Note that the zstd cache must be decompressed before use

README.md

@@ -1,8 +1,20 @@
 # open-source
 
-Local copies of [open-thoughts/OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M), kept under the `open-source` folder.
+## 使用前必须解压
 
-- `open-source/hub/`: original parquet shards and the upstream dataset card. License: Apache-2.0.
-- `open-source/datasets/`: local Hugging Face `datasets` arrow cache of the same data.
+`open-source/datasets/cache-b0687316e784f78b.arrow.zst` 是压缩文件，不能直接当作数据使用。必须先解压，得到原来的 `.arrow` 文件后才能正常读取。
 
-This is a backup of files already on this machine. The upstream dataset remains open-thoughts/OpenThoughts3-1.2M.
+```bash
+zstd -d cache-b0687316e784f78b.arrow.zst -o cache-b0687316e784f78b.arrow
+```
+
+解压前大约 37GB，解压后大约 121GB。不解压的话，Hugging Face `datasets` 和普通 Arrow 读取都会失败。
+
+This file is zstd-compressed. Decompress it before use. The `.zst` file itself is not a usable dataset.
+
+## 内容
+
+这是本机上 [open-thoughts/OpenThoughts3-1.2M](https://huggingface.co/datasets/open-thoughts/OpenThoughts3-1.2M) 的本地副本，放在 `open-source` 文件夹里。上游数据集仍以 open-thoughts/OpenThoughts3-1.2M 为准，许可证是 Apache-2.0。
+
+- `open-source/hub/`：原始 parquet 分片和上游数据集说明。
+- `open-source/datasets/`：本地 Hugging Face `datasets` 的 arrow 缓存。其中 `cache-b0687316e784f78b.arrow.zst` 是上面那个必须解压的文件，其余 `.arrow` 分片可以直接使用。