From 555f17c1eea94a2a433b142169e253251139b113 Mon Sep 17 00:00:00 2001
From: Zhangchi Feng <64362896+BUAADreamer@users.noreply.github.com>
Date: Wed, 15 Jan 2025 17:30:37 +0800
Subject: [PATCH] [data] Fix minicpmv/o dpo training (#6657)

* fix template name

* tiny fix

* support minicpm-o-2.6

* support inference of minicpmv

* update readme

* support dpo of minicpmv

Former-commit-id: 027942789bf3a28b2506a5730c05c8392ef5c885
---
 src/llamafactory/data/collator.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/src/llamafactory/data/collator.py b/src/llamafactory/data/collator.py
index c14f17e0..ba6a3da7 100644
--- a/src/llamafactory/data/collator.py
+++ b/src/llamafactory/data/collator.py
@@ -155,7 +155,7 @@ class MultiModalDataCollatorForSeq2Seq(DataCollatorForSeq2Seq):
         if "image_bound" in features:  # for minicpmv inputs
             bsz, seq_length = features["input_ids"].shape
             features["position_ids"] = torch.arange(seq_length).long().repeat(bsz, 1)
-            return {"data": features, "labels": features["labels"]}
+            return {"data": features, "input_ids": features["input_ids"], "labels": features["labels"]}
 
         return features