Duplicate from HuggingFaceM4/Florence-2-DocVQA

Co-authored-by: Andres Marafioti <[email protected]>

.gitattributes

@@ -0,0 +1,35 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+pipeline_tag: image-text-to-text
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+This is Microsoft's Florence-2 model trained for 1 day with Docmatix (5% of the data) with a learning rate of 1e-6.
+The code for this fine-tuning can be found here: https://github.com/andimarafioti/florence2-finetuning
+And here's a blog explaining how to fine tune Florence: https://huggingface.co/blog/finetune-florence2
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. It has been automatically generated.
+
+- **Developed by:** Andi Marafioti
+- **Funded by [optional]:** Hugging Face 🤗
+- **Language(s) (NLP):** English
+- **License:** MIT
+- **Finetuned from model:** [Florence-2-large-ft](https://huggingface.co/microsoft/Florence-2-large-ft)
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

added_tokens.json

@@ -0,0 +1,1026 @@
+{
+  "</cap>": 51270,
+  "</dcap>": 51274,
+  "</grounding>": 51276,
+  "</ncap>": 51272,
+  "</ocr>": 50268,
+  "</od>": 50266,
+  "</poly>": 51287,
+  "</proposal>": 51285,
+  "</region_cap>": 51281,
+  "</region_to_desciption>": 51283,
+  "</seg>": 51278,
+  "<and>": 51288,
+  "<cap>": 51269,
+  "<dcap>": 51273,
+  "<grounding>": 51275,
+  "<loc_0>": 50269,
+  "<loc_100>": 50369,
+  "<loc_101>": 50370,
+  "<loc_102>": 50371,
+  "<loc_103>": 50372,
+  "<loc_104>": 50373,
+  "<loc_105>": 50374,
+  "<loc_106>": 50375,
+  "<loc_107>": 50376,
+  "<loc_108>": 50377,
+  "<loc_109>": 50378,
+  "<loc_10>": 50279,
+  "<loc_110>": 50379,
+  "<loc_111>": 50380,
+  "<loc_112>": 50381,
+  "<loc_113>": 50382,
+  "<loc_114>": 50383,
+  "<loc_115>": 50384,
+  "<loc_116>": 50385,
+  "<loc_117>": 50386,
+  "<loc_118>": 50387,
+  "<loc_119>": 50388,
+  "<loc_11>": 50280,
+  "<loc_120>": 50389,
+  "<loc_121>": 50390,
+  "<loc_122>": 50391,
+  "<loc_123>": 50392,
+  "<loc_124>": 50393,
+  "<loc_125>": 50394,
+  "<loc_126>": 50395,
+  "<loc_127>": 50396,
+  "<loc_128>": 50397,
+  "<loc_129>": 50398,
+  "<loc_12>": 50281,
+  "<loc_130>": 50399,
+  "<loc_131>": 50400,
+  "<loc_132>": 50401,
+  "<loc_133>": 50402,
+  "<loc_134>": 50403,
+  "<loc_135>": 50404,
+  "<loc_136>": 50405,
+  "<loc_137>": 50406,
+  "<loc_138>": 50407,
+  "<loc_139>": 50408,
+  "<loc_13>": 50282,
+  "<loc_140>": 50409,
+  "<loc_141>": 50410,
+  "<loc_142>": 50411,
+  "<loc_143>": 50412,
+  "<loc_144>": 50413,
+  "<loc_145>": 50414,
+  "<loc_146>": 50415,
+  "<loc_147>": 50416,
+  "<loc_148>": 50417,
+  "<loc_149>": 50418,
+  "<loc_14>": 50283,
+  "<loc_150>": 50419,
+  "<loc_151>": 50420,
+  "<loc_152>": 50421,
+  "<loc_153>": 50422,
+  "<loc_154>": 50423,
+  "<loc_155>": 50424,
+  "<loc_156>": 50425,
+  "<loc_157>": 50426,
+  "<loc_158>": 50427,
+  "<loc_159>": 50428,
+  "<loc_15>": 50284,
+  "<loc_160>": 50429,
+  "<loc_161>": 50430,
+  "<loc_162>": 50431,
+  "<loc_163>": 50432,
+  "<loc_164>": 50433,
+  "<loc_165>": 50434,
+  "<loc_166>": 50435,
+  "<loc_167>": 50436,
+  "<loc_168>": 50437,
+  "<loc_169>": 50438,
+  "<loc_16>": 50285,
+  "<loc_170>": 50439,
+  "<loc_171>": 50440,
+  "<loc_172>": 50441,
+  "<loc_173>": 50442,
+  "<loc_174>": 50443,
+  "<loc_175>": 50444,
+  "<loc_176>": 50445,
+  "<loc_177>": 50446,
+  "<loc_178>": 50447,
+  "<loc_179>": 50448,
+  "<loc_17>": 50286,
+  "<loc_180>": 50449,
+  "<loc_181>": 50450,
+  "<loc_182>": 50451,
+  "<loc_183>": 50452,
+  "<loc_184>": 50453,
+  "<loc_185>": 50454,
+  "<loc_186>": 50455,
+  "<loc_187>": 50456,
+  "<loc_188>": 50457,
+  "<loc_189>": 50458,
+  "<loc_18>": 50287,
+  "<loc_190>": 50459,
+  "<loc_191>": 50460,
+  "<loc_192>": 50461,
+  "<loc_193>": 50462,
+  "<loc_194>": 50463,
+  "<loc_195>": 50464,
+  "<loc_196>": 50465,
+  "<loc_197>": 50466,
+  "<loc_198>": 50467,
+  "<loc_199>": 50468,
+  "<loc_19>": 50288,
+  "<loc_1>": 50270,
+  "<loc_200>": 50469,
+  "<loc_201>": 50470,
+  "<loc_202>": 50471,
+  "<loc_203>": 50472,
+  "<loc_204>": 50473,
+  "<loc_205>": 50474,
+  "<loc_206>": 50475,
+  "<loc_207>": 50476,
+  "<loc_208>": 50477,
+  "<loc_209>": 50478,
+  "<loc_20>": 50289,
+  "<loc_210>": 50479,
+  "<loc_211>": 50480,
+  "<loc_212>": 50481,
+  "<loc_213>": 50482,
+  "<loc_214>": 50483,
+  "<loc_215>": 50484,
+  "<loc_216>": 50485,
+  "<loc_217>": 50486,
+  "<loc_218>": 50487,
+  "<loc_219>": 50488,
+  "<loc_21>": 50290,
+  "<loc_220>": 50489,
+  "<loc_221>": 50490,
+  "<loc_222>": 50491,
+  "<loc_223>": 50492,
+  "<loc_224>": 50493,
+  "<loc_225>": 50494,
+  "<loc_226>": 50495,
+  "<loc_227>": 50496,
+  "<loc_228>": 50497,
+  "<loc_229>": 50498,
+  "<loc_22>": 50291,
+  "<loc_230>": 50499,
+  "<loc_231>": 50500,
+  "<loc_232>": 50501,
+  "<loc_233>": 50502,
+  "<loc_234>": 50503,
+  "<loc_235>": 50504,
+  "<loc_236>": 50505,
+  "<loc_237>": 50506,
+  "<loc_238>": 50507,
+  "<loc_239>": 50508,
+  "<loc_23>": 50292,
+  "<loc_240>": 50509,
+  "<loc_241>": 50510,
+  "<loc_242>": 50511,
+  "<loc_243>": 50512,
+  "<loc_244>": 50513,
+  "<loc_245>": 50514,
+  "<loc_246>": 50515,
+  "<loc_247>": 50516,
+  "<loc_248>": 50517,
+  "<loc_249>": 50518,
+  "<loc_24>": 50293,
+  "<loc_250>": 50519,
+  "<loc_251>": 50520,
+  "<loc_252>": 50521,
+  "<loc_253>": 50522,
+  "<loc_254>": 50523,
+  "<loc_255>": 50524,
+  "<loc_256>": 50525,
+  "<loc_257>": 50526,
+  "<loc_258>": 50527,
+  "<loc_259>": 50528,
+  "<loc_25>": 50294,
+  "<loc_260>": 50529,
+  "<loc_261>": 50530,
+  "<loc_262>": 50531,
+  "<loc_263>": 50532,
+  "<loc_264>": 50533,
+  "<loc_265>": 50534,
+  "<loc_266>": 50535,
+  "<loc_267>": 50536,
+  "<loc_268>": 50537,
+  "<loc_269>": 50538,
+  "<loc_26>": 50295,
+  "<loc_270>": 50539,
+  "<loc_271>": 50540,
+  "<loc_272>": 50541,
+  "<loc_273>": 50542,
+  "<loc_274>": 50543,
+  "<loc_275>": 50544,
+  "<loc_276>": 50545,
+  "<loc_277>": 50546,
+  "<loc_278>": 50547,
+  "<loc_279>": 50548,
+  "<loc_27>": 50296,
+  "<loc_280>": 50549,
+  "<loc_281>": 50550,
+  "<loc_282>": 50551,
+  "<loc_283>": 50552,
+  "<loc_284>": 50553,
+  "<loc_285>": 50554,
+  "<loc_286>": 50555,
+  "<loc_287>": 50556,
+  "<loc_288>": 50557,
+  "<loc_289>": 50558,
+  "<loc_28>": 50297,
+  "<loc_290>": 50559,
+  "<loc_291>": 50560,
+  "<loc_292>": 50561,
+  "<loc_293>": 50562,
+  "<loc_294>": 50563,
+  "<loc_295>": 50564,
+  "<loc_296>": 50565,
+  "<loc_297>": 50566,
+  "<loc_298>": 50567,
+  "<loc_299>": 50568,
+  "<loc_29>": 50298,
+  "<loc_2>": 50271,
+  "<loc_300>": 50569,
+  "<loc_301>": 50570,
+  "<loc_302>": 50571,
+  "<loc_303>": 50572,
+  "<loc_304>": 50573,
+  "<loc_305>": 50574,
+  "<loc_306>": 50575,
+  "<loc_307>": 50576,
+  "<loc_308>": 50577,
+  "<loc_309>": 50578,
+  "<loc_30>": 50299,
+  "<loc_310>": 50579,
+  "<loc_311>": 50580,
+  "<loc_312>": 50581,
+  "<loc_313>": 50582,
+  "<loc_314>": 50583,
+  "<loc_315>": 50584,
+  "<loc_316>": 50585,
+  "<loc_317>": 50586,
+  "<loc_318>": 50587,
+  "<loc_319>": 50588,
+  "<loc_31>": 50300,
+  "<loc_320>": 50589,
+  "<loc_321>": 50590,
+  "<loc_322>": 50591,
+  "<loc_323>": 50592,
+  "<loc_324>": 50593,
+  "<loc_325>": 50594,
+  "<loc_326>": 50595,
+  "<loc_327>": 50596,
+  "<loc_328>": 50597,
+  "<loc_329>": 50598,
+  "<loc_32>": 50301,
+  "<loc_330>": 50599,
+  "<loc_331>": 50600,
+  "<loc_332>": 50601,
+  "<loc_333>": 50602,
+  "<loc_334>": 50603,
+  "<loc_335>": 50604,
+  "<loc_336>": 50605,
