
logging.warning(f"Could not parse JSON line in {batch_file.name}")
|
| 870 |
+
|
| 871 |
+
# Log stats
|
| 872 |
+
logging.info(f"Batch has {total_records_count} records, {new_records_count} are new")
|
| 873 |
+
|
| 874 |
+
# Mark batch as processed
|
| 875 |
+
merged_batch_tracker.add(batch_file.name)
|
| 876 |
+
|
| 877 |
+
# Clean up batch file if enabled
|
| 878 |
+
if CLEAN_AFTER_PERIODIC_MERGE:
|
| 879 |
+
try:
|
| 880 |
+
batch_file.unlink()
|
| 881 |
+
logging.debug(f"Removed processed batch file: {batch_file}")
|
| 882 |
+
except Exception as e:
|
| 883 |
+
logging.warning(f"Could not remove batch file: {e}")
|
| 884 |
+
|
| 885 |
+
# Force memory cleanup after each batch
|
| 886 |
+
gc.collect()
|
| 887 |
+
|
| 888 |
+
except Exception as e:
|
| 889 |
+
logging.error(f"Error processing batch file {batch_file}: {e}")
|
| 890 |
+
except Exception as e:
|
| 891 |
+
logging.error(f"Error in metadata merge process: {e}", exc_info=True)
|
| 892 |
+
|
| 893 |
+
# Force memory cleanup between metadata and embeddings
|
| 894 |
+
gc.collect()
|
| 895 |
+
|
| 896 |
+
# --- Process embeddings files (similar approach) ---
|
| 897 |
+
if embed_batch_files:
|
| 898 |
+
try:
|
| 899 |
+
# Load existing record CIDs from JSONL to avoid duplicates
|
| 900 |
+
existing_cids = set()
|
| 901 |
+
|
| 902 |
+
# Check if JSONL exists from previous run and load CIDs
|
| 903 |
+
if embed_jsonl_path.exists():
|
| 904 |
+
logging.info(f"Scanning existing JSONL for CIDs: {embed_jsonl_path}")
|
| 905 |
+
with open(embed_jsonl_path, 'r') as f:
|
| 906 |
+
for line in f:
|
| 907 |
+
try:
|
| 908 |
+
record = json.loads(line)
|
| 909 |
+
if 'record_cid' in record:
|
| 910 |
+
existing_cids.add(record['record_cid'])
|
| 911 |
+
except:
|
| 912 |
+
pass
|
| 913 |
+
logging.info(f"Found {len(existing_cids)} existing CIDs in embeddings JSONL")
|
| 914 |
+
|
| 915 |
+
# Open JSONL in append mode
|
| 916 |
+
with open(embed_jsonl_path, 'a') as jsonl_out:
|
| 917 |
+
# Process each batch file
|
| 918 |
+
for batch_file in embed_batch_files:
|
| 919 |
+
try:
|
| 920 |
+
logging.info(f"Processing embeddings batch: {batch_file.name}")
|
| 921 |
+
|
| 922 |
+
# Process the JSONL batch file line by line
|
| 923 |
+
new_records_count = 0
|
| 924 |
+
total_records_count = 0
|
| 925 |
+
|
| 926 |
+
with open(batch_file, 'r') as batch_in:
|
| 927 |
+
for line in batch_in:
|
| 928 |
+
total_records_count += 1
|
| 929 |
+
try:
|
| 930 |
+
record = json.loads(line)
|
| 931 |
+
# Filter out records with CIDs that already exist
|
| 932 |
+
if 'record_cid' in record and record['record_cid'] not in existing_cids:
|
| 933 |
+
# Write new record to output JSONL
|
| 934 |
+
jsonl_out.write(line)
|
| 935 |
+
# Add to existing CIDs to avoid future duplicates
|
| 936 |
+
existing_cids.add(record['record_cid'])
|
| 937 |
+
new_records_count += 1
|
| 938 |
+
except json.JSONDecodeError:
|
| 939 |
+
logging.warning(f"Could not parse JSON line in {batch_file.name}")
|
| 940 |
+
|
| 941 |
+
# Log stats
|
| 942 |
+
logging.info(f"Batch has {total_records_count} records, {new_records_count} are new")
|
| 943 |
+
|
| 944 |
+
# Mark batch as processed
|
| 945 |
+
merged_batch_tracker.add(batch_file.name)
|
| 946 |
+
|
| 947 |
+
# Clean up batch file if enabled
|
| 948 |
+
if CLEAN_AFTER_PERIODIC_MERGE:
|
| 949 |
+
try:
|
| 950 |
+
batch_file.unlink()
|
| 951 |
+
logging.debug(f"Removed processed batch file: {batch_file}")
|
| 952 |
+
except Exception as e:
|
| 953 |
+
logging.warning(f"Could not remove batch file: {e}")
|
| 954 |
+
|
| 955 |
+
# Force memory cleanup after each batch
|
| 956 |
+
gc.collect()
|
| 957 |
+
|
| 958 |
+
except Exception as e:
|
| 959 |
+
logging.error(f"Error processing batch file {batch_file}: {e}")
|
| 960 |
+
except Exception as e:
|
| 961 |
+
logging.error(f"Error in embeddings merge process: {e}", exc_info=True)
|
| 962 |
+
|
| 963 |
+
# -- Only sync the log file, no Parquet files during runtime --
|
| 964 |
+
try:
|
| 965 |
+
if local_temp_log_path.is_file() and local_temp_log_path.stat().st_size > 0:
|
| 966 |
+
final_log_path.parent.mkdir(parents=True, exist_ok=True)
|
| 967 |
+
shutil.copyfile(local_temp_log_path, final_log_path)
|
| 968 |
+
logging.info("Log file sync successful.")
