summaryrefslogtreecommitdiff
path: root/README.md
diff options
context:
space:
mode:
Diffstat (limited to 'README.md')
-rw-r--r--README.md3
1 files changed, 2 insertions, 1 deletions
diff --git a/README.md b/README.md
index 9c038d5..c89bb5c 100644
--- a/README.md
+++ b/README.md
@@ -16,8 +16,9 @@ product are digital. See the main zobp repo for the method, theory (NSR ≈ c·d
- `configs/` — model sizes (60m/124m/350m/1b) x training arms (bp / zbp_n16 / zbp_n4)
## Run
+Data prep runs **on the training node** (H200), not on a dev machine; `data/` and `runs/` are gitignored.
```
-python scripts/prepare_data.py --dataset fineweb-edu --tokens 3e9 --out data/fineweb
+python scripts/prepare_data.py --tokens 3e9 --out data/fineweb # on the H200 node
torchrun --nproc_per_node=8 scripts/train.py --model configs/model/m124.yaml --train configs/train/zbp_n16.yaml
```
Global batch is fixed in the train config; per-rank micro-batch and accumulation adapt to world size.