Skip to content

feat: hebrew trainer env params (WO20 substrate) #112

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
ronaldtse merged 1 commit into from
Oct 7, 2026
Merged

feat: hebrew trainer env params (WO20 substrate) #112

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 11 additions & 7 deletions train_hebrew_plane_hf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,7 @@
-v hf://buckets/Interscript/isx-training:/ckpt:rw \
-v /tmp/isx-hf-code:/code:ro \
train_hebrew_plane_hf.py
# arms via env: HEB_RUN / HEB_BACKBONE / HEB_EPOCHS / HEB_K / HEB_BS / HEB_ACCUM
"""

from __future__ import annotations
Expand All @@ -36,16 +37,19 @@
sys.path.insert(0, "/code")
sys.path.insert(0, "/code/src")

import os

DATA = Path("/train_data")
CKPT = Path("/ckpt")
RUN = "run-022-heb-plane-base"
RUN = os.environ.get("HEB_RUN", "run-022-heb-plane-base")
BACKBONE = os.environ.get("HEB_BACKBONE", "google/byt5-base")
EPOCHS = int(os.environ.get("HEB_EPOCHS", "4"))
K_PASSES = int(os.environ.get("HEB_K", "3"))
MICRO_BS = int(os.environ.get("HEB_BS", "8"))
GRAD_ACCUM = int(os.environ.get("HEB_ACCUM", "2"))
UNIT_BYTES = 1400
N_VAL = 2_000
EPOCHS = 4
MAX_LINES = 600_000
K_PASSES = 3
MICRO_BS = 8
GRAD_ACCUM = 2

import torch
import torch.nn as nn
Expand Down Expand Up @@ -113,7 +117,7 @@ def main() -> None:

labels_digest = hashlib.sha256(inv_path.read_bytes()).hexdigest()[:12]

tok = AutoTokenizer.from_pretrained("google/byt5-base")
tok = AutoTokenizer.from_pretrained(BACKBONE)

def encode_unit(text: str):
skel, labels = NP.split_planes(text)
Expand Down Expand Up @@ -159,7 +163,7 @@ def collate(batch):
print(f"[data] train={len(train_units)} val={len(val_rows)} steps={total_steps} "
f"classes={n_classes}", flush=True)

backbone = AutoModelForSeq2SeqLM.from_pretrained("google/byt5-base")
backbone = AutoModelForSeq2SeqLM.from_pretrained(BACKBONE)
encoder = backbone.encoder.cuda().train()
d_model = encoder.config.d_model

Expand Down
Loading