|
| 969 |
+
except Exception as e:
|
| 970 |
+
logging.error(f"Failed to sync log file: {e}")
|
| 971 |
+
|
| 972 |
+
# Final cleanup
|
| 973 |
+
for _ in range(3):
|
| 974 |
+
gc.collect()
|
| 975 |
+
|
| 976 |
+
# Update memory tracking for next call
|
| 977 |
+
end_memory = process.memory_info().rss / (1024 * 1024)
|
| 978 |
+
logging.info(f"Memory at end of merge: {end_memory:.2f} MB (Change: {end_memory - start_memory:.2f} MB)")
|
| 979 |
+
last_merge_memory_usage = end_memory
|
| 980 |
+
|
| 981 |
+
return len(meta_batch_files) + len(embed_batch_files)
|
| 982 |
+
|
| 983 |
+
|
| 984 |
+
|
| 985 |
+
# --- UPDATED: Main Embedding Generation Function with CID-based Primary Key ---
|
| 986 |
+
def create_embedding_dataset(
|
| 987 |
+
input_jsonl_filepath: Path,
|
| 988 |
+
final_metadata_parquet_path: Path,
|
| 989 |
+
final_embeddings_parquet_path: Path,
|
| 990 |
+
local_temp_metadata_path: Path,
|
| 991 |
+
local_temp_embeddings_path: Path,
|
| 992 |
+
local_temp_log_path: Path,
|
| 993 |
+
final_log_filepath: Path,
|
| 994 |
+
max_records: Optional[int] = None,
|
| 995 |
+
batch_size: int = 32,
|
| 996 |
+
embedding_model_name: str = EMBEDDING_MODEL_NAME,
|
| 997 |
+
process_config: bool = PROCESS_CONFIG_JSON,
|
| 998 |
+
process_readme: bool = PROCESS_README_CONTENT,
|
| 999 |
+
):
|
| 1000 |
+
"""
|
| 1001 |
+
JSONL-only workflow that reads metadata, generates CIDs & embeddings,
|
| 1002 |
+
and saves all outputs as JSONL until the very end.
|
| 1003 |
+
"""
|
| 1004 |
+
# --- Setup batch directory ---
|
| 1005 |
+
batch_save_dir = LOCAL_WORK_DIR / BATCH_SAVE_DIR_NAME
|
| 1006 |
+
batch_save_dir.mkdir(parents=True, exist_ok=True)
|
| 1007 |
+
|
| 1008 |
+
# --- Define JSONL paths by converting Parquet paths ---
|
| 1009 |
+
meta_jsonl_path = Path(str(local_temp_metadata_path).replace('.parquet', '.jsonl'))
|
| 1010 |
+
embed_jsonl_path = Path(str(local_temp_embeddings_path).replace('.parquet', '.jsonl'))
|
| 1011 |
+
|
| 1012 |
+
# --- Configure logging to use the local temp log file ---
|
| 1013 |
+
log_file_handler = logging.FileHandler(local_temp_log_path)
|
| 1014 |
+
log_stream_handler = logging.StreamHandler()
|
| 1015 |
+
for handler in logging.root.handlers[:]: logging.root.removeHandler(handler)
|
| 1016 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s', handlers=[log_file_handler, log_stream_handler])
|
| 1017 |
+
logging.getLogger('huggingface_hub.repocard_data').setLevel(logging.ERROR)
|
| 1018 |
+
|
| 1019 |
+
# --- Log configuration ---
|
| 1020 |
+
logging.info(f"--- Starting Embedding Generation with JSONL-only workflow ---")
|
| 1021 |
+
logging.info(f"Input JSONL: '{input_jsonl_filepath}'")
|
| 1022 |
+
logging.info(f"Metadata JSONL Output: '{meta_jsonl_path}'")
|
| 1023 |
+
logging.info(f"Embeddings JSONL Output: '{embed_jsonl_path}'")
|
| 1024 |
+
logging.info(f"Final Metadata Parquet Output (post-processing): '{final_metadata_parquet_path}'")
|
| 1025 |
+
logging.info(f"Final Embeddings Parquet Output (post-processing): '{final_embeddings_parquet_path}'")
|
| 1026 |
+
logging.info(f"Batch Save Directory: '{batch_save_dir}'")
|
| 1027 |
+
logging.info(f"Batch Save Threshold: {BATCH_SAVE_THRESHOLD}")
|
| 1028 |
+
logging.info(f"Periodic Merge Frequency: {PERIODIC_MERGE_FREQUENCY} batches")
|
| 1029 |
+
logging.info(f"Clean After Periodic Merge: {CLEAN_AFTER_PERIODIC_MERGE}")
|
| 1030 |
+
logging.info(f"Local Temp Log: '{local_temp_log_path}'")
|
| 1031 |
+
logging.info(f"Final Log Output: '{final_log_filepath}'")
|
| 1032 |
+
logging.info(f"Embedding Model: '{embedding_model_name}', Batch Size: {batch_size}")
|
| 1033 |
+
logging.info(f"Process Config: {process_config}, Process README: {process_readme}")
|
| 1034 |
+
logging.info(f"Max Records: {'All' if max_records is None else max_records}")
|
| 1035 |
+
|
| 1036 |
+
# --- Load Embedding Model ---
|
| 1037 |
+
try:
|
| 1038 |
+
logging.info(f"Loading embedding model: {embedding_model_name}")
|
| 1039 |
+
|
| 1040 |
+
# Check for MPS (Apple Silicon GPU) availability first, then CUDA, then fall back to CPU
|
| 1041 |
+
if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
|
| 1042 |
+
device = 'mps'
|
| 1043 |
+
logging.info(f"Using Apple Silicon GPU (MPS)")
|
| 1044 |
+
elif torch.cuda.is_available():
|
| 1045 |
+
device = 'cuda'
|
| 1046 |
+
logging.info(f"Using NVIDIA GPU (CUDA)")
|
| 1047 |
+
else:
|
| 1048 |
+
device = 'cpu'
|
| 1049 |
+
logging.info(f"Using CPU (no GPU acceleration available)")
|
| 1050 |
+
|
| 1051 |
+
model = SentenceTransformer(embedding_model_name, device=device)
|
| 1052 |
+
cid_generator = ipfs_multiformats_py() # Initialize CID generator
|
| 1053 |
+
logging.info("Embedding model & CID generator loaded.")
|
| 1054 |
+
except Exception as e:
|
| 1055 |
+
logging.error(f"Failed to load embedding model or init CID generator: {e}", exc_info=True)
|
| 1056 |
+
return None, None # Return None for both paths
|
| 1057 |
+
|
| 1058 |
+
# --- Load processed CIDs from existing JSONL files ---
|
| 1059 |
+
processed_cids = set()
|
| 1060 |
+
|
| 1061 |
+
# 1. Check the main embeddings JSONL file
|
| 1062 |
+
if embed_jsonl_path.exists():
|
| 1063 |
+
logging.info(f"Found existing embeddings JSONL: {embed_jsonl_path}. Loading processed CIDs...")
|
| 1064 |
+
try:
|
| 1065 |
+
with open(embed_jsonl_path, 'r') as f:
|
| 1066 |
+
for line in f:
|
| 1067 |
+
try:
|
| 1068 |
+
record = json.loads(line)
|
| 1069 |
+
if 'record_cid' in record:
|
| 1070 |
+
processed_cids.add(record['record_cid'])
|
| 1071 |
+
except:
|
| 1072 |
+
pass
|
| 1073 |
+
logging.info(f"Loaded {len(processed_cids)} CIDs from existing embeddings JSONL.")