+  "<loc_337>": 50606,
+  "<loc_338>": 50607,
+  "<loc_339>": 50608,
+  "<loc_33>": 50302,
+  "<loc_340>": 50609,
+  "<loc_341>": 50610,
+  "<loc_342>": 50611,
+  "<loc_343>": 50612,
+  "<loc_344>": 50613,
+  "<loc_345>": 50614,
+  "<loc_346>": 50615,
+  "<loc_347>": 50616,
+  "<loc_348>": 50617,
+  "<loc_349>": 50618,
+  "<loc_34>": 50303,
+  "<loc_350>": 50619,
+  "<loc_351>": 50620,
+  "<loc_352>": 50621,
+  "<loc_353>": 50622,
+  "<loc_354>": 50623,
+  "<loc_355>": 50624,
+  "<loc_356>": 50625,
+  "<loc_357>": 50626,
+  "<loc_358>": 50627,
+  "<loc_359>": 50628,
+  "<loc_35>": 50304,
+  "<loc_360>": 50629,
+  "<loc_361>": 50630,
+  "<loc_362>": 50631,
+  "<loc_363>": 50632,
+  "<loc_364>": 50633,
+  "<loc_365>": 50634,
+  "<loc_366>": 50635,
+  "<loc_367>": 50636,
+  "<loc_368>": 50637,
+  "<loc_369>": 50638,
+  "<loc_36>": 50305,
+  "<loc_370>": 50639,
+  "<loc_371>": 50640,
+  "<loc_372>": 50641,
+  "<loc_373>": 50642,
+  "<loc_374>": 50643,
+  "<loc_375>": 50644,
+  "<loc_376>": 50645,
+  "<loc_377>": 50646,
+  "<loc_378>": 50647,
+  "<loc_379>": 50648,
+  "<loc_37>": 50306,
+  "<loc_380>": 50649,
+  "<loc_381>": 50650,
+  "<loc_382>": 50651,
+  "<loc_383>": 50652,
+  "<loc_384>": 50653,
+  "<loc_385>": 50654,
+  "<loc_386>": 50655,
+  "<loc_387>": 50656,
+  "<loc_388>": 50657,
+  "<loc_389>": 50658,
+  "<loc_38>": 50307,
+  "<loc_390>": 50659,
+  "<loc_391>": 50660,
+  "<loc_392>": 50661,
+  "<loc_393>": 50662,
+  "<loc_394>": 50663,
+  "<loc_395>": 50664,
+  "<loc_396>": 50665,
+  "<loc_397>": 50666,
+  "<loc_398>": 50667,
+  "<loc_399>": 50668,
+  "<loc_39>": 50308,
+  "<loc_3>": 50272,
+  "<loc_400>": 50669,
+  "<loc_401>": 50670,
+  "<loc_402>": 50671,
+  "<loc_403>": 50672,
+  "<loc_404>": 50673,
+  "<loc_405>": 50674,
+  "<loc_406>": 50675,
+  "<loc_407>": 50676,
+  "<loc_408>": 50677,
+  "<loc_409>": 50678,
+  "<loc_40>": 50309,
+  "<loc_410>": 50679,
+  "<loc_411>": 50680,
+  "<loc_412>": 50681,
+  "<loc_413>": 50682,
+  "<loc_414>": 50683,
+  "<loc_415>": 50684,
+  "<loc_416>": 50685,
+  "<loc_417>": 50686,
+  "<loc_418>": 50687,
+  "<loc_419>": 50688,
+  "<loc_41>": 50310,
+  "<loc_420>": 50689,
+  "<loc_421>": 50690,
+  "<loc_422>": 50691,
+  "<loc_423>": 50692,
+  "<loc_424>": 50693,
+  "<loc_425>": 50694,
+  "<loc_426>": 50695,
+  "<loc_427>": 50696,
+  "<loc_428>": 50697,
+  "<loc_429>": 50698,
+  "<loc_42>": 50311,
+  "<loc_430>": 50699,
+  "<loc_431>": 50700,
+  "<loc_432>": 50701,
+  "<loc_433>": 50702,
+  "<loc_434>": 50703,
+  "<loc_435>": 50704,
+  "<loc_436>": 50705,
+  "<loc_437>": 50706,
+  "<loc_438>": 50707,
+  "<loc_439>": 50708,
+  "<loc_43>": 50312,
+  "<loc_440>": 50709,
+  "<loc_441>": 50710,
+  "<loc_442>": 50711,
+  "<loc_443>": 50712,
+  "<loc_444>": 50713,
+  "<loc_445>": 50714,
+  "<loc_446>": 50715,
+  "<loc_447>": 50716,
+  "<loc_448>": 50717,
+  "<loc_449>": 50718,
+  "<loc_44>": 50313,
+  "<loc_450>": 50719,
+  "<loc_451>": 50720,
+  "<loc_452>": 50721,
+  "<loc_453>": 50722,
+  "<loc_454>": 50723,
+  "<loc_455>": 50724,
+  "<loc_456>": 50725,
+  "<loc_457>": 50726,
+  "<loc_458>": 50727,
+  "<loc_459>": 50728,
+  "<loc_45>": 50314,
+  "<loc_460>": 50729,
+  "<loc_461>": 50730,
+  "<loc_462>": 50731,
+  "<loc_463>": 50732,
+  "<loc_464>": 50733,
+  "<loc_465>": 50734,
+  "<loc_466>": 50735,
+  "<loc_467>": 50736,
+  "<loc_468>": 50737,
+  "<loc_469>": 50738,
+  "<loc_46>": 50315,
+  "<loc_470>": 50739,
+  "<loc_471>": 50740,
+  "<loc_472>": 50741,
+  "<loc_473>": 50742,
+  "<loc_474>": 50743,
+  "<loc_475>": 50744,
+  "<loc_476>": 50745,
+  "<loc_477>": 50746,
+  "<loc_478>": 50747,
+  "<loc_479>": 50748,
+  "<loc_47>": 50316,
+  "<loc_480>": 50749,
+  "<loc_481>": 50750,
+  "<loc_482>": 50751,
+  "<loc_483>": 50752,
+  "<loc_484>": 50753,
+  "<loc_485>": 50754,
+  "<loc_486>": 50755,
+  "<loc_487>": 50756,
+  "<loc_488>": 50757,
+  "<loc_489>": 50758,
+  "<loc_48>": 50317,
+  "<loc_490>": 50759,
+  "<loc_491>": 50760,
+  "<loc_492>": 50761,
+  "<loc_493>": 50762,
+  "<loc_494>": 50763,
+  "<loc_495>": 50764,
+  "<loc_496>": 50765,
+  "<loc_497>": 50766,
+  "<loc_498>": 50767,
+  "<loc_499>": 50768,
+  "<loc_49>": 50318,
+  "<loc_4>": 50273,
+  "<loc_500>": 50769,
+  "<loc_501>": 50770,
+  "<loc_502>": 50771,
+  "<loc_503>": 50772,
+  "<loc_504>": 50773,
+  "<loc_505>": 50774,
+  "<loc_506>": 50775,
+  "<loc_507>": 50776,
+  "<loc_508>": 50777,
+  "<loc_509>": 50778,
+  "<loc_50>": 50319,
+  "<loc_510>": 50779,
+  "<loc_511>": 50780,
+  "<loc_512>": 50781,
+  "<loc_513>": 50782,
+  "<loc_514>": 50783,
+  "<loc_515>": 50784,
+  "<loc_516>": 50785,
+  "<loc_517>": 50786,
+  "<loc_518>": 50787,
+  "<loc_519>": 50788,
+  "<loc_51>": 50320,
+  "<loc_520>": 50789,
+  "<loc_521>": 50790,
+  "<loc_522>": 50791,
+  "<loc_523>": 50792,
+  "<loc_524>": 50793,
+  "<loc_525>": 50794,
+  "<loc_526>": 50795,
+  "<loc_527>": 50796,
+  "<loc_528>": 50797,
+  "<loc_529>": 50798,
+  "<loc_52>": 50321,
+  "<loc_530>": 50799,
+  "<loc_531>": 50800,
+  "<loc_532>": 50801,
+  "<loc_533>": 50802,
+  "<loc_534>": 50803,
+  "<loc_535>": 50804,
+  "<loc_536>": 50805,
+  "<loc_537>": 50806,
+  "<loc_538>": 50807,
+  "<loc_539>": 50808,
+  "<loc_53>": 50322,
+  "<loc_540>": 50809,
+  "<loc_541>": 50810,
+  "<loc_542>": 50811,
+  "<loc_543>": 50812,
+  "<loc_544>": 50813,
+  "<loc_545>": 50814,
+  "<loc_546>": 50815,
+  "<loc_547>": 50816,
+  "<loc_548>": 50817,
+  "<loc_549>": 50818,
+  "<loc_54>": 50323,
+  "<loc_550>": 50819,
+  "<loc_551>": 50820,
+  "<loc_552>": 50821,
+  "<loc_553>": 50822,
+  "<loc_554>": 50823,
+  "<loc_555>": 50824,
+  "<loc_556>": 50825,
+  "<loc_557>": 50826,
+  "<loc_558>": 50827,
+  "<loc_559>": 50828,
+  "<loc_55>": 50324,
+  "<loc_560>": 50829,
+  "<loc_561>": 50830,
+  "<loc_562>": 50831,
+  "<loc_563>": 50832,
+  "<loc_564>": 50833,
+  "<loc_565>": 50834,
+  "<loc_566>": 50835,
+  "<loc_567>": 50836,
+  "<loc_568>": 50837,
+  "<loc_569>": 50838,
+  "<loc_56>": 50325,
+  "<loc_570>": 50839,
+  "<loc_571>": 50840,
+  "<loc_572>": 50841,
+  "<loc_573>": 50842,
+  "<loc_574>": 50843,
+  "<loc_575>": 50844,
+  "<loc_576>": 50845,
+  "<loc_577>": 50846,
+  "<loc_578>": 50847,
+  "<loc_579>": 50848,
+  "<loc_57>": 50326,
+  "<loc_580>": 50849,
+  "<loc_581>": 50850,
+  "<loc_582>": 50851,
+  "<loc_583>": 50852,
+  "<loc_584>": 50853,
+  "<loc_585>": 50854,
+  "<loc_586>": 50855,
+  "<loc_587>": 50856,
+  "<loc_588>": 50857,
+  "<loc_589>": 50858,
+  "<loc_58>": 50327,
+  "<loc_590>": 50859,
+  "<loc_591>": 50860,
+  "<loc_592>": 50861,
+  "<loc_593>": 50862,
+  "<loc_594>": 50863,
+  "<loc_595>": 50864,
+  "<loc_596>": 50865,
+  "<loc_597>": 50866,
+  "<loc_598>": 50867,
+  "<loc_599>": 50868,
+  "<loc_59>": 50328,
+  "<loc_5>": 50274,
+  "<loc_600>": 50869,
+  "<loc_601>": 50870,
+  "<loc_602>": 50871,
+  "<loc_603>": 50872,
+  "<loc_604>": 50873,
+  "<loc_605>": 50874,
+  "<loc_606>": 50875,
+  "<loc_607>": 50876,
+  "<loc_608>": 50877,
+  "<loc_609>": 50878,
+  "<loc_60>": 50329,
+  "<loc_610>": 50879,
+  "<loc_611>": 50880,
+  "<loc_612>": 50881,
+  "<loc_613>": 50882,
+  "<loc_614>": 50883,
+  "<loc_615>": 50884,
+  "<loc_616>": 50885,
+  "<loc_617>": 50886,
+  "<loc_618>": 50887,
+  "<loc_619>": 50888,
+  "<loc_61>": 50330,
+  "<loc_620>": 50889,
+  "<loc_621>": 50890,
+  "<loc_622>": 50891,
+  "<loc_623>": 50892,
+  "<loc_624>": 50893,
+  "<loc_625>": 50894,
+  "<loc_626>": 50895,
