Which tool would be good to polish PacBio or ONT with Illumina?
Thank you in advance
PacBio and ONT are very different technologies; it would be helpful to be more specific. Regardless, I can only give my understanding as it relates to ONT reads....
TLDR: use medaka. Don't use Illumina reads.
Nanopore reads called using the most recent super-accuracy basecallers have a very high consensus accuracy (q40-q50) for SNPs and non-homopolymeric INDELs after nanopore-only correction. ONT provide their own tool, medaka, which uses assembly-mapped nanopore reads to correct the assembly. Medaka uses the base calling model from Guppy to predict the most likely sequence, given the input reads and the assembly.
If you're using Illumina reads to polish a nanopore assembly, it should only be used to correct INDEL errors (and ideally, only homopolymer errors). The two most commonly-used tools that can correct nanopore reads using Illumina reads are Racon and Pilon. Based on the recency of the commit histories, I'd guess that Racon would be the most likely to work well. It looks like Racon can do both assembly correction and read correction.
For PacBio HiFi reads (or other techs of comparable quality), it is not recommended to polish with Illumina at all. You run the risk of collapsing true heterozygosity/variation in your underlying genome and yielding a haploid assembly that does not represent either haplotype.
If that underlying variation is an issue because you need a haploid assembly, you are free to use a deduplication technique such as purge_haplotigs or similar (reviewed here) to obtain a primary set of contigs/scaffolds. Assigning p and q haplotypes is something that you can do with a tool like falcon-phase (if you have Hi-C) or a similar technique (I haven't been following the literature for a couple years, possibly there's something a lot better now!).
For a recent review of recommended strategies you can read the T2T consortium benchmarking paper on the topic.