|
| 1074 |
+
except Exception as e:
|
| 1075 |
+
logging.warning(f"Could not load CIDs from '{embed_jsonl_path}': {e}")
|
| 1076 |
+
|
| 1077 |
+
# 2. Check batch files
|
| 1078 |
+
batch_files = list(batch_save_dir.glob("embeddings_batch_*.jsonl"))
|
| 1079 |
+
if batch_files:
|
| 1080 |
+
logging.info(f"Found {len(batch_files)} embedding batch JSONL files.")
|
| 1081 |
+
batch_cids_count = 0
|
| 1082 |
+
|
| 1083 |
+
for batch_file in batch_files:
|
| 1084 |
+
try:
|
| 1085 |
+
with open(batch_file, 'r') as f:
|
| 1086 |
+
for line in f:
|
| 1087 |
+
try:
|
| 1088 |
+
record = json.loads(line)
|
| 1089 |
+
if 'record_cid' in record:
|
| 1090 |
+
processed_cids.add(record['record_cid'])
|
| 1091 |
+
batch_cids_count += 1
|
| 1092 |
+
except:
|
| 1093 |
+
pass
|
| 1094 |
+
except Exception as e:
|
| 1095 |
+
logging.warning(f"Error loading CIDs from batch file {batch_file}: {e}")
|
| 1096 |
+
|
| 1097 |
+
logging.info(f"Loaded {batch_cids_count} additional CIDs from JSONL batch files.")
|
| 1098 |
+
|
| 1099 |
+
initial_processed_count = len(processed_cids)
|
| 1100 |
+
logging.info(f"Resuming from {initial_processed_count} records already processed.")
|
| 1101 |
+
|
| 1102 |
+
# --- Batch saving and periodic merge setup ---
|
| 1103 |
+
batch_counter = 0
|
| 1104 |
+
records_since_last_save = 0
|
| 1105 |
+
merged_batch_tracker = set() # Track which batch files have been merged
|
| 1106 |
+
|
| 1107 |
+
# Keep a lookup of model_id to record_cid for this session
|
| 1108 |
+
model_id_to_record_cid = {}
|
| 1109 |
+
|
| 1110 |
+
# --- Process JSONL File ---
|
| 1111 |
+
metadata_records_list = [] # Holds dicts for metadata
|
| 1112 |
+
embeddings_records_list = [] # Holds dicts for embeddings
|
| 1113 |
+
batch_inputs = [] # Holds tuples for batch processing
|
| 1114 |
+
record_count_from_jsonl = 0; processed_count_this_run = 0; skipped_resume_count = 0; skipped_error_count = 0
|
| 1115 |
+
start_time = None
|
| 1116 |
+
|
| 1117 |
+
try:
|
| 1118 |
+
logging.info(f"Opening input JSONL file: {input_jsonl_filepath}")
|
| 1119 |
+
start_time = time.time()
|
| 1120 |
+
with input_jsonl_filepath.open('r', encoding='utf-8') as f_jsonl:
|
| 1121 |
+
pbar = tqdm(f_jsonl, desc="Processing JSONL", unit="record")
|
| 1122 |
+
for line in pbar:
|
| 1123 |
+
record_count_from_jsonl += 1
|
| 1124 |
+
if max_records is not None and processed_count_this_run >= max_records:
|
| 1125 |
+
logging.info(f"Reached max_records limit ({max_records}). Stopping.");
|
| 1126 |
+
break
|
| 1127 |
+
|
| 1128 |
+
try:
|
| 1129 |
+
line = line.strip()
|
| 1130 |
+
if not line: continue
|
| 1131 |
+
data = json.loads(line) # Original metadata dictionary
|
| 1132 |
+
model_id = data.get('id')
|
| 1133 |
+
if not model_id or not isinstance(model_id, str):
|
| 1134 |
+
logging.warning(f"Skip record {record_count_from_jsonl}: missing/invalid 'id'.");
|
| 1135 |
+
skipped_error_count += 1;
|
| 1136 |
+
continue
|
| 1137 |
+
|
| 1138 |
+
# --- Extract text for embedding & CID generation ---
|
| 1139 |
+
config_text = ""; config_cid = None; config_dict_or_str = data.get('config')
|
| 1140 |
+
if process_config and config_dict_or_str is not None:
|
| 1141 |
+
if isinstance(config_dict_or_str, dict):
|
| 1142 |
+
try:
|
| 1143 |
+
config_text = json.dumps(config_dict_or_str, separators=(',', ':'));
|
| 1144 |
+
config_cid = cid_generator.get_cid(config_text) # Use compact string for CID
|
| 1145 |
+
except TypeError:
|
| 1146 |
+
logging.warning(f"Cannot serialize config for {model_id}. Skip CID/embed.")
|
| 1147 |
+
elif isinstance(config_dict_or_str, str): # Handle if config is already a string
|
| 1148 |
+
config_text = config_dict_or_str;
|
| 1149 |
+
config_cid = cid_generator.get_cid(config_text)
|
| 1150 |
+
else:
|
| 1151 |
+
logging.warning(f"Config for {model_id} type {type(config_dict_or_str)}. Skip CID/embed.")