+  "<loc_627>": 50896,
+  "<loc_628>": 50897,
+  "<loc_629>": 50898,
+  "<loc_62>": 50331,
+  "<loc_630>": 50899,
+  "<loc_631>": 50900,
+  "<loc_632>": 50901,
+  "<loc_633>": 50902,
+  "<loc_634>": 50903,
+  "<loc_635>": 50904,
+  "<loc_636>": 50905,
+  "<loc_637>": 50906,
+  "<loc_638>": 50907,
+  "<loc_639>": 50908,
+  "<loc_63>": 50332,
+  "<loc_640>": 50909,
+  "<loc_641>": 50910,
+  "<loc_642>": 50911,
+  "<loc_643>": 50912,
+  "<loc_644>": 50913,
+  "<loc_645>": 50914,
+  "<loc_646>": 50915,
+  "<loc_647>": 50916,
+  "<loc_648>": 50917,
+  "<loc_649>": 50918,
+  "<loc_64>": 50333,
+  "<loc_650>": 50919,
+  "<loc_651>": 50920,
+  "<loc_652>": 50921,
+  "<loc_653>": 50922,
+  "<loc_654>": 50923,
+  "<loc_655>": 50924,
+  "<loc_656>": 50925,
+  "<loc_657>": 50926,
+  "<loc_658>": 50927,
+  "<loc_659>": 50928,
+  "<loc_65>": 50334,
+  "<loc_660>": 50929,
+  "<loc_661>": 50930,
+  "<loc_662>": 50931,
+  "<loc_663>": 50932,
+  "<loc_664>": 50933,
+  "<loc_665>": 50934,
+  "<loc_666>": 50935,
+  "<loc_667>": 50936,
+  "<loc_668>": 50937,
+  "<loc_669>": 50938,
+  "<loc_66>": 50335,
+  "<loc_670>": 50939,
+  "<loc_671>": 50940,
+  "<loc_672>": 50941,
+  "<loc_673>": 50942,
+  "<loc_674>": 50943,
+  "<loc_675>": 50944,
+  "<loc_676>": 50945,
+  "<loc_677>": 50946,
+  "<loc_678>": 50947,
+  "<loc_679>": 50948,
+  "<loc_67>": 50336,
+  "<loc_680>": 50949,
+  "<loc_681>": 50950,
+  "<loc_682>": 50951,
+  "<loc_683>": 50952,
+  "<loc_684>": 50953,
+  "<loc_685>": 50954,
+  "<loc_686>": 50955,
+  "<loc_687>": 50956,
+  "<loc_688>": 50957,
+  "<loc_689>": 50958,
+  "<loc_68>": 50337,
+  "<loc_690>": 50959,
+  "<loc_691>": 50960,
+  "<loc_692>": 50961,
+  "<loc_693>": 50962,
+  "<loc_694>": 50963,
+  "<loc_695>": 50964,
+  "<loc_696>": 50965,
+  "<loc_697>": 50966,
+  "<loc_698>": 50967,
+  "<loc_699>": 50968,
+  "<loc_69>": 50338,
+  "<loc_6>": 50275,
+  "<loc_700>": 50969,
+  "<loc_701>": 50970,
+  "<loc_702>": 50971,
+  "<loc_703>": 50972,
+  "<loc_704>": 50973,
+  "<loc_705>": 50974,
+  "<loc_706>": 50975,
+  "<loc_707>": 50976,
+  "<loc_708>": 50977,
+  "<loc_709>": 50978,
+  "<loc_70>": 50339,
+  "<loc_710>": 50979,
+  "<loc_711>": 50980,
+  "<loc_712>": 50981,
+  "<loc_713>": 50982,
+  "<loc_714>": 50983,
+  "<loc_715>": 50984,
+  "<loc_716>": 50985,
+  "<loc_717>": 50986,
+  "<loc_718>": 50987,
+  "<loc_719>": 50988,
+  "<loc_71>": 50340,
+  "<loc_720>": 50989,
+  "<loc_721>": 50990,
+  "<loc_722>": 50991,
+  "<loc_723>": 50992,
+  "<loc_724>": 50993,
+  "<loc_725>": 50994,
+  "<loc_726>": 50995,
+  "<loc_727>": 50996,
+  "<loc_728>": 50997,
+  "<loc_729>": 50998,
+  "<loc_72>": 50341,
+  "<loc_730>": 50999,
+  "<loc_731>": 51000,
+  "<loc_732>": 51001,
+  "<loc_733>": 51002,
+  "<loc_734>": 51003,
+  "<loc_735>": 51004,
+  "<loc_736>": 51005,
+  "<loc_737>": 51006,
+  "<loc_738>": 51007,
+  "<loc_739>": 51008,
+  "<loc_73>": 50342,
+  "<loc_740>": 51009,
+  "<loc_741>": 51010,
+  "<loc_742>": 51011,
+  "<loc_743>": 51012,
+  "<loc_744>": 51013,
+  "<loc_745>": 51014,
+  "<loc_746>": 51015,
+  "<loc_747>": 51016,
+  "<loc_748>": 51017,
+  "<loc_749>": 51018,
+  "<loc_74>": 50343,
+  "<loc_750>": 51019,
+  "<loc_751>": 51020,
+  "<loc_752>": 51021,
+  "<loc_753>": 51022,
+  "<loc_754>": 51023,
+  "<loc_755>": 51024,
+  "<loc_756>": 51025,
+  "<loc_757>": 51026,
+  "<loc_758>": 51027,
+  "<loc_759>": 51028,
+  "<loc_75>": 50344,
+  "<loc_760>": 51029,
+  "<loc_761>": 51030,
+  "<loc_762>": 51031,
+  "<loc_763>": 51032,
+  "<loc_764>": 51033,
+  "<loc_765>": 51034,
+  "<loc_766>": 51035,
+  "<loc_767>": 51036,
+  "<loc_768>": 51037,
+  "<loc_769>": 51038,
+  "<loc_76>": 50345,
+  "<loc_770>": 51039,
+  "<loc_771>": 51040,
+  "<loc_772>": 51041,
+  "<loc_773>": 51042,
+  "<loc_774>": 51043,
+  "<loc_775>": 51044,
+  "<loc_776>": 51045,
+  "<loc_777>": 51046,
+  "<loc_778>": 51047,
+  "<loc_779>": 51048,
+  "<loc_77>": 50346,
+  "<loc_780>": 51049,
+  "<loc_781>": 51050,
+  "<loc_782>": 51051,
+  "<loc_783>": 51052,
+  "<loc_784>": 51053,
+  "<loc_785>": 51054,
+  "<loc_786>": 51055,
+  "<loc_787>": 51056,
+  "<loc_788>": 51057,
+  "<loc_789>": 51058,
+  "<loc_78>": 50347,
+  "<loc_790>": 51059,
+  "<loc_791>": 51060,
+  "<loc_792>": 51061,
+  "<loc_793>": 51062,
+  "<loc_794>": 51063,
+  "<loc_795>": 51064,
+  "<loc_796>": 51065,
+  "<loc_797>": 51066,
+  "<loc_798>": 51067,
+  "<loc_799>": 51068,
+  "<loc_79>": 50348,
+  "<loc_7>": 50276,
+  "<loc_800>": 51069,
+  "<loc_801>": 51070,
+  "<loc_802>": 51071,
+  "<loc_803>": 51072,
+  "<loc_804>": 51073,
+  "<loc_805>": 51074,
+  "<loc_806>": 51075,
+  "<loc_807>": 51076,
+  "<loc_808>": 51077,
+  "<loc_809>": 51078,
+  "<loc_80>": 50349,
+  "<loc_810>": 51079,
+  "<loc_811>": 51080,
+  "<loc_812>": 51081,
+  "<loc_813>": 51082,
+  "<loc_814>": 51083,
+  "<loc_815>": 51084,
+  "<loc_816>": 51085,
+  "<loc_817>": 51086,
+  "<loc_818>": 51087,
+  "<loc_819>": 51088,
+  "<loc_81>": 50350,
+  "<loc_820>": 51089,
+  "<loc_821>": 51090,
+  "<loc_822>": 51091,
+  "<loc_823>": 51092,
+  "<loc_824>": 51093,
+  "<loc_825>": 51094,
+  "<loc_826>": 51095,
+  "<loc_827>": 51096,
+  "<loc_828>": 51097,
+  "<loc_829>": 51098,
+  "<loc_82>": 50351,
+  "<loc_830>": 51099,
+  "<loc_831>": 51100,
+  "<loc_832>": 51101,
+  "<loc_833>": 51102,
+  "<loc_834>": 51103,
+  "<loc_835>": 51104,
+  "<loc_836>": 51105,
+  "<loc_837>": 51106,
+  "<loc_838>": 51107,
+  "<loc_839>": 51108,
+  "<loc_83>": 50352,
+  "<loc_840>": 51109,
+  "<loc_841>": 51110,
+  "<loc_842>": 51111,
+  "<loc_843>": 51112,
+  "<loc_844>": 51113,
+  "<loc_845>": 51114,
+  "<loc_846>": 51115,
+  "<loc_847>": 51116,
+  "<loc_848>": 51117,
+  "<loc_849>": 51118,
+  "<loc_84>": 50353,
+  "<loc_850>": 51119,
+  "<loc_851>": 51120,
+  "<loc_852>": 51121,
+  "<loc_853>": 51122,
+  "<loc_854>": 51123,
+  "<loc_855>": 51124,
+  "<loc_856>": 51125,
+  "<loc_857>": 51126,
+  "<loc_858>": 51127,
+  "<loc_859>": 51128,
+  "<loc_85>": 50354,
+  "<loc_860>": 51129,
+  "<loc_861>": 51130,
+  "<loc_862>": 51131,
+  "<loc_863>": 51132,
+  "<loc_864>": 51133,
+  "<loc_865>": 51134,
+  "<loc_866>": 51135,
+  "<loc_867>": 51136,
+  "<loc_868>": 51137,
+  "<loc_869>": 51138,
+  "<loc_86>": 50355,
+  "<loc_870>": 51139,
+  "<loc_871>": 51140,
+  "<loc_872>": 51141,
+  "<loc_873>": 51142,
+  "<loc_874>": 51143,
+  "<loc_875>": 51144,
+  "<loc_876>": 51145,
+  "<loc_877>": 51146,
+  "<loc_878>": 51147,
+  "<loc_879>": 51148,
+  "<loc_87>": 50356,
+  "<loc_880>": 51149,
+  "<loc_881>": 51150,
+  "<loc_882>": 51151,
+  "<loc_883>": 51152,
+  "<loc_884>": 51153,
+  "<loc_885>": 51154,
+  "<loc_886>": 51155,
+  "<loc_887>": 51156,
+  "<loc_888>": 51157,
+  "<loc_889>": 51158,
+  "<loc_88>": 50357,
+  "<loc_890>": 51159,
+  "<loc_891>": 51160,
+  "<loc_892>": 51161,
+  "<loc_893>": 51162,
+  "<loc_894>": 51163,
+  "<loc_895>": 51164,
+  "<loc_896>": 51165,
+  "<loc_897>": 51166,
+  "<loc_898>": 51167,
+  "<loc_899>": 51168,
+  "<loc_89>": 50358,
+  "<loc_8>": 50277,
+  "<loc_900>": 51169,
+  "<loc_901>": 51170,
+  "<loc_902>": 51171,
+  "<loc_903>": 51172,
+  "<loc_904>": 51173,
+  "<loc_905>": 51174,
+  "<loc_906>": 51175,
+  "<loc_907>": 51176,
+  "<loc_908>": 51177,
+  "<loc_909>": 51178,
+  "<loc_90>": 50359,
+  "<loc_910>": 51179,
+  "<loc_911>": 51180,
+  "<loc_912>": 51181,
+  "<loc_913>": 51182,
+  "<loc_914>": 51183,
+  "<loc_915>": 51184,
+  "<loc_916>": 51185,
+  "<loc_917>": 51186,
