README: script paths (pretraining/passk/)

README.md

@@ -17,7 +17,7 @@ as in the authors' code ([rosieyzh/openrlhf-pretrain](https://github.com/rosieyz
 (vocab 32000; the copy shipped with the authors' checkpoints, identical to `meta-llama/Llama-2-7b-hf`),
 one `</s>` (id 2) after every document, no BOS, documents shuffled with seed 0. Question-answer sources are
 `question + "\n\n" + solution` with no chat template. Produced by
-`PRETRAINING-PASSK/OLMo-mathonly/flame/data/tokenize_all.sh`.
+`PRETRAINING-PASSK/OLMo-mathonly/pretraining/passk/flame/data/tokenize_all.sh`.
 
 ## Layout
 
@@ -41,7 +41,7 @@ token count.
 
 The paper's mixtures (model-card naming: `as` Algebraic-Stack, `fm3` FineMath-3+, `tg` TinyGSM, `omi1`/`omi2`
 OpenMathInstruct-1/2, `mmqa` MetaMathQA, `<n>x` = n passes) are built from these pieces as directories of
-symlinks, n copies per pass (`flame/data/make_mix.py`, the authors' `mix_datasets.py`), and trained for
+symlinks, n copies per pass (`pretraining/passk/data/make_mix.py`, the authors' `mix_datasets.py`), and trained for
 one epoch at 512 x 2048 tokens per step.
 
 Note: the authors' OpenMathInstruct-2 script globs `*.parquet`, which on the hub also matches the
@@ -51,7 +51,7 @@ only, as the paper describes.
 ## Fetching
 
 ```bash
-bash PRETRAINING-PASSK/OLMo-mathonly/flame/data/download_from_hf.sh /tmp/$USER-mathonly-data   # parallel resumable curl
+bash PRETRAINING-PASSK/OLMo-mathonly/pretraining/passk/flame/data/download_from_hf.sh /tmp/$USER-mathonly-data   # parallel resumable curl
 ```
 
 Licenses follow the upstream datasets (ODC-By for FineMath, the Proof-Pile-2 subset licenses for