|
| 1152 |
+
|
| 1153 |
+
readme_text = ""; readme_cid = None
|
| 1154 |
+
if process_readme:
|
| 1155 |
+
card_data = data.get('cardData')
|
| 1156 |
+
if isinstance(card_data, dict):
|
| 1157 |
+
readme_text = card_data.get('text', '') or ''
|
| 1158 |
+
elif isinstance(card_data, str):
|
| 1159 |
+
readme_text = card_data # If cardData itself is the string
|
| 1160 |
+
if not readme_text and isinstance(data.get('description'), str):
|
| 1161 |
+
readme_text = data['description'] # Fallback
|
| 1162 |
+
if readme_text:
|
| 1163 |
+
readme_cid = cid_generator.get_cid(readme_text)
|
| 1164 |
+
|
| 1165 |
+
# --- Generate record_cid (primary key) ---
|
| 1166 |
+
record_cid = generate_record_cid(cid_generator, model_id, config_cid, readme_cid)
|
| 1167 |
+
|
| 1168 |
+
# Store in lookup for future reference
|
| 1169 |
+
model_id_to_record_cid[model_id] = record_cid
|
| 1170 |
+
|
| 1171 |
+
# Skip if this record_cid has already been processed
|
| 1172 |
+
if record_cid in processed_cids:
|
| 1173 |
+
skipped_resume_count += 1
|
| 1174 |
+
continue
|
| 1175 |
+
|
| 1176 |
+
processed_count_this_run += 1
|
| 1177 |
+
pbar.set_postfix_str(f"Batching: {model_id}", refresh=True)
|
| 1178 |
+
|
| 1179 |
+
# Add to batch for embedding
|
| 1180 |
+
batch_inputs.append((data, config_text, readme_text, config_cid, readme_cid, record_cid))
|
| 1181 |
+
|
| 1182 |
+
# --- Process Batch when full ---
|
| 1183 |
+
if len(batch_inputs) >= batch_size:
|
| 1184 |
+
pbar.set_postfix_str(f"Embedding batch ({len(batch_inputs)})...", refresh=True)
|
| 1185 |
+
try:
|
| 1186 |
+
original_data_batch = [item[0] for item in batch_inputs]
|
| 1187 |
+
config_texts_batch = [item[1] for item in batch_inputs]
|
| 1188 |
+
readme_texts_batch = [item[2] for item in batch_inputs]
|
| 1189 |
+
config_cids_batch = [item[3] for item in batch_inputs]
|
| 1190 |
+
readme_cids_batch = [item[4] for item in batch_inputs]
|
| 1191 |
+
record_cids_batch = [item[5] for item in batch_inputs]
|
| 1192 |
+
|
| 1193 |
+
# Generate embeddings
|
| 1194 |
+
config_embeddings = model.encode(config_texts_batch, batch_size=batch_size, show_progress_bar=False) if process_config else [None] * len(batch_inputs)
|
| 1195 |
+
readme_embeddings = model.encode(readme_texts_batch, batch_size=batch_size, show_progress_bar=False) if process_readme else [None] * len(batch_inputs)
|
| 1196 |
+
|
| 1197 |
+
# --- Create records for BOTH data formats ---
|
| 1198 |
+
for i, original_data in enumerate(original_data_batch):
|
| 1199 |
+
current_model_id = original_data.get('id')
|
| 1200 |
+
current_record_cid = record_cids_batch[i]
|
| 1201 |
+
if not current_model_id or not current_record_cid: continue
|
| 1202 |
+
|
| 1203 |
+
# 1. Metadata Record
|
| 1204 |
+
metadata_record = original_data.copy() # Start with all original metadata
|
| 1205 |
+
# Remove bulky/embedded fields if they exist, keep CIDs
|
| 1206 |
+
metadata_record.pop('config_embedding', None)
|
| 1207 |
+
metadata_record.pop('readme_embedding', None)
|
| 1208 |
+
# Add CIDs
|
| 1209 |
+
metadata_record['record_cid'] = current_record_cid # Primary key
|
| 1210 |
+
if process_config: metadata_record['config_cid'] = config_cids_batch[i]
|
| 1211 |
+
if process_readme: metadata_record['readme_cid'] = readme_cids_batch[i]
|
| 1212 |
+
metadata_records_list.append(metadata_record)
|
| 1213 |
+
|
| 1214 |
+
# 2. Embedding Record
|
| 1215 |
+
embedding_record = {
|
| 1216 |
+
'record_cid': current_record_cid, # Primary key
|
| 1217 |
+
'model_id': current_model_id # Keep model_id for reference
|
| 1218 |
+
}
|
| 1219 |
+
if process_config:
|
| 1220 |
+
embedding_record['config_embedding'] = config_embeddings[i].tolist() if config_texts_batch[i] else None
|
| 1221 |
+
if process_readme:
|
| 1222 |
+
embedding_record['readme_embedding'] = readme_embeddings[i].tolist() if readme_texts_batch[i] else None
|
| 1223 |
+
embeddings_records_list.append(embedding_record)
|
| 1224 |
+
|
| 1225 |
+
# Mark this record as processed to avoid reprocessing if script restarts
|
| 1226 |
+
processed_cids.add(current_record_cid)
|
| 1227 |
+
|
| 1228 |
+
# Increment counter for batch saving
|
| 1229 |
+
records_since_last_save += 1
|
| 1230 |
+
|
| 1231 |
+
logging.debug(f"Processed batch. Metadata size: {len(metadata_records_list)}, Embeddings size: {len(embeddings_records_list)}")
|
| 1232 |
+
|
| 1233 |
+
# --- Save batch if we've reached the threshold ---
|
| 1234 |
+
if records_since_last_save >= BATCH_SAVE_THRESHOLD:
|
| 1235 |
+
batch_counter += 1
|
| 1236 |
+
timestamp = int(time.time())
|
| 1237 |
+
|
| 1238 |
+
# Save metadata batch as JSONL
|
| 1239 |
+
meta_batch_file = batch_save_dir / f"metadata_batch_{batch_counter}_{timestamp}.jsonl"
|
| 1240 |
+
success_meta = True
|
| 1241 |
+
try:
|
| 1242 |
+
with open(meta_batch_file, 'w') as f:
|
| 1243 |
+
for record in metadata_records_list:
|
| 1244 |
+
f.write(json.dumps(safe_serialize_dict(record)) + '\n')
|
| 1245 |
+
logging.info(f"Saved metadata batch {batch_counter} as JSONL with {len(metadata_records_list)} records")
|
| 1246 |
+
except Exception as e:
|
| 1247 |
+
logging.error(f"Error saving metadata batch as JSONL: {e}")
|
| 1248 |
+
success_meta = False
|
| 1249 |
+
|
| 1250 |
+
# Save embeddings batch as JSONL
|
| 1251 |
+
embed_batch_file = batch_save_dir / f"embeddings_batch_{batch_counter}_{timestamp}.jsonl"
|
| 1252 |
+
success_embed = True
|
| 1253 |
+
try:
|
| 1254 |
+
with open(embed_batch_file, 'w') as f:
|
| 1255 |
+
for record in embeddings_records_list:
|
| 1256 |
+
f.write(json.dumps(safe_serialize_dict(record)) + '\n')
|
| 1257 |
+
logging.info(f"Saved embeddings batch {batch_counter} as JSONL with {len(embeddings_records_list)} records")
|
| 1258 |
+
except Exception as e:
|
| 1259 |
+
logging.error(f"Error saving embeddings batch as JSONL: {e}")
|
| 1260 |
+
success_embed = False
|
| 1261 |
+
|
| 1262 |
+
if success_meta and success_embed:
|
| 1263 |
+
logging.info(f"Saved batch {batch_counter} with {len(embeddings_records_list)} records")
|
| 1264 |
+
else:
|
| 1265 |
+
logging.warning(f"Batch {batch_counter} save had issues. Check logs.")