+  "<loc_918>": 51187,
+  "<loc_919>": 51188,
+  "<loc_91>": 50360,
+  "<loc_920>": 51189,
+  "<loc_921>": 51190,
+  "<loc_922>": 51191,
+  "<loc_923>": 51192,
+  "<loc_924>": 51193,
+  "<loc_925>": 51194,
+  "<loc_926>": 51195,
+  "<loc_927>": 51196,
+  "<loc_928>": 51197,
+  "<loc_929>": 51198,
+  "<loc_92>": 50361,
+  "<loc_930>": 51199,
+  "<loc_931>": 51200,
+  "<loc_932>": 51201,
+  "<loc_933>": 51202,
+  "<loc_934>": 51203,
+  "<loc_935>": 51204,
+  "<loc_936>": 51205,
+  "<loc_937>": 51206,
+  "<loc_938>": 51207,
+  "<loc_939>": 51208,
+  "<loc_93>": 50362,
+  "<loc_940>": 51209,
+  "<loc_941>": 51210,
+  "<loc_942>": 51211,
+  "<loc_943>": 51212,
+  "<loc_944>": 51213,
+  "<loc_945>": 51214,
+  "<loc_946>": 51215,
+  "<loc_947>": 51216,
+  "<loc_948>": 51217,
+  "<loc_949>": 51218,
+  "<loc_94>": 50363,
+  "<loc_950>": 51219,
+  "<loc_951>": 51220,
+  "<loc_952>": 51221,
+  "<loc_953>": 51222,
+  "<loc_954>": 51223,
+  "<loc_955>": 51224,
+  "<loc_956>": 51225,
+  "<loc_957>": 51226,
+  "<loc_958>": 51227,
+  "<loc_959>": 51228,
+  "<loc_95>": 50364,
+  "<loc_960>": 51229,
+  "<loc_961>": 51230,
+  "<loc_962>": 51231,
+  "<loc_963>": 51232,
+  "<loc_964>": 51233,
+  "<loc_965>": 51234,
+  "<loc_966>": 51235,
+  "<loc_967>": 51236,
+  "<loc_968>": 51237,
+  "<loc_969>": 51238,
+  "<loc_96>": 50365,
+  "<loc_970>": 51239,
+  "<loc_971>": 51240,
+  "<loc_972>": 51241,
+  "<loc_973>": 51242,
+  "<loc_974>": 51243,
+  "<loc_975>": 51244,
+  "<loc_976>": 51245,
+  "<loc_977>": 51246,
+  "<loc_978>": 51247,
+  "<loc_979>": 51248,
+  "<loc_97>": 50366,
+  "<loc_980>": 51249,
+  "<loc_981>": 51250,
+  "<loc_982>": 51251,
+  "<loc_983>": 51252,
+  "<loc_984>": 51253,
+  "<loc_985>": 51254,
+  "<loc_986>": 51255,
+  "<loc_987>": 51256,
+  "<loc_988>": 51257,
+  "<loc_989>": 51258,
+  "<loc_98>": 50367,
+  "<loc_990>": 51259,
+  "<loc_991>": 51260,
+  "<loc_992>": 51261,
+  "<loc_993>": 51262,
+  "<loc_994>": 51263,
+  "<loc_995>": 51264,
+  "<loc_996>": 51265,
+  "<loc_997>": 51266,
+  "<loc_998>": 51267,
+  "<loc_999>": 51268,
+  "<loc_99>": 50368,
+  "<loc_9>": 50278,
+  "<ncap>": 51271,
+  "<ocr>": 50267,
+  "<od>": 50265,
+  "<poly>": 51286,
+  "<proposal>": 51284,
+  "<region_cap>": 51280,
+  "<region_to_desciption>": 51282,
+  "<seg>": 51277,
+  "<sep>": 51279
+}

config.json

@@ -0,0 +1,237 @@
+{
+  "_name_or_path": "model_checkpoints/vqainstruct_no_lora/epoch_5",
+  "architectures": [
+    "Florence2ForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_florence2.Florence2Config",
+    "AutoModelForCausalLM": "modeling_florence2.Florence2ForConditionalGeneration"
+  },
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "ignore_index": -100,
+  "is_encoder_decoder": true,
+  "model_type": "florence2",
+  "pad_token_id": 1,
+  "projection_dim": 1024,
+  "text_config": {
+    "_name_or_path": "",
+    "activation_dropout": 0.1,
+    "activation_function": "gelu",
+    "add_bias_logits": false,
+    "add_cross_attention": false,
+    "add_final_layer_norm": false,
+    "architectures": null,
+    "attention_dropout": 0.1,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": 0,
+    "chunk_size_feed_forward": 0,
+    "classif_dropout": 0.1,
+    "classifier_dropout": 0.0,
+    "cross_attention_hidden_size": null,
+    "d_model": 1024,
+    "decoder_attention_heads": 16,
+    "decoder_ffn_dim": 4096,
+    "decoder_layerdrop": 0.0,
+    "decoder_layers": 12,
+    "decoder_start_token_id": 2,
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "dropout": 0.1,
+    "early_stopping": true,
+    "encoder_attention_heads": 16,
+    "encoder_ffn_dim": 4096,
+    "encoder_layerdrop": 0.0,
+    "encoder_layers": 12,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": 2,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": 0,
+    "forced_eos_token_id": 2,
+    "gradient_checkpointing": false,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1",
+      "2": "LABEL_2"
+    },
+    "init_std": 0.02,
+    "is_decoder": false,
+    "is_encoder_decoder": true,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1,
+      "LABEL_2": 2
+    },
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "max_position_embeddings": 1024,
+    "min_length": 0,
+    "model_type": "florence2_language",
+    "no_repeat_ngram_size": 3,
+    "normalize_before": false,
+    "num_beam_groups": 1,
+    "num_beams": 3,
+    "num_hidden_layers": 12,
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": 1,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "scale_embedding": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "use_cache": true,
+    "vocab_size": 51289
+  },
+  "torch_dtype": "float32",
+  "transformers_version": "4.41.2",
+  "vision_config": {
+    "_name_or_path": "",
+    "add_cross_attention": false,
+    "architectures": null,
+    "bad_words_ids": null,
+    "begin_suppress_tokens": null,
+    "bos_token_id": null,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "depths": [
+      1,
+      1,
+      9,
+      1
+    ],
+    "dim_embed": [
+      256,
+      512,
+      1024,
+      2048
+    ],
+    "diversity_penalty": 0.0,
+    "do_sample": false,
+    "drop_path_rate": 0.1,
+    "early_stopping": false,
+    "enable_checkpoint": false,
+    "encoder_no_repeat_ngram_size": 0,
+    "eos_token_id": null,
+    "exponential_decay_length_penalty": null,
+    "finetuning_task": null,
+    "forced_bos_token_id": null,
+    "forced_eos_token_id": null,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "image_feature_source": [
+      "spatial_avg_pool",
+      "temporal_avg_pool"
+    ],
+    "image_pos_embed": {
+      "max_pos_embeddings": 50,
+      "type": "learned_abs_2d"
+    },
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "length_penalty": 1.0,
+    "max_length": 20,
+    "min_length": 0,
+    "model_type": "davit",
+    "no_repeat_ngram_size": 0,
+    "num_beam_groups": 1,
+    "num_beams": 1,
+    "num_groups": [
+      8,
+      16,
+      32,
+      64
+    ],
+    "num_heads": [
+      8,
+      16,
+      32,
+      64
+    ],
+    "num_return_sequences": 1,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "output_scores": false,
+    "pad_token_id": null,
+    "patch_padding": [
+      3,
+      1,
+      1,
+      1
+    ],
+    "patch_prenorm": [
+      false,
+      true,
+      true,
+      true
+    ],
+    "patch_size": [
+      7,
+      3,
+      3,
+      3
+    ],
+    "patch_stride": [
+      4,
+      2,
+      2,
+      2
+    ],
+    "prefix": null,
+    "problem_type": null,
+    "projection_dim": 1024,
+    "pruned_heads": {},
+    "remove_invalid_values": false,
+    "repetition_penalty": 1.0,
+    "return_dict": true,
+    "return_dict_in_generate": false,
+    "sep_token_id": null,
+    "suppress_tokens": null,
+    "task_specific_params": null,
+    "temperature": 1.0,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": true,
+    "tokenizer_class": null,
+    "top_k": 50,
+    "top_p": 1.0,
+    "torch_dtype": null,
+    "torchscript": false,
+    "typical_p": 1.0,
+    "use_bfloat16": false,
+    "visual_temporal_embedding": {
+      "max_temporal_embeddings": 100,
+      "type": "COSINE"
+    },
+    "window_size": 12
+  },
+  "vocab_size": 51289
+}

configuration_florence2.py

@@ -0,0 +1,340 @@
+# coding=utf-8
+# Copyright 2024 Microsoft and the HuggingFace Inc. team. All rights reserved.
