From 2200120133404869ef8af07acd361c41a8d4bb42 Mon Sep 17 00:00:00 2001
From: marauder-actual <marauder@saiden.dev>
Date: Mon, 1 Jun 2026 04:27:24 +0200
Subject: [PATCH] fix: lower seqlen to 512 for short calibration examples

Training examples are ~500-1500 tokens. seqlen=2048 causes
'no data has been cached' error. Also remove deprecated format param.
---
 quantize_awq.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/quantize_awq.py b/quantize_awq.py
index c065160..c193afc 100644
--- a/quantize_awq.py
+++ b/quantize_awq.py
@@ -46,10 +46,9 @@ rounder = AutoRound(
     dataset=calib,
     bits=4,
     group_size=128,
-    seqlen=2048,
+    seqlen=512,
     nsamples=min(128, len(calib)),
     iters=200,
-    format="auto_round",
 )
 
 print("Running AutoRound INT4 quantization...")