|
| 1266 |
+
|
| 1267 |
+
# Clear the lists to start a new batch and reset counter
|
| 1268 |
+
metadata_records_list = []
|
| 1269 |
+
embeddings_records_list = []
|
| 1270 |
+
records_since_last_save = 0
|
| 1271 |
+
|
| 1272 |
+
# --- Periodic merge to final JSONL files ---
|
| 1273 |
+
if PERIODIC_MERGE_FREQUENCY > 0 and batch_counter % PERIODIC_MERGE_FREQUENCY == 0:
|
| 1274 |
+
pbar.set_postfix_str(f"Periodic merge to JSONL...", refresh=True)
|
| 1275 |
+
batches_merged = perform_periodic_merge(
|
| 1276 |
+
batch_save_dir=batch_save_dir,
|
| 1277 |
+
merged_batch_tracker=merged_batch_tracker,
|
| 1278 |
+
local_temp_metadata_path=local_temp_metadata_path,
|
| 1279 |
+
local_temp_embeddings_path=local_temp_embeddings_path,
|
| 1280 |
+
final_log_path=final_log_filepath,
|
| 1281 |
+
local_temp_log_path=local_temp_log_path
|
| 1282 |
+
)
|
| 1283 |
+
pbar.set_postfix_str(f"Merged {batches_merged} batches to JSONL", refresh=True)
|
| 1284 |
+
|
| 1285 |
+
except Exception as e_embed:
|
| 1286 |
+
logging.error(f"Error embedding batch: {e_embed}", exc_info=True)
|
| 1287 |
+
skipped_error_count += len(batch_inputs) # Count whole batch as skipped
|
| 1288 |
+
|
| 1289 |
+
batch_inputs = [] # Clear batch
|
| 1290 |
+
|
| 1291 |
+
# Handle line processing errors
|
| 1292 |
+
except json.JSONDecodeError:
|
| 1293 |
+
logging.warning(f"Skip record {record_count_from_jsonl}: JSON decode error.");
|
| 1294 |
+
skipped_error_count += 1
|
| 1295 |
+
except Exception as e_line:
|
| 1296 |
+
logging.error(f"Skip record {record_count_from_jsonl}: Error - {e_line}", exc_info=False);
|
| 1297 |
+
skipped_error_count += 1
|
| 1298 |
+
# --- End reading JSONL file ---
|
| 1299 |
+
|
| 1300 |
+
# --- Process Final Remaining Batch ---
|
| 1301 |
+
if batch_inputs:
|
| 1302 |
+
pbar.set_postfix_str(f"Embedding final batch ({len(batch_inputs)})...", refresh=True)
|
| 1303 |
+
try:
|
| 1304 |
+
# Process just like the main batch
|
| 1305 |
+
original_data_batch = [item[0] for item in batch_inputs]
|
| 1306 |
+
config_texts_batch = [item[1] for item in batch_inputs]
|
| 1307 |
+
readme_texts_batch = [item[2] for item in batch_inputs]
|
| 1308 |
+
config_cids_batch = [item[3] for item in batch_inputs]
|
| 1309 |
+
readme_cids_batch = [item[4] for item in batch_inputs]
|
| 1310 |
+
record_cids_batch = [item[5] for item in batch_inputs]
|
| 1311 |
+
|
| 1312 |
+
config_embeddings = model.encode(config_texts_batch, batch_size=batch_size, show_progress_bar=False) if process_config else [None] * len(batch_inputs)
|
| 1313 |
+
readme_embeddings = model.encode(readme_texts_batch, batch_size=batch_size, show_progress_bar=False) if process_readme else [None] * len(batch_inputs)
|
| 1314 |
+
|
| 1315 |
+
for i, original_data in enumerate(original_data_batch):
|
| 1316 |
+
current_model_id = original_data.get('id')
|
| 1317 |
+
current_record_cid = record_cids_batch[i]
|
| 1318 |
+
if not current_model_id or not current_record_cid: continue
|
| 1319 |
+
|
| 1320 |
+
metadata_record = original_data.copy()
|
| 1321 |
+
metadata_record.pop('config_embedding', None)
|
| 1322 |
+
metadata_record.pop('readme_embedding', None)
|
| 1323 |
+
metadata_record['record_cid'] = current_record_cid # Primary key
|
| 1324 |
+
if process_config: metadata_record['config_cid'] = config_cids_batch[i]
|
| 1325 |
+
if process_readme: metadata_record['readme_cid'] = readme_cids_batch[i]
|
| 1326 |
+
metadata_records_list.append(metadata_record)
|
| 1327 |
+
|
| 1328 |
+
embedding_record = {
|
| 1329 |
+
'record_cid': current_record_cid, # Primary key
|
| 1330 |
+
'model_id': current_model_id # Keep model_id for reference
|
| 1331 |
+
}
|
| 1332 |
+
if process_config:
|
| 1333 |
+
embedding_record['config_embedding'] = config_embeddings[i].tolist() if config_texts_batch[i] else None
|
| 1334 |
+
if process_readme:
|
| 1335 |
+
embedding_record['readme_embedding'] = readme_embeddings[i].tolist() if readme_texts_batch[i] else None
|
| 1336 |
+
embeddings_records_list.append(embedding_record)
|
| 1337 |
+
|
| 1338 |
+
# Mark as processed
|
| 1339 |
+
processed_cids.add(current_record_cid)
|
| 1340 |
+
records_since_last_save += 1
|
| 1341 |
+
|
| 1342 |
+
logging.debug(f"Processed final batch. Metadata size: {len(metadata_records_list)}, Embeddings size: {len(embeddings_records_list)}")
|
| 1343 |
+
except Exception as e_embed_final:
|
| 1344 |
+
logging.error(f"Error embedding final batch: {e_embed_final}", exc_info=True)
|
| 1345 |
+
skipped_error_count += len(batch_inputs)
|
| 1346 |
+
# --- End processing batches ---
|
| 1347 |
+
|
| 1348 |
+
# --- Save any remaining records as a final batch ---
|
| 1349 |
+
if metadata_records_list:
|
| 1350 |
+
batch_counter += 1
|
| 1351 |
+
timestamp = int(time.time())
|
| 1352 |
+
|
| 1353 |
+
# Save final metadata batch as JSONL
|
| 1354 |
+
meta_batch_file = batch_save_dir / f"metadata_batch_{batch_counter}_{timestamp}.jsonl"
|
| 1355 |
+
success_meta = True
|
| 1356 |
+
try:
|
| 1357 |
+
with open(meta_batch_file, 'w') as f:
|
| 1358 |
+
for record in metadata_records_list:
|
| 1359 |
+
f.write(json.dumps(safe_serialize_dict(record)) + '\n')
|
| 1360 |
+
logging.info(f"Saved final metadata batch as JSONL with {len(metadata_records_list)} records")
|
| 1361 |
+
except Exception as e:
|
| 1362 |
+
logging.error(f"Error saving final metadata batch as JSONL: {e}")
|
| 1363 |
+
success_meta = False
|
| 1364 |
+
|
| 1365 |
+
# Save final embeddings batch as JSONL
|
| 1366 |
+
embed_batch_file = batch_save_dir / f"embeddings_batch_{batch_counter}_{timestamp}.jsonl"
|
| 1367 |
+
success_embed = True
|
| 1368 |
+
try:
|
| 1369 |
+
with open(embed_batch_file, 'w') as f:
|
| 1370 |
+
for record in embeddings_records_list:
|
| 1371 |
+
f.write(json.dumps(safe_serialize_dict(record)) + '\n')
|
| 1372 |
+
logging.info(f"Saved final embeddings batch as JSONL with {len(embeddings_records_list)} records")
|
| 1373 |
+
except Exception as e:
|
| 1374 |
+
logging.error(f"Error saving final embeddings batch as JSONL: {e}")
|
| 1375 |
+
success_embed = False
|
| 1376 |
+
|
| 1377 |
+
if success_meta and success_embed:
|
| 1378 |
+
logging.info(f"Saved final batch {batch_counter} with {len(embeddings_records_list)} records")
|
| 1379 |
+
else:
|
| 1380 |
+
logging.warning(f"Final batch {batch_counter} save had issues. Check logs.")
