ONT model problems

[Checunmily]

Hi, recently I tried to phasing ont hg002 data by margin, the basecaller version was Dorado0.4.3 with hac mode, and the vcf was generated by PEPPER-Margin-Deepvariant r0.8 with the model "--ont_r10_q20", when using margin, I used singularity to run the image:
singularity exec pepper_r0.8_with_all_para_for_margin.sif \
margin phase \
${dorado.0.4.3_trim}/align_to_hg38.all.bam \
${ref_hg38} \
${vcf_dir}/r043_trimmed_ont_r10_q20.vcf.gz \
/opt/margin_dir/params/phase/${para}.json \
-t 40 \
-o ${outdir}/R1041.dorado0.4.3.margin.${para}
The parameters I tested here were "allParams.haplotag.ont-r94g422.json", "allParams.haplotag.ont-r94g507.json" and "allParams.haplotag.ont-r104q20.json". The "allParams.haplotag.ont-r104q20.json" was not contained in pepper_deepvariant r0.8 so that I downloaded and added it into the image. Finally, I used whatshap v2.1 compare command line to check the switch error rate . I used GIAB v4.2.1 as the truth dataset and found that the result with the parameter "allParams.haplotag.ont-r94g422.json" seemed to be the best. Later I also tried Guppy6.3.8 with hac mode on ONT HG002 data and just got the same result. Did I miss something?