Method	Captioning		VQA				Comprehensive
	COCO	12Paragraph	VQAv2	OKVQA	VizWiz	TextVQA	MM-Vet

Qwen-InternViT-SD3.5	106.4	10.7	73.9	54.2	49.1	54.8	44.0
Qwen-InternViT-SD3.5*	102.1	10.9	73.0	53.6	48.6	55.2	45.7
Qwen-InternViT-SD3.5-DPO	64.6	11.6	74.2	50.9	48.9	55.8	44.7

Qwen-CLIP-SD3.5	99.9	9.7	72.9	52.3	49.0	44.0	39.8
Qwen-CLIP-SD3.5*	99.1	10.2	72.7	51.1	49.1	43.9	42.1

Qwen-CLIP-SD2.1	82.8	9.1	72.5	52.4	49.4	43.6	42
Qwen-CLIP-SD2.1*	97.3	10.8	72.4	50.4	49.9	43.2	39.0

Vicuna-CLIP-SD2.1	115.4	17.4	56.6	44.3	45.8	34.9	35.9
Vicuna-CLIP-SD2.1*	103.7	8.4	72.9	52.2	49.3	41.8	36.6

Image Generation Evaluation

Method	MS-COCO

Qwen-InternViT-SD3.5-Stage1	11.72
Qwen-InternViT-SD3.5	11.11
Qwen-InternViT-SD3.5-DPO	11.33

Qwen-CLIP-SD3.5-Stage1	11.72
Qwen-CLIP-SD3.5	11.61

Qwen-CLIP-SD2.1-Stage1	13.94
Qwen-CLIP-SD2.1	12.26

Vicuna-CLIP-SD2.1-Stage1	8.76(+~2)
Vicuna-CLIP-SD2.1	8.46(+~2)

In the original text of DreamLLm, Vicuna-CLIP-SD2.1 and Vicuna-CLIP-SD2.1 were run 8 times, and the best one among 8 images was selected for each figure. All my models were only tested once, with an approximate error of 2~3.

FolSpark/DreamLLM-Qwen2.5-CLIP-SD2.1

Model performance

Multimodal Comprehension Assessment

Image Generation Evaluation

FolSpark/
DreamLLM-Qwen2.5-CLIP-SD2.1