|
| 1381 |
+
|
| 1382 |
+
# Clear lists
|
| 1383 |
+
metadata_records_list = []
|
| 1384 |
+
embeddings_records_list = []
|
| 1385 |
+
|
| 1386 |
+
pbar.close()
|
| 1387 |
+
logging.info("Finished processing records from JSONL.")
|
| 1388 |
+
|
| 1389 |
+
# --- Merge all remaining batches into the final JSONL files ---
|
| 1390 |
+
# Process any remaining batches that haven't been merged
|
| 1391 |
+
if PERIODIC_MERGE_FREQUENCY > 0:
|
| 1392 |
+
logging.info("Performing final merge of any remaining batches...")
|
| 1393 |
+
batches_merged = perform_periodic_merge(
|
| 1394 |
+
batch_save_dir=batch_save_dir,
|
| 1395 |
+
merged_batch_tracker=merged_batch_tracker,
|
| 1396 |
+
local_temp_metadata_path=local_temp_metadata_path,
|
| 1397 |
+
local_temp_embeddings_path=local_temp_embeddings_path,
|
| 1398 |
+
final_log_path=final_log_filepath,
|
| 1399 |
+
local_temp_log_path=local_temp_log_path
|
| 1400 |
+
)
|
| 1401 |
+
logging.info(f"Final merge: processed {batches_merged} remaining batches")
|
| 1402 |
+
|
| 1403 |
+
# Return the JSONL paths for final conversion
|
| 1404 |
+
return meta_jsonl_path, embed_jsonl_path
|
| 1405 |
+
|
| 1406 |
+
# Handle file/main processing errors
|
| 1407 |
+
except FileNotFoundError:
|
| 1408 |
+
logging.error(f"CRITICAL: Input JSONL file not found: {input_jsonl_filepath}.");
|
| 1409 |
+
return None, None
|
| 1410 |
+
except Exception as e_main:
|
| 1411 |
+
logging.error(f"CRITICAL error: {e_main}", exc_info=True);
|
| 1412 |
+
return None, None
|
| 1413 |
+
|
| 1414 |
+
# --- Final Summary ---
|
| 1415 |
+
finally:
|
| 1416 |
+
total_processed_in_run = processed_count_this_run
|
| 1417 |
+
total_batches_saved = batch_counter
|
| 1418 |
+
total_batches_merged = len(merged_batch_tracker)
|
| 1419 |
+
|
| 1420 |
+
logging.info("--- Embedding Generation Summary ---")
|
| 1421 |
+
logging.info(f"Records read from JSONL: {record_count_from_jsonl}")
|
| 1422 |
+
logging.info(f"Records skipped (resume): {skipped_resume_count}")
|
| 1423 |
+
logging.info(f"Records processed this run: {total_processed_in_run}")
|
| 1424 |
+
logging.info(f"Records skipped (errors): {skipped_error_count}")
|
| 1425 |
+
logging.info(f"Total batches saved: {total_batches_saved}")
|
| 1426 |
+
logging.info(f"Total batches merged: {total_batches_merged}")
|
| 1427 |
+
logging.info(f"Total unique records processed (including previous runs): {len(processed_cids)}")
|
| 1428 |
+
if start_time:
|
| 1429 |
+
logging.info(f"Total processing time: {time.time() - start_time:.2f} seconds")
|
| 1430 |
+
logging.info("------------------------------------")
|
| 1431 |
+
|
| 1432 |
+
|
| 1433 |
+
|
| 1434 |
+
# --- Upload Function (Modified for two files) ---
|
| 1435 |
+
def upload_files_to_hub(
|
| 1436 |
+
local_metadata_path: Path,
|
| 1437 |
+
local_embeddings_path: Path,
|
| 1438 |
+
repo_id: str,
|
| 1439 |
+
repo_type: str = "dataset",
|
| 1440 |
+
metadata_path_in_repo: Optional[str] = None,
|
| 1441 |
+
embeddings_path_in_repo: Optional[str] = None,
|
| 1442 |
+
hf_token: Union[str, bool, None] = None
|
| 1443 |
+
):
|
| 1444 |
+
"""Uploads the generated Parquet files to the Hugging Face Hub."""
|
| 1445 |
+
api = HfApi(token=hf_token)
|
| 1446 |
+
uploaded_meta = False
|
| 1447 |
+
uploaded_embed = False
|
| 1448 |
+
|
| 1449 |
+
# Upload Metadata (Parquet or CSV)
|
| 1450 |
+
if local_metadata_path and local_metadata_path.exists():
|
| 1451 |
+
path_in_repo_meta = metadata_path_in_repo or local_metadata_path.name
|
| 1452 |
+
logging.info(f"Uploading Metadata: {local_metadata_path} to {repo_id} as {path_in_repo_meta}...")
|
| 1453 |
+
try:
|
| 1454 |
+
api.upload_file(
|
| 1455 |
+
path_or_fileobj=str(local_metadata_path), path_in_repo=path_in_repo_meta, repo_id=repo_id, repo_type=repo_type,
|
| 1456 |
+
commit_message=f"Update metadata ({local_metadata_path.suffix}) {time.strftime('%Y-%m-%d %H:%M:%S')}"
|
| 1457 |
+
); logging.info("Metadata upload successful."); uploaded_meta = True
|
| 1458 |
+
except Exception as e: logging.error(f"Metadata upload failed: {e}", exc_info=True)
|
| 1459 |
+
else: logging.warning("Local metadata file not found or not specified. Skipping metadata upload.")
