From b40fcc86c7c9bf9c8b7135b95f8e65dca3fcbde1 Mon Sep 17 00:00:00 2001
From: ViperEkura <3081035982@qq.com>
Date: Sat, 4 Jul 2026 22:31:45 +0800
Subject: [PATCH] Refine abstract, switch to Times font, remove date, fix
 overfull hboxes with microtype

---
 main.tex | 24 +++++++++++++-----------
 1 file changed, 13 insertions(+), 11 deletions(-)

diff --git a/main.tex b/main.tex
index 94a1107..2c2ba65 100644
--- a/main.tex
+++ b/main.tex
@@ -3,8 +3,9 @@
 % ===== Packages =====
 \usepackage[utf8]{inputenc}
 \usepackage[T1]{fontenc}
+\usepackage{newtxtext,newtxmath}
 \usepackage[margin=1in]{geometry}
-\usepackage{amsmath,amssymb}
+\usepackage{amsmath}
 \usepackage{booktabs}
 \usepackage{graphicx}
 \usepackage{hyperref}
@@ -12,6 +13,7 @@
 \usepackage{caption}
 \usepackage{enumitem}
 \usepackage{url}
+\usepackage{microtype}
 
 \DeclareMathOperator{\Var}{Var}
 
@@ -19,7 +21,7 @@
        \large Data Pipeline, Distributed Training, and BF16 Numerical Stability via Residual Scaling}
 
 \author{AstrAI Contributors}
-\date{June 2026}
+\date{}
 
 \begin{document}
 
@@ -36,14 +38,14 @@ a 24-layer decoder-only architecture with Grouped Query Attention and
 SwiGLU, and distributed training via DDP/FSDP with cosine scheduling.
 A central focus is the numerical stability of BF16-precision training
 in deep Transformers. Through variance propagation analysis, we show
-that GPT-2 residual scaling ($\sigma_o = 0.02 / \sqrt{2L}$) on output
-projections reduces per-block residual variance by a factor of $2L=48$,
-containing post-24-layer variance at $1.34$ compared to $17.5$ without
-scaling. Empirical evaluations over 15B training tokens demonstrate that
-residual scaling consistently outperforms Kaiming initialization, with
-the gap widening to $0.79$ in the mid-training regime before narrowing
-to $0.38$ at convergence. These results establish residual scaling as a
-practical necessity for BF16 Transformer training at scale.
+that GPT-2 residual scaling on output projections reduces per-block
+residual variance by a factor of 48, containing post-24-layer variance
+at 1.34 compared to 17.5 without scaling. Empirical evaluations over
+15B training tokens demonstrate that residual scaling consistently
+outperforms Kaiming initialization, with the gap widening to 0.79 in
+the mid-training regime before narrowing to 0.38 at convergence. These
+results establish residual scaling as a practical necessity for BF16
+Transformer training at scale.
 \end{abstract}
 
 % ======================================================================
@@ -74,7 +76,7 @@ via a JSON specification that defines:
         (e.g.,~mask user input, compute loss on assistant response).
   \item \textbf{Packing}: Documents concatenated via \texttt{simple}
         (sequential), \texttt{bfd} (best-fit decreasing), or
-        \texttt{bfd\_split} strategies.
+        \texttt{bfd\_\allowbreak{}split} strategies.
   \item \textbf{Position IDs}: \texttt{none}, \texttt{doc\_reset} (per-document
         boundary), or \texttt{continuous}.
   \item \textbf{Output}: Tokenized sequences written to \texttt{.h5} or