[WIP] add florence2

[Sanster]

• Did you read the contributor guideline?
• Did you write any new necessary tests? Added two test in test_mm_plugin.py
• Fix the issues in the saved configuration file.

Running example data training:

llamafactory-cli train examples/train_full/florence2_full_sft.yaml

TODO: After training finished, modify saves/florence2-large/full/sft/config.json file(I still haven't figured out why the saved config has the following issues that need to be modified; it shouldn't be necessary.):

Change auto_map, from:

  "auto_map": {
    "AutoConfig": "configuration_florence2.Florence2Config",
    "AutoModel": "modeling_florence2.Florence2ForConditionalGeneration",
    "AutoModelForCausalLM": "microsoft/Florence-2-large-ft--modeling_florence2.Florence2ForConditionalGeneration"
  }

to

  "auto_map": {
    "AutoConfig": "configuration_florence2.Florence2Config",
    "AutoModelForCausalLM": "modeling_florence2.Florence2ForConditionalGeneration"
  }

Change vision_config's model_type

{
  "vision_config": {
    "model_type": "davit"
  }
}

Document of training table and figure detection on the yifeihu/TF-ID-arxiv-papers: https://github.com/Sanster/llama_factory_florence2_example