|
| 1460 |
+
|
| 1461 |
+
# Upload Embeddings (Parquet or CSV)
|
| 1462 |
+
if local_embeddings_path and local_embeddings_path.exists():
|
| 1463 |
+
path_in_repo_embed = embeddings_path_in_repo or local_embeddings_path.name
|
| 1464 |
+
logging.info(f"Uploading Embeddings: {local_embeddings_path} to {repo_id} as {path_in_repo_embed}...")
|
| 1465 |
+
try:
|
| 1466 |
+
api.upload_file(
|
| 1467 |
+
path_or_fileobj=str(local_embeddings_path), path_in_repo=path_in_repo_embed, repo_id=repo_id, repo_type=repo_type,
|
| 1468 |
+
commit_message=f"Update embeddings ({local_embeddings_path.suffix}) {time.strftime('%Y-%m-%d %H:%M:%S')}"
|
| 1469 |
+
); logging.info("Embeddings upload successful."); uploaded_embed = True
|
| 1470 |
+
except Exception as e: logging.error(f"Embeddings upload failed: {e}", exc_info=True)
|
| 1471 |
+
else: logging.warning("Local embeddings file not found or not specified. Skipping embeddings upload.")
|
| 1472 |
+
|
| 1473 |
+
return uploaded_meta and uploaded_embed # Return overall success
|
| 1474 |
+
|
| 1475 |
+
# --- Script Execution (`if __name__ == "__main__":`) ---
|
| 1476 |
+
if __name__ == "__main__":
|
| 1477 |
+
# --- Determine Paths ---
|
| 1478 |
+
print("--- Determining Output Paths ---")
|
| 1479 |
+
gdrive_base = Path(GDRIVE_MOUNT_POINT); gdrive_target_dir = gdrive_base / GDRIVE_FOLDER_NAME
|
| 1480 |
+
local_fallback_dir = Path(LOCAL_FOLDER_NAME); effective_final_dir = None;
|
| 1481 |
+
print(f"Checking GDrive: {gdrive_base}");
|
| 1482 |
+
if gdrive_base.is_dir() and gdrive_base.exists():
|
| 1483 |
+
print(f"Mount OK. Checking target: {gdrive_target_dir}");
|
| 1484 |
+
|
| 1485 |
+
if gdrive_target_dir.is_dir():
|
| 1486 |
+
print(f"Target Google Drive directory found. Using Google Drive.")
|
| 1487 |
+
effective_final_dir = gdrive_target_dir
|
| 1488 |
+
else:
|
| 1489 |
+
print(f"Target Google Drive directory '{gdrive_target_dir}' not found. Will attempt to create.")
|
| 1490 |
+
try:
|
| 1491 |
+
gdrive_target_dir.mkdir(parents=True, exist_ok=True)
|
| 1492 |
+
print(f"Successfully created Google Drive directory.")
|
| 1493 |
+
effective_final_dir = gdrive_target_dir
|
| 1494 |
+
except Exception as e:
|
| 1495 |
+
print(f"Error creating Google Drive directory '{gdrive_target_dir}': {e}")
|
| 1496 |
+
print("Falling back to local directory.")
|
| 1497 |
+
effective_final_dir = local_target_dir
|
| 1498 |
+
|
| 1499 |
+
else:
|
| 1500 |
+
local_fallback_dir.mkdir(parents=True, exist_ok=True)
|
| 1501 |
+
print(f"Mount not found. Using local fallback: {local_fallback_dir}")
|
| 1502 |
+
effective_final_dir = local_fallback_dir
|
| 1503 |
+
|
| 1504 |
+
effective_final_dir.mkdir(parents=True, exist_ok=True); LOCAL_WORK_DIR.mkdir(parents=True, exist_ok=True); print(f"Effective final destination directory: {effective_final_dir}");
|
| 1505 |
+
|
| 1506 |
+
# Define final destination paths
|
| 1507 |
+
final_metadata_filepath = effective_final_dir / FINAL_METADATA_PARQUET_FILENAME
|
| 1508 |
+
final_embeddings_filepath = effective_final_dir / FINAL_EMBEDDINGS_PARQUET_FILENAME
|
| 1509 |
+
final_log_filepath = effective_final_dir / FINAL_LOG_FILENAME
|
| 1510 |
+
input_jsonl_filepath = effective_final_dir / INPUT_JSONL_FILENAME # Assume input is also in final dir
|
| 1511 |
+
|
| 1512 |
+
# Define local working paths
|
| 1513 |
+
local_temp_metadata_path = LOCAL_WORK_DIR / LOCAL_TEMP_METADATA_JSONL
|
| 1514 |
+
local_temp_embeddings_path = LOCAL_WORK_DIR / LOCAL_TEMP_EMBEDDINGS_JSONL
|
| 1515 |
+
local_temp_log_path = LOCAL_WORK_DIR / LOCAL_TEMP_LOG_FILENAME
|
| 1516 |
+
|
| 1517 |
+
print(f"Input JSONL path: {input_jsonl_filepath}")
|
| 1518 |
+
print(f"Final Metadata Parquet path: {final_metadata_filepath}")
|
| 1519 |
+
print(f"Final Embeddings Parquet path: {final_embeddings_filepath}")
|
| 1520 |
+
print(f"Final log file path: {final_log_filepath}")
|
| 1521 |
+
print(f"Local temp Metadata path: {local_temp_metadata_path}")
|
| 1522 |
+
print(f"Local temp Embeddings path: {local_temp_embeddings_path}")
|
| 1523 |
+
print(f"Local temp log file path: {local_temp_log_path}")
|
| 1524 |
+
print("-" * 30)
|
| 1525 |
+
|
| 1526 |
+
|
| 1527 |
+
# Check for existing local temp files (for resumption)
|
| 1528 |
+
resuming_from_previous_run = False
|
| 1529 |
+
if local_temp_metadata_path.exists() and local_temp_embeddings_path.exists():
|
| 1530 |
+
file_size_meta = local_temp_metadata_path.stat().st_size
|
| 1531 |
+
file_size_embed = local_temp_embeddings_path.stat().st_size
|
| 1532 |
+
|
| 1533 |
+
if file_size_meta > 0 and file_size_embed > 0:
|
| 1534 |
+
print(f"Found existing temp files, will resume processing:")
|
| 1535 |
+
print(f" - Metadata file: {local_temp_metadata_path} ({file_size_meta} bytes)")
|
| 1536 |
+
print(f" - Embeddings file: {local_temp_embeddings_path} ({file_size_embed} bytes)")
|
| 1537 |
+
resuming_from_previous_run = True
|
| 1538 |
+
else:
|
| 1539 |
+
print(f"Found existing temp files but they're empty, removing them:")
|
| 1540 |
+
if file_size_meta == 0:
|
| 1541 |
+
print(f" - Removing empty metadata file: {local_temp_metadata_path}")
|
| 1542 |
+
local_temp_metadata_path.unlink()
|
| 1543 |
+
if file_size_embed == 0:
|
| 1544 |
+
print(f" - Removing empty embeddings file: {local_temp_embeddings_path}")
|
| 1545 |
+
local_temp_embeddings_path.unlink()
|
| 1546 |
+
else:
|
| 1547 |
+
print(f"No existing temp files found, starting fresh processing run.")
