From b303dba7df07b52d2687451f7765c2461b59f018 Mon Sep 17 00:00:00 2001
From: "coderabbitai[bot]"
 <136622811+coderabbitai[bot]@users.noreply.github.com>
Date: Mon, 6 Jul 2026 10:29:02 +0000
Subject: [PATCH] =?UTF-8?q?=F0=9F=93=9D=20CodeRabbit=20Chat:=20Update=20da?=
 =?UTF-8?q?taset=20blend=20preparation=20example?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 examples/dataset/prepare_data_blend.py | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/examples/dataset/prepare_data_blend.py b/examples/dataset/prepare_data_blend.py
index b5cc2607aa8..7ffebd5fe26 100644
--- a/examples/dataset/prepare_data_blend.py
+++ b/examples/dataset/prepare_data_blend.py
@@ -84,6 +84,8 @@ def _prepare_sources(
     workers = min(32, os.cpu_count() or 1)
     blend: list[tuple[float, str]] = []  # (weight, shared .bin/.idx path without extension)
     allocated_tokens = 0
+    # Weights are relative, not required to sum to 100 (matching data_blend.txt semantics).
+    weight_sum = sum(float(source["weight"]) for source in sources)
 
     for index, source in enumerate(sources):
         weight = float(source["weight"])
@@ -92,7 +94,7 @@ def _prepare_sources(
         elif index == len(sources) - 1:
             source_tokens = total_tokens - allocated_tokens
         else:
-            source_tokens = round(total_tokens * weight / 100)
+            source_tokens = round(total_tokens * weight / weight_sum)
             allocated_tokens += source_tokens
 
         dataset = source["hf_dataset"]