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import warnings
+""" Florence-2 configuration"""
+
+from typing import Optional
+
+from transformers import AutoConfig
+from transformers.configuration_utils import PretrainedConfig
+from transformers.utils import logging
+
+logger = logging.get_logger(__name__)
+
+class Florence2VisionConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Florence2VisionModel`]. It is used to instantiate a Florence2VisionModel
+    according to the specified arguments, defining the model architecture. Instantiating a configuration with the 
+    defaults will yield a similar configuration to that of the Florence2VisionModel architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        drop_path_rate (`float`, *optional*, defaults to 0.1):
+            The dropout rate of the drop path layer.
+        patch_size (`List[int]`, *optional*, defaults to [7, 3, 3, 3]):
+            The patch size of the image.
+        patch_stride (`List[int]`, *optional*, defaults to [4, 2, 2, 2]):
+            The patch stride of the image.
+        patch_padding (`List[int]`, *optional*, defaults to [3, 1, 1, 1]):
+            The patch padding of the image.
+        patch_prenorm (`List[bool]`, *optional*, defaults to [false, true, true, true]):
+            Whether to apply layer normalization before the patch embedding layer.
+        enable_checkpoint (`bool`, *optional*, defaults to False):
+            Whether to enable checkpointing.
+        dim_embed (`List[int]`, *optional*, defaults to [256, 512, 1024, 2048]):
+            The dimension of the embedding layer.
+        num_heads (`List[int]`, *optional*, defaults to [8, 16, 32, 64]):
+            The number of attention heads.
+        num_groups (`List[int]`, *optional*, defaults to [8, 16, 32, 64]):
+            The number of groups.
+        depths (`List[int]`, *optional*, defaults to [1, 1, 9, 1]):
+            The depth of the model.
+        window_size (`int`, *optional*, defaults to 12):
+            The window size of the model.
+        projection_dim (`int`, *optional*, defaults to 1024):
+            The dimension of the projection layer.
+        visual_temporal_embedding (`dict`, *optional*):
+            The configuration of the visual temporal embedding.
+        image_pos_embed (`dict`, *optional*):
+            The configuration of the image position embedding.
+        image_feature_source (`List[str]`, *optional*, defaults to ["spatial_avg_pool", "temporal_avg_pool"]):
+            The source of the image feature.
+    Example:
+
+    ```python
+    >>> from transformers import Florence2VisionConfig, Florence2VisionModel
+
+    >>> # Initializing a Florence2 Vision style configuration
+    >>> configuration = Florence2VisionConfig()
+
+    >>> # Initializing a model (with random weights)
+    >>> model = Florence2VisionModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "florence2_vision"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        drop_path_rate=0.1,
+        patch_size=[7, 3, 3, 3],
+        patch_stride=[4, 2, 2, 2],
+        patch_padding=[3, 1, 1, 1],
+        patch_prenorm=[False, True, True, True],
+        enable_checkpoint=False,
+        dim_embed=[256, 512, 1024, 2048],
+        num_heads=[8, 16, 32, 64],
+        num_groups=[8, 16, 32, 64],
+        depths=[1, 1, 9, 1],
+        window_size=12,
+        projection_dim=1024,
+        visual_temporal_embedding=None,
+        image_pos_embed=None,
+        image_feature_source=["spatial_avg_pool", "temporal_avg_pool"],
+        **kwargs,
+    ):
+        self.drop_path_rate = drop_path_rate
+        self.patch_size = patch_size
+        self.patch_stride = patch_stride
+        self.patch_padding = patch_padding
+        self.patch_prenorm = patch_prenorm
+        self.enable_checkpoint = enable_checkpoint
+        self.dim_embed = dim_embed
+        self.num_heads = num_heads
+        self.num_groups = num_groups
+        self.depths = depths
+        self.window_size = window_size
+        self.projection_dim = projection_dim
+        self.visual_temporal_embedding = visual_temporal_embedding
+        self.image_pos_embed = image_pos_embed
+        self.image_feature_source = image_feature_source
+
+        super().__init__(**kwargs)
+
+
+
+class Florence2LanguageConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Florence2LanguagePreTrainedModel`]. It is used to instantiate a BART
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the BART
+    [facebook/bart-large](https://huggingface.co/facebook/bart-large) architecture.
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 51289):
+            Vocabulary size of the Florence2Language model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`Florence2LanguageModel`].
+        d_model (`int`, *optional*, defaults to 1024):
+            Dimensionality of the layers and the pooler layer.
+        encoder_layers (`int`, *optional*, defaults to 12):
+            Number of encoder layers.
+        decoder_layers (`int`, *optional*, defaults to 12):
+            Number of decoder layers.
+        encoder_attention_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        decoder_attention_heads (`int`, *optional*, defaults to 16):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        decoder_ffn_dim (`int`, *optional*, defaults to 4096):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in decoder.
+        encoder_ffn_dim (`int`, *optional*, defaults to 4096):
+            Dimensionality of the "intermediate" (often named feed-forward) layer in decoder.
+        activation_function (`str` or `function`, *optional*, defaults to `"gelu"`):
+            The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`,
+            `"relu"`, `"silu"` and `"gelu_new"` are supported.
+        dropout (`float`, *optional*, defaults to 0.1):
+            The dropout probability for all fully connected layers in the embeddings, encoder, and pooler.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        activation_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for activations inside the fully connected layer.
+        classifier_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for classifier.
+        max_position_embeddings (`int`, *optional*, defaults to 1024):
+            The maximum sequence length that this model might ever be used with. Typically set this to something large
+            just in case (e.g., 512 or 1024 or 2048).
+        init_std (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        encoder_layerdrop (`float`, *optional*, defaults to 0.0):
+            The LayerDrop probability for the encoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556)
+            for more details.
+        decoder_layerdrop (`float`, *optional*, defaults to 0.0):
+            The LayerDrop probability for the decoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556)
+            for more details.
+        scale_embedding (`bool`, *optional*, defaults to `False`):
+            Scale embeddings by diving by sqrt(d_model).
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models).
+        num_labels (`int`, *optional*, defaults to 3):
+            The number of labels to use in [`Florence2LanguageForSequenceClassification`].
+        forced_eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the token to force as the last generated token when `max_length` is reached. Usually set to
+            `eos_token_id`.
+
+    Example:
+
+    ```python
+    >>> from transformers import Florence2LanguageConfig, Florence2LanguageModel
+
+    >>> # Initializing a Florence2 Language style configuration
+    >>> configuration = Florence2LanguageConfig()
+
+    >>> # Initializing a model (with random weights)
+    >>> model = Florence2LangaugeModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "florence2_language"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    attribute_map = {"num_attention_heads": "encoder_attention_heads", "hidden_size": "d_model"}
+
+    def __init__(
+        self,
+        vocab_size=51289,
+        max_position_embeddings=1024,
+        encoder_layers=12,
+        encoder_ffn_dim=4096,
+        encoder_attention_heads=16,
+        decoder_layers=12,
+        decoder_ffn_dim=4096,
+        decoder_attention_heads=16,
+        encoder_layerdrop=0.0,
+        decoder_layerdrop=0.0,
+        activation_function="gelu",
+        d_model=1024,
+        dropout=0.1,
+        attention_dropout=0.0,
+        activation_dropout=0.0,
+        init_std=0.02,
+        classifier_dropout=0.0,
+        scale_embedding=False,
+        use_cache=True,
+        num_labels=3,
+        pad_token_id=1,
+        bos_token_id=0,
+        eos_token_id=2,
+        is_encoder_decoder=True,
+        decoder_start_token_id=2,
+        forced_eos_token_id=2,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.d_model = d_model
+        self.encoder_ffn_dim = encoder_ffn_dim
+        self.encoder_layers = encoder_layers
+        self.encoder_attention_heads = encoder_attention_heads
+        self.decoder_ffn_dim = decoder_ffn_dim
+        self.decoder_layers = decoder_layers
+        self.decoder_attention_heads = decoder_attention_heads
+        self.dropout = dropout
+        self.attention_dropout = attention_dropout
+        self.activation_dropout = activation_dropout
+        self.activation_function = activation_function
+        self.init_std = init_std
+        self.encoder_layerdrop = encoder_layerdrop
+        self.decoder_layerdrop = decoder_layerdrop
+        self.classifier_dropout = classifier_dropout
+        self.use_cache = use_cache
+        self.num_hidden_layers = encoder_layers
+        self.scale_embedding = scale_embedding  # scale factor will be sqrt(d_model) if True
+
+        super().__init__(
+            num_labels=num_labels,
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            is_encoder_decoder=is_encoder_decoder,
+            decoder_start_token_id=decoder_start_token_id,
+            forced_eos_token_id=forced_eos_token_id,
+            **kwargs,
+        )
+
+        # ensure backward compatibility for BART CNN models
+        if self.forced_bos_token_id is None and kwargs.get("force_bos_token_to_be_generated", False):
+            self.forced_bos_token_id = self.bos_token_id
+            warnings.warn(
+                f"Please make sure the config includes `forced_bos_token_id={self.bos_token_id}` in future versions. "
+                "The config can simply be saved and uploaded again to be fixed."
+            )
+
+class Florence2Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`Florence2ForConditionalGeneration`]. It is used to instantiate an
+    Florence-2 model according to the specified arguments, defining the model architecture. 
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+    Args:
+        vision_config (`Florence2VisionConfig`,  *optional*):
+            Custom vision config or dict
+        text_config (`Union[AutoConfig, dict]`, *optional*):
+            The config object of the text backbone. 
+        ignore_index (`int`, *optional*, defaults to -100):
+            The ignore index for the loss function.
+        vocab_size (`int`, *optional*, defaults to 51289):
+            Vocabulary size of the Florence2model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`~Florence2ForConditionalGeneration`]
+        projection_dim (`int`, *optional*, defaults to 1024):
+            Dimension of the multimodal projection space.