|
| 1548 |
+
|
| 1549 |
+
|
| 1550 |
+
# --- Run the Embedding Generation ---
|
| 1551 |
+
# Returns paths to the *local* temp parquet files if successful
|
| 1552 |
+
local_meta_path, local_embed_path = create_embedding_dataset(
|
| 1553 |
+
input_jsonl_filepath=input_jsonl_filepath,
|
| 1554 |
+
final_metadata_parquet_path=final_metadata_filepath, # For loading resume
|
| 1555 |
+
final_embeddings_parquet_path=final_embeddings_filepath, # For loading resume
|
| 1556 |
+
local_temp_metadata_path=local_temp_metadata_path, # Local save dest
|
| 1557 |
+
local_temp_embeddings_path=local_temp_embeddings_path, # Local save dest
|
| 1558 |
+
local_temp_log_path=local_temp_log_path, # Local log dest
|
| 1559 |
+
final_log_filepath=final_log_filepath, # Final log for logging clarity
|
| 1560 |
+
max_records=MAX_RECORDS_TO_PROCESS,
|
| 1561 |
+
batch_size=BATCH_SIZE,
|
| 1562 |
+
embedding_model_name=EMBEDDING_MODEL_NAME,
|
| 1563 |
+
process_config=PROCESS_CONFIG_JSON,
|
| 1564 |
+
process_readme=PROCESS_README_CONTENT,
|
| 1565 |
+
)
|
| 1566 |
+
|
| 1567 |
+
# --- Sync final local files to Drive/Destination ---
|
| 1568 |
+
if local_meta_path or local_embed_path: # Check if at least one file was created
|
| 1569 |
+
logging.info("Attempting to sync final local files to destination...")
|
| 1570 |
+
|
| 1571 |
+
# After all processing is complete
|
| 1572 |
+
meta_jsonl_path = LOCAL_WORK_DIR / LOCAL_TEMP_METADATA_JSONL
|
| 1573 |
+
embed_jsonl_path = LOCAL_WORK_DIR / LOCAL_TEMP_EMBEDDINGS_JSONL
|
| 1574 |
+
|
| 1575 |
+
# After all processing is complete
|
| 1576 |
+
# Define Parquet output paths
|
| 1577 |
+
local_temp_metadata_parquet = LOCAL_WORK_DIR / FINAL_METADATA_PARQUET_FILENAME
|
| 1578 |
+
local_temp_embeddings_parquet = LOCAL_WORK_DIR / FINAL_EMBEDDINGS_PARQUET_FILENAME
|
| 1579 |
+
|
| 1580 |
+
# One-time conversion from JSONL to Parquet at the very end
|
| 1581 |
+
convert_jsonl_to_parquet(
|
| 1582 |
+
meta_jsonl_path=meta_jsonl_path,
|
| 1583 |
+
embed_jsonl_path=embed_jsonl_path,
|
| 1584 |
+
local_temp_metadata_path=local_temp_metadata_parquet,
|
| 1585 |
+
local_temp_embeddings_path=local_temp_embeddings_parquet,
|
| 1586 |
+
chunk_size=50000, # Starting chunk size (will adapt)
|
| 1587 |
+
max_memory_mb=2000 # Memory threshold in MB
|
| 1588 |
+
)
|
| 1589 |
+
|
| 1590 |
+
sync_success = sync_local_files_to_final(
|
| 1591 |
+
local_metadata_path=local_temp_metadata_parquet, # Use the defined local path vars
|
| 1592 |
+
local_embeddings_path=local_temp_embeddings_path,
|
| 1593 |
+
local_log_path=local_temp_log_path,
|
| 1594 |
+
final_metadata_path=final_metadata_filepath,
|
| 1595 |
+
final_embeddings_path=final_embeddings_filepath,
|
| 1596 |
+
final_log_path=final_log_filepath
|
| 1597 |
+
)
|
| 1598 |
+
|
| 1599 |
+
if sync_success:
|
| 1600 |
+
logging.info("Final sync to destination successful.")
|
| 1601 |
+
# --- Upload final Parquet from Destination to Hub (Optional) ---
|
| 1602 |
+
if UPLOAD_TO_HUB:
|
| 1603 |
+
upload_files_to_hub(
|
| 1604 |
+
local_metadata_path=final_metadata_filepath, # Upload from final dest
|
| 1605 |
+
local_embeddings_path=final_embeddings_filepath,
|
| 1606 |
+
repo_id=TARGET_REPO_ID,
|
| 1607 |
+
repo_type=TARGET_REPO_TYPE,
|
| 1608 |
+
metadata_path_in_repo=METADATA_FILENAME_IN_REPO,
|
| 1609 |
+
embeddings_path_in_repo=EMBEDDINGS_FILENAME_IN_REPO,
|
| 1610 |
+
hf_token=None # Uses login
|
| 1611 |
+
)
|
| 1612 |
+
else: logging.info("Hub upload skipped by configuration.")
|
| 1613 |
+
else: logging.error("Final sync to destination FAILED. Cannot upload to Hub.")
|
| 1614 |
+
else: logging.warning("Local Parquet file creation failed or no data processed. Skipping final sync and Hub upload.")
|
| 1615 |
+
|
| 1616 |
+
'''
|
| 1617 |
+
# --- Clean up local temp files ---
|
| 1618 |
+
logging.info("Attempting final cleanup of local temp files...")
|
| 1619 |
+
try:
|
| 1620 |
+
if local_temp_metadata_path.is_file(): local_temp_metadata_path.unlink(); logging.info(f"Cleaned {local_temp_metadata_path}")
|
| 1621 |
+
if local_temp_embeddings_path.is_file(): local_temp_embeddings_path.unlink(); logging.info(f"Cleaned {local_temp_embeddings_path}")
|
| 1622 |
+
if local_temp_log_path.is_file(): local_temp_log_path.unlink(); logging.info(f"Cleaned {local_temp_log_path}")
|
| 1623 |
+
except Exception as clean_e: logging.warning(f"Could not clean up local temp files: {clean_e}")
|
| 1624 |
+
'''
|
| 1625 |
+
logging.info("Script finished.")
|
| 1626 |
+
|