+
+    Example:
+
+    ```python
+    >>> from transformers import Florence2ForConditionalGeneration, Florence2Config, CLIPVisionConfig, BartConfig
+
+    >>> # Initializing a clip-like vision config
+    >>> vision_config = CLIPVisionConfig()
+
+    >>> # Initializing a Bart config
+    >>> text_config = BartConfig()
+
+    >>> # Initializing a Florence-2 configuration
+    >>> configuration = Florence2Config(vision_config, text_config)
+
+    >>> # Initializing a model from the florence-2 configuration
+    >>> model = Florence2ForConditionalGeneration(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "florence2"
+    is_composition = False
+
+    def __init__(
+        self,
+        vision_config=None,
+        text_config=None,
+        ignore_index=-100,
+        vocab_size=51289,
+        projection_dim=1024,
+        **kwargs,
+    ):
+        self.ignore_index = ignore_index
+        self.vocab_size = vocab_size
+        self.projection_dim = projection_dim
+        if vision_config is not None:
+            vision_config = PretrainedConfig(**vision_config)
+        self.vision_config = vision_config
+        self.vocab_size = self.vocab_size
+
+        self.text_config = text_config
+        if text_config is not None:
+            self.text_config = Florence2LanguageConfig(**text_config)
+
+
+        super().__init__(**kwargs)
+

generation_config.json

@@ -0,0 +1,4 @@
+{
+  "num_beams": 3,
+  "transformers_version": "4.41.2"
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d9a3bc6abcace5e9820630945fe26cfa961fe2577f8adeb48256acba876123e
+size 3291921348

preprocessor_config.json

@@ -0,0 +1,50 @@
+{
+  "_valid_processor_keys": [
+    "images",
+    "do_resize",
+    "size",
+    "resample",
+    "do_center_crop",
+    "crop_size",
+    "do_rescale",
+    "rescale_factor",
+    "do_normalize",
+    "image_mean",
+    "image_std",
+    "do_convert_rgb",
+    "return_tensors",
+    "data_format",
+    "input_data_format"
+  ],
+  "auto_map": {
+    "AutoProcessor": "processing_florence2.Florence2Processor"
+  },
+  "crop_size": {
+    "height": 768,
+    "width": 768
+  },
+  "do_center_crop": false,
+  "do_convert_rgb": null,
+  "do_normalize": true,
+  "do_rescale": true,
+  "do_resize": true,
+  "image_mean": [
+    0.485,
+    0.456,
+    0.406
+  ],
+  "image_processor_type": "CLIPImageProcessor",
+  "image_seq_length": 577,
+  "image_std": [
+    0.229,
+    0.224,
+    0.225
+  ],
+  "processor_class": "Florence2Processor",
+  "resample": 3,
+  "rescale_factor": 0.00392156862745098,
+  "size": {
+    "height": 768,
+    "width": 768
+  }
+}

processing_florence2.py

@@ -0,0 +1,1088 @@
+# coding=utf-8
+# Copyright 2024 Microsoft and The HuggingFace Inc. team.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""
+Processor class for Florence-2.
+"""
+
+import re
+import logging
+from typing import List, Optional, Union
+import numpy as np
+
+import torch
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.image_utils import ImageInput, is_valid_image
+from transformers.processing_utils import ProcessorMixin
+from transformers.tokenization_utils_base import (
+    PaddingStrategy,
+    PreTokenizedInput,
+    TextInput,
+    TruncationStrategy,
+)
+from transformers.utils import TensorType
+
+
+logger = logging.getLogger(__name__)
+
+# Copied from transformers.models.idefics2.processing_idefics2.is_url
+def is_url(val) -> bool:
+    return isinstance(val, str) and val.startswith("http")
+
+# Copied from transformers.models.idefics2.processing_idefics2.is_image_or_image_url
+def is_image_or_image_url(elem):
+    return is_url(elem) or is_valid_image(elem)
+
+
+def _is_str_or_image(elem):
+    return isinstance(elem, (str)) or is_image_or_image_url(elem)
+
+
+class Florence2Processor(ProcessorMixin):
+    r"""
+    Constructs a Florence2 processor which wraps a Florence2 image processor and a Florence2 tokenizer into a single processor.
+
+    [`Florence2Processor`] offers all the functionalities of [`CLIPImageProcessor`] and [`BartTokenizerFast`]. See the
+    [`~Florence2Processor.__call__`] and [`~Florence2Processor.decode`] for more information.
+
+    Args:
+        image_processor ([`CLIPImageProcessor`], *optional*):
+            The image processor is a required input.
+        tokenizer ([`BartTokenizerFast`], *optional*):
+            The tokenizer is a required input.
+    """
+
+    attributes = ["image_processor", "tokenizer"]
+    image_processor_class = "CLIPImageProcessor"
+    tokenizer_class = ("BartTokenizer", "BartTokenizerFast")
+
+    def __init__(
+        self,
+        image_processor=None,
+        tokenizer=None,
+    ):
+        if image_processor is None:
+            raise ValueError("You need to specify an `image_processor`.")
+        if tokenizer is None:
+            raise ValueError("You need to specify a `tokenizer`.")
+        if not hasattr(image_processor, "image_seq_length"):
+            raise ValueError("Image processor is missing an `image_seq_length` attribute.")
+
+        self.image_seq_length = image_processor.image_seq_length
+
+        tokens_to_add = {
+                'additional_special_tokens': \
+                    tokenizer.additional_special_tokens + \
+                    ['<od>', '</od>', '<ocr>', '</ocr>'] + \
+                    [f'<loc_{x}>' for x in range(1000)] + \
+                    ['<cap>', '</cap>', '<ncap>', '</ncap>','<dcap>', '</dcap>', '<grounding>', '</grounding>', '<seg>', '</seg>', '<sep>', '<region_cap>', '</region_cap>', '<region_to_desciption>', '</region_to_desciption>', '<proposal>', '</proposal>', '<poly>', '</poly>', '<and>']
+            }
+        tokenizer.add_special_tokens(tokens_to_add)
+
+        self.tasks_answer_post_processing_type = {
+            '<OCR>': 'pure_text',
+            '<OCR_WITH_REGION>': 'ocr',
+            '<CAPTION>': 'pure_text',
+            '<DETAILED_CAPTION>': 'pure_text',
+            '<MORE_DETAILED_CAPTION>': 'pure_text',
+            '<OD>': 'description_with_bboxes',
+            '<DENSE_REGION_CAPTION>': 'description_with_bboxes',
+            '<CAPTION_TO_PHRASE_GROUNDING>': "phrase_grounding",
+            '<REFERRING_EXPRESSION_SEGMENTATION>': 'polygons',
+            '<REGION_TO_SEGMENTATION>': 'polygons',
+            '<OPEN_VOCABULARY_DETECTION>': 'description_with_bboxes_or_polygons',
+            '<REGION_TO_CATEGORY>': 'pure_text',
+            '<REGION_TO_DESCRIPTION>': 'pure_text',
+            '<REGION_TO_OCR>': 'pure_text',
+            '<REGION_PROPOSAL>': 'bboxes'
+        }
+
+        self.task_prompts_without_inputs = {
+            '<OCR>': 'What is the text in the image?',
+            '<OCR_WITH_REGION>': 'What is the text in the image, with regions?',
+            '<CAPTION>': 'What does the image describe?',
+            '<DETAILED_CAPTION>': 'Describe in detail what is shown in the image.',
+            '<MORE_DETAILED_CAPTION>': 'Describe with a paragraph what is shown in the image.',
+            '<OD>': 'Locate the objects with category name in the image.',
+            '<DENSE_REGION_CAPTION>': 'Locate the objects in the image, with their descriptions.',
+            '<REGION_PROPOSAL>': 'Locate the region proposals in the image.'
+        }
+
+        self.task_prompts_with_input = {
+            '<CAPTION_TO_PHRASE_GROUNDING>': "Locate the phrases in the caption: {input}",
+            '<REFERRING_EXPRESSION_SEGMENTATION>': 'Locate {input} in the image with mask',
+            '<REGION_TO_SEGMENTATION>': 'What is the polygon mask of region {input}',
+            '<OPEN_VOCABULARY_DETECTION>': 'Locate {input} in the image.',
+            '<REGION_TO_CATEGORY>': 'What is the region {input}?',
+            '<REGION_TO_DESCRIPTION>': 'What does the region {input} describe?',
+            '<REGION_TO_OCR>': 'What text is in the region {input}?',
+        }
+
+        self.post_processor = Florence2PostProcesser(tokenizer=tokenizer)
+
+
+        super().__init__(image_processor, tokenizer)
+    
+    def _construct_prompts(self, text):
+        # replace the task tokens with the task prompts if task token is in the text
+        prompts = []
+        for _text in text:
+            # 1. fixed task prompts without additional inputs
+            for task_token, task_prompt in self.task_prompts_without_inputs.items():
+                if task_token in _text:
+                    assert _text == task_token, f"Task token {task_token} should be the only token in the text."
+                    _text = task_prompt
+                    break
+            # 2. task prompts with additional inputs 
+            for task_token, task_prompt in self.task_prompts_with_input.items():
+                if task_token in _text:
+                    _text = task_prompt.format(input=_text.replace(task_token, ''))
+                    break
+            prompts.append(_text)
+        return prompts
+
+    def __call__(
+        self,
+        text: Union[TextInput, PreTokenizedInput, List[TextInput], List[PreTokenizedInput]] = None,
+        images: ImageInput = None,
+        tokenize_newline_separately: bool = True,
+        padding: Union[bool, str, PaddingStrategy] = False,
+        truncation: Union[bool, str, TruncationStrategy] = None,
+        max_length=None,
+        return_tensors: Optional[Union[str, TensorType]] = TensorType.PYTORCH,
+        do_resize: bool = None,
+        do_normalize: bool = None,
+        image_mean: Optional[Union[float, List[float]]] = None,
+        image_std: Optional[Union[float, List[float]]] = None,
+        data_format: Optional["ChannelDimension"] = "channels_first",  # noqa: F821
+        input_data_format: Optional[
+            Union[str, "ChannelDimension"]  # noqa: F821
+        ] = None,
+        resample: "PILImageResampling" = None,  # noqa: F821
+        do_convert_rgb: bool = None,
+        do_thumbnail: bool = None,
+        do_align_long_axis: bool = None,
+        do_rescale: bool = None,
+    ) -> BatchFeature:
+        """
+        Main method to prepare for the model one or several sequences(s) and image(s). This method forwards the `text`
+        and `kwargs` arguments to BartTokenizerFast's [`~BartTokenizerFast.__call__`] if `text` is not `None` to encode
+        the text. To prepare the image(s), this method forwards the `images` and `kwrags` arguments to
+        CLIPImageProcessor's [`~CLIPImageProcessor.__call__`] if `images` is not `None`. Please refer to the doctsring
+        of the above two methods for more information.
+
+        Args:
+            text (`str`, `List[str]`, `List[List[str]]`):
+                The sequence or batch of sequences to be encoded. Each sequence can be a string or a list of strings
+                (pretokenized string). If the sequences are provided as list of strings (pretokenized), you must set
+                `is_split_into_words=True` (to lift the ambiguity with a batch of sequences).
+            images (`PIL.Image.Image`, `np.ndarray`, `torch.Tensor`, `List[PIL.Image.Image]`, `List[np.ndarray]`, `List[torch.Tensor]`):
+                The image or batch of images to be prepared. Each image can be a PIL image, NumPy array or PyTorch
+                tensor. In case of a NumPy array/PyTorch tensor, each image should be of shape (C, H, W), where C is a
+                number of channels, H and W are image height and width.
+            tokenize_newline_separately (`bool`, defaults to `True`):
+                Adds a separately tokenized '\n' at the end of the prompt.
+            padding (`bool`, `str` or [`~utils.PaddingStrategy`], *optional*, defaults to `False`):
+                Select a strategy to pad the returned sequences (according to the model's padding side and padding
+                index) among:
+                - `True` or `'longest'`: Pad to the longest sequence in the batch (or no padding if only a single
+                  sequence if provided).
+                - `'max_length'`: Pad to a maximum length specified with the argument `max_length` or to the maximum
+                  acceptable input length for the model if that argument is not provided.
+                - `False` or `'do_not_pad'` (default): No padding (i.e., can output a batch with sequences of different
+                  lengths).
+            max_length (`int`, *optional*):
+                Maximum length of the returned list and optionally padding length (see above).
+            truncation (`bool`, *optional*):
+                Activates truncation to cut input sequences longer than `max_length` to `max_length`.
+            return_tensors (`str` or [`~utils.TensorType`], *optional*):
+                If set, will return tensors of a particular framework. Acceptable values are:
+
+                - `'tf'`: Return TensorFlow `tf.constant` objects.
+                - `'pt'`: Return PyTorch `torch.Tensor` objects.
+                - `'np'`: Return NumPy `np.ndarray` objects.
+                - `'jax'`: Return JAX `jnp.ndarray` objects.
+
+        Returns:
+            [`BatchFeature`]: A [`BatchFeature`] with the following fields:
+
+            - **input_ids** -- List of token ids to be fed to a model. Returned when `text` is not `None`. If `suffix`
+              is provided, the `input_ids` will also contain the suffix input ids.
+            - **attention_mask** -- List of indices specifying which tokens should be attended to by the model (when
+              `return_attention_mask=True` or if *"attention_mask"* is in `self.model_input_names` and if `text` is not
+              `None`).
+            - **pixel_values** -- Pixel values to be fed to a model. Returned when `images` is not `None`.
+            - **labels** -- Labels compatible with training if `suffix` is not None
+        """
+
+        return_token_type_ids = False
+
+        if images is None:
+            raise ValueError("`images` are expected as arguments to a `Florence2Processor` instance.")
+        if text is None:
+            logger.warning_once(
+                "You are using Florence-2 without a text prompt."
+            )
+            text = ""
+
+        if isinstance(text, List) and isinstance(images, List):
+            if len(images) < len(text):
+                raise ValueError(
+                    f"Received {len(images)} images for {len(text)} prompts. Each prompt should be associated with an image."
+                )
+        if _is_str_or_image(text):
+            text = [text]
+        elif isinstance(text, list) and _is_str_or_image(text[0]):
+            pass
+
+        pixel_values = self.image_processor(
+            images,
+            do_resize=do_resize,
+            do_normalize=do_normalize,
+            return_tensors=return_tensors,
+            image_mean=image_mean,
+            image_std=image_std,
+            input_data_format=input_data_format,
+            data_format=data_format,
+            resample=resample,
+            do_convert_rgb=do_convert_rgb,
+        )["pixel_values"]
+
+        if max_length is not None:
+            max_length -= self.image_seq_length  # max_length has to account for the image tokens
+
+        text = self._construct_prompts(text)
+
+        inputs = self.tokenizer(
+            text,
+            return_tensors=return_tensors,
+            padding=padding,
+            max_length=max_length,
+            truncation=truncation,
+            return_token_type_ids=return_token_type_ids,
+        )
+
+        return_data = {**inputs, "pixel_values": pixel_values}
+
+        if return_token_type_ids:
+            labels = inputs["input_ids"].masked_fill(inputs["token_type_ids"] == 0, -100)
+            return_data.update({"labels": labels})
+        return BatchFeature(data=return_data)
+
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.batch_decode with CLIP->Florence2
+    def batch_decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to BartTokenizerFast's [`~PreTrainedTokenizer.batch_decode`]. Please
+        refer to the docstring of this method for more information.
+        """
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.decode with CLIP->Florence2
+    def decode(self, *args, **kwargs):
+        """
+        This method forwards all its arguments to BartTokenizerFast's [`~PreTrainedTokenizer.decode`]. Please refer to
+        the docstring of this method for more information.
+        """
+        return self.tokenizer.decode(*args, **kwargs)
+
+    @property
+    # Copied from transformers.models.clip.processing_clip.CLIPProcessor.model_input_names with CLIP->Florence2
+    def model_input_names(self):
+        tokenizer_input_names = self.tokenizer.model_input_names
+        image_processor_input_names = self.image_processor.model_input_names
+        return list(dict.fromkeys(tokenizer_input_names + image_processor_input_names))
+
+    def post_process_generation(self, text, task, image_size):
+        """
+        Post-process the output of the model to each of the task outputs.
+
+        Args:
+            text (`str`): The text to post-process.
+            task (`str`): The task to post-process the text for.
+            image_size (`Tuple[int, int]`): The size of the image. height x width.
+        """
+
+        task_answer_post_processing_type = self.tasks_answer_post_processing_type.get(task, 'pure_text')
+        task_answer = self.post_processor(
+            text=text,
+            image_size=image_size,
+            parse_tasks=task_answer_post_processing_type,
+        )[task_answer_post_processing_type]
+
+        if task_answer_post_processing_type == 'pure_text':
+            final_answer = task_answer
+            # remove the special tokens
+            final_answer = final_answer.replace('<s>', '').replace('</s>', '')
+        elif task_answer_post_processing_type in ['od', 'description_with_bboxes', 'bboxes']:
+            od_instances = task_answer
+            bboxes_od = [_od_instance['bbox'] for _od_instance in od_instances]
+            labels_od = [str(_od_instance['cat_name']) for _od_instance in od_instances]
+            final_answer = {'bboxes': bboxes_od, 'labels': labels_od}
+        elif task_answer_post_processing_type in ['ocr']:
+            bboxes = [_od_instance['quad_box'] for _od_instance in task_answer]
+            labels = [str(_od_instance['text']) for _od_instance in task_answer]
+            final_answer = {'quad_boxes': bboxes, 'labels': labels}
+        elif task_answer_post_processing_type in ['phrase_grounding']:
+            bboxes = []
+            labels = []
+            for _grounded_phrase in task_answer:
+                for _bbox in _grounded_phrase['bbox']:
+                    bboxes.append(_bbox)
+                    labels.append(_grounded_phrase['cat_name'])
+            final_answer = {'bboxes': bboxes, 'labels': labels}
+        elif task_answer_post_processing_type in ['description_with_polygons', 'polygons']:
+            labels = []
+            polygons = []
+            for result in task_answer:
+                label = result['cat_name']
+                _polygons = result['polygons']
+                labels.append(label)
+                polygons.append(_polygons)
+            final_answer = {'polygons': polygons, 'labels': labels}
+        elif task_answer_post_processing_type in ['description_with_bboxes_or_polygons']:
+            bboxes = []
+            bboxes_labels = []
+            polygons = []
+            polygons_labels = []
+            for result in task_answer:
+                label = result['cat_name']
+                if 'polygons' in result:
+                    _polygons = result['polygons']
+                    polygons.append(_polygons)
+                    polygons_labels.append(label)
+                else:
+                    _bbox = result['bbox']
+                    bboxes.append(_bbox)
+                    bboxes_labels.append(label)
+            final_answer = {'bboxes': bboxes, 'bboxes_labels': bboxes_labels, 'polygons': polygons, 'polygons_labels': polygons_labels}
+        else:
+            raise ValueError('Unknown task answer post processing type: {}'.format(task_answer_post_processing_type))
+
+        final_answer = {
+            task: final_answer}
+        return final_answer 
+
+class BoxQuantizer(object):
+    def __init__(self, mode, bins):
+        self.mode = mode
+        self.bins = bins
+
+    def quantize(self, boxes: torch.Tensor, size):
+        bins_w, bins_h = self.bins  # Quantization bins.
+        size_w, size_h = size       # Original image size.
+        size_per_bin_w = size_w / bins_w
+        size_per_bin_h = size_h / bins_h
+        xmin, ymin, xmax, ymax = boxes.split(1, dim=-1)  # Shape: 4 * [N, 1].
+
+        if self.mode == 'floor':
+            quantized_xmin = (
+                xmin / size_per_bin_w).floor().clamp(0, bins_w - 1)
+            quantized_ymin = (
+                ymin / size_per_bin_h).floor().clamp(0, bins_h - 1)
+            quantized_xmax = (
+                xmax / size_per_bin_w).floor().clamp(0, bins_w - 1)
+            quantized_ymax = (
+                ymax / size_per_bin_h).floor().clamp(0, bins_h - 1)
+
+        elif self.mode == 'round':
+            raise NotImplementedError()
+
+        else:
+            raise ValueError('Incorrect quantization type.')
+
+        quantized_boxes = torch.cat(
+            (quantized_xmin, quantized_ymin, quantized_xmax, quantized_ymax), dim=-1
+        ).int()
+
+        return quantized_boxes
+
+    def dequantize(self, boxes: torch.Tensor, size):
+        bins_w, bins_h = self.bins  # Quantization bins.
+        size_w, size_h = size       # Original image size.
+        size_per_bin_w = size_w / bins_w
+        size_per_bin_h = size_h / bins_h
+        xmin, ymin, xmax, ymax = boxes.split(1, dim=-1)  # Shape: 4 * [N, 1].
+
+        if self.mode == 'floor':
+            # Add 0.5 to use the center position of the bin as the coordinate.
+            dequantized_xmin = (xmin + 0.5) * size_per_bin_w
+            dequantized_ymin = (ymin + 0.5) * size_per_bin_h
+            dequantized_xmax = (xmax + 0.5) * size_per_bin_w
+            dequantized_ymax = (ymax + 0.5) * size_per_bin_h
+
+        elif self.mode == 'round':
+            raise NotImplementedError()
+
+        else:
+            raise ValueError('Incorrect quantization type.')
+
+        dequantized_boxes = torch.cat(
+            (dequantized_xmin, dequantized_ymin,
+             dequantized_xmax, dequantized_ymax), dim=-1
+        )
+
+        return dequantized_boxes
+
+
+class CoordinatesQuantizer(object):
+    """
+    Quantize coornidates (Nx2)
+    """
+
+    def __init__(self, mode, bins):
+        self.mode = mode
+        self.bins = bins
+
+    def quantize(self, coordinates: torch.Tensor, size):
+        bins_w, bins_h = self.bins  # Quantization bins.
+        size_w, size_h = size       # Original image size.
+        size_per_bin_w = size_w / bins_w
+        size_per_bin_h = size_h / bins_h
+        assert coordinates.shape[-1] == 2, 'coordinates should be shape (N, 2)'
+        x, y = coordinates.split(1, dim=-1)  # Shape: 4 * [N, 1].
+
+        if self.mode == 'floor':
+            quantized_x = (x / size_per_bin_w).floor().clamp(0, bins_w - 1)
+            quantized_y = (y / size_per_bin_h).floor().clamp(0, bins_h - 1)
+
+        elif self.mode == 'round':
+            raise NotImplementedError()
+
+        else:
+            raise ValueError('Incorrect quantization type.')
+
+        quantized_coordinates = torch.cat(
+            (quantized_x, quantized_y), dim=-1
+        ).int()
+
+        return quantized_coordinates
+
+    def dequantize(self, coordinates: torch.Tensor, size):
+        bins_w, bins_h = self.bins  # Quantization bins.
+        size_w, size_h = size       # Original image size.
+        size_per_bin_w = size_w / bins_w
+        size_per_bin_h = size_h / bins_h
+        assert coordinates.shape[-1] == 2, 'coordinates should be shape (N, 2)'
+        x, y = coordinates.split(1, dim=-1)  # Shape: 4 * [N, 1].
+
+        if self.mode == 'floor':
+            # Add 0.5 to use the center position of the bin as the coordinate.
+            dequantized_x = (x + 0.5) * size_per_bin_w
+            dequantized_y = (y + 0.5) * size_per_bin_h
+
+        elif self.mode == 'round':
+            raise NotImplementedError()
+
+        else:
+            raise ValueError('Incorrect quantization type.')
+
+        dequantized_coordinates = torch.cat(
+            (dequantized_x, dequantized_y), dim=-1
+        )
+
+        return dequantized_coordinates
+
+
+class Florence2PostProcesser(object):
+    """
+    Florence-2 post process for converting text prediction to various tasks results. 
+
+    Args:
+        config: A dict of configs.
+        tokenizer: A tokenizer for decoding text to spans.
+        sample config:
+            UNIFIED_POST_PROCESS:
+                # commom configs
+                NUM_BBOX_HEIGHT_BINS: 1000
+                NUM_BBOX_WIDTH_BINS: 1000
+                COORDINATES_HEIGHT_BINS: 1000
+                COORDINATES_WIDTH_BINS: 1000
+                # task specific configs, override the common configs
+                PRASE_TASKS:
+                    - TASK_NAME: 'video_dense_caption'
+                      PATTERN: 'r<time_(\d+)><time_(\d+)>([a-zA-Z0-9 ]+)'
+                      SCORE_MODE: 'avg_cat_name_scores'
+                      NUM_BINS: 100
+                    - TASK_NAME: 'od'
+                      PATTERN: 'r<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>([a-zA-Z0-9 ]+)'
+                      SCORE_MODE: 'avg_cat_name_scores'
+
+    Returns:
+        parsed_dict (dict): A dict of parsed results.
+    """
+    def __init__(
+        self,
+        tokenizer=None
+    ):
+        parse_tasks = []
+        parse_task_configs = {}
+        config = self._create_default_config()
+        for task in config['PARSE_TASKS']:
+            parse_tasks.append(task['TASK_NAME'])
+            parse_task_configs[task['TASK_NAME']] = task
+
+        self.config = config
+        self.parse_tasks = parse_tasks
+        self.parse_tasks_configs = parse_task_configs
+
+        self.tokenizer =  tokenizer
+        if self.tokenizer is not None:
+            self.all_special_tokens = set(self.tokenizer.all_special_tokens)
+
+        self.init_quantizers()
+        self.black_list_of_phrase_grounding = self._create_black_list_of_phrase_grounding()
+
+    def _create_black_list_of_phrase_grounding(self):
+        black_list = {}
+
+        if 'phrase_grounding' in self.parse_tasks and self.parse_tasks_configs['phrase_grounding']['FILTER_BY_BLACK_LIST']:
+            black_list =  set(
+                ['it', 'I', 'me', 'mine',
+                 'you', 'your', 'yours',
+                 'he', 'him', 'his',
+                 'she', 'her', 'hers',
+                 'they', 'them', 'their', 'theirs',
+                 'one', 'oneself',
+                 'we', 'us', 'our', 'ours',
+                 'you', 'your', 'yours',
+                 'they', 'them', 'their', 'theirs',
+                 'mine', 'yours', 'his', 'hers', 'its',
+                 'ours', 'yours', 'theirs',
+                 'myself', 'yourself', 'himself', 'herself', 'itself',
+                 'ourselves', 'yourselves', 'themselves',
+                 'this', 'that',
+                 'these', 'those',
+                 'who', 'whom', 'whose', 'which', 'what',
+                 'who', 'whom', 'whose', 'which', 'that',
+                 'all', 'another', 'any', 'anybody', 'anyone', 'anything',
+                 'each', 'everybody', 'everyone', 'everything',
+                 'few', 'many', 'nobody', 'none', 'one', 'several',
+                 'some', 'somebody', 'someone', 'something',
+                 'each other', 'one another',
+                 'myself', 'yourself', 'himself', 'herself', 'itself',
+                 'ourselves', 'yourselves', 'themselves',
+                 'the image', 'image', 'images', 'the', 'a', 'an', 'a group',
+                 'other objects', 'lots', 'a set',
+                 ]
+            )
+
+        return black_list
+    
+    def _create_default_config(self):
+        config = {
+            'NUM_BBOX_HEIGHT_BINS': 1000,
+            'NUM_BBOX_WIDTH_BINS': 1000,
+            'BOX_QUANTIZATION_MODE': 'floor',
+            'COORDINATES_HEIGHT_BINS': 1000,
+            'COORDINATES_WIDTH_BINS': 1000,
+            'COORDINATES_QUANTIZATION_MODE': 'floor',
+            'PARSE_TASKS': [
+                {
+                    'TASK_NAME': 'od',
+                    'PATTERN': r'([a-zA-Z0-9 ]+)<loc_(\\d+)><loc_(\\d+)><loc_(\\d+)><loc_(\\d+)>'
+                },
+                {
+                    'TASK_NAME': 'ocr',
+                    'PATTERN':  r'(.+?)<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>',
+                    'AREA_THRESHOLD': 0.01
+                },
+                {
+                    'TASK_NAME': 'phrase_grounding',
+                    'FILTER_BY_BLACK_LIST': True
+                },
+                {
+                    'TASK_NAME': 'pure_text',
+                },
+                {
+                    'TASK_NAME': 'description_with_bboxes',
+                },
+                {
+                    'TASK_NAME': 'description_with_polygons',
+                },
+                {
+                    'TASK_NAME': 'polygons',
+                },
+                {
+                    'TASK_NAME': 'bboxes',
+                },
+                {
+                    'TASK_NAME': 'description_with_bboxes_or_polygons',
+                }
+            ]
+        }
+
+        return config
+
+    def init_quantizers(self):
+        # we have box_quantizer (od, grounding) and coordinates_quantizer (ocr, referring_segmentation)
+        num_bbox_height_bins = self.config.get('NUM_BBOX_HEIGHT_BINS', 1000)
+        num_bbox_width_bins = self.config.get('NUM_BBOX_WIDTH_BINS', 1000)
+        box_quantization_mode = self.config.get('BOX_QUANTIZATION_MODE', 'floor')
+        self.box_quantizer = BoxQuantizer(
+            box_quantization_mode,
+            (num_bbox_width_bins, num_bbox_height_bins),
+        )
+        
+        num_bbox_height_bins = self.config['COORDINATES_HEIGHT_BINS'] if 'COORDINATES_HEIGHT_BINS' in self.config else self.config.get('NUM_BBOX_HEIGHT_BINS', 1000)
+        num_bbox_width_bins = self.config['COORDINATES_WIDTH_BINS'] if 'COORDINATES_WIDTH_BINS' in self.config else self.config.get('NUM_BBOX_WIDTH_BINS', 1000)
+        box_quantization_mode = self.config.get('COORDINATES_QUANTIZATION_MODE') if 'COORDINATES_QUANTIZATION_MODE' in self.config else self.config.get('BOX_QUANTIZATION_MODE', 'floor')
+        self.coordinates_quantizer = CoordinatesQuantizer(
+            box_quantization_mode,
+            (num_bbox_width_bins, num_bbox_height_bins),
+        )
+
+    def decode_with_spans(self, tokenizer, token_ids):
+        filtered_tokens = tokenizer.convert_ids_to_tokens(
+            token_ids, skip_special_tokens=False)
+        assert len(filtered_tokens) == len(token_ids)
+
+        # To avoid mixing byte-level and unicode for byte-level BPT
+        # we need to build string separately for added tokens and byte-level tokens
+        # cf. https://github.com/huggingface/transformers/issues/1133
+        sub_texts = []
+        for token in filtered_tokens:
+            if token in self.all_special_tokens:
+                sub_texts.append(token)
+            else:
+                if isinstance(tokenizer, (BartTokenizer, BartTokenizerFast)):
+                    sub_text = tokenizer.convert_tokens_to_string([token])
+                elif isinstance(tokenizer, (T5Tokenizer, T5TokenizerFast)):
+                    # Ref: https://github.com/google/sentencepiece#whitespace-is-treated-as-a-basic-symbol
+                    # Note: Do not strip sub_text as it may have functional whitespace
+                    sub_text = token.replace('▁', ' ')
+                else:
+                    raise ValueError(f'type {type(tokenizer)} not supported')
+                sub_texts.append(sub_text)
+
+        text = ''
+        spans = []
+        for sub_text in sub_texts:
+            span = (len(text), len(text) + len(sub_text))  # [start index, end index).
+            text += sub_text
+            spans.append(span)
+
+        # Text format:
+        # 1. T5Tokenizer/T5TokenizerFast: 
+        #      "<loc_1><loc_2><loc_3><loc_4> transplanting dog<loc_1><loc_2><loc_3><loc_4> cat</s>"
+        #    Equivalent to t5_tokenizer.decode(input_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False, spaces_between_special_tokens=False)
+        # 2. BartTokenizer (need to double check):
+        #      "<s><loc_1><loc_2><loc_3><loc_4>transplanting dog<loc_1><loc_2><loc_3><loc_4>cat</s>"
+        #    Equivalent to bart_tokenizer.decode(input_ids, skip_special_tokens=False, clean_up_tokenization_spaces=False, spaces_between_special_tokens=False)
+        return text, spans
+
+    def parse_od_from_text_and_spans(
+        self,
+        text,
+        pattern,
+        image_size,
+        phrase_centric=False
+    ):
+        parsed = list(re.finditer(pattern, text))
+
+        instances = []
+        for i in range(len(parsed)):
+            # Prepare instance.
+            instance = {}
+
+            if phrase_centric:
+                bbox_bins = [int(parsed[i].group(j)) for j in range(2, 6)]
+            else:
+                bbox_bins = [int(parsed[i].group(j)) for j in range(1, 5)]
+            instance['bbox'] = self.box_quantizer.dequantize(
+                boxes=torch.tensor(bbox_bins),
+                size=image_size
+            ).tolist()  
+
+            if phrase_centric:
+                instance['cat_name'] = parsed[i].group(1).lower().strip()
+            else:
+                instance['cat_name'] = parsed[i].group(5).lower().strip()
+            instances.append(instance)
+
+        return instances
+
+    def parse_ocr_from_text_and_spans(self, 
+                                    text, 
+                                     pattern, 
+                                     image_size,
+                                     area_threshold=-1.0,
+        ):
+        bboxes = []
+        labels = []
+        text = text.replace('<s>', '')
+        # ocr with regions
+        parsed = re.findall(pattern, text)
+        instances = []
+        image_width, image_height = image_size
+
+        for ocr_line in parsed:
+            ocr_content = ocr_line[0]
+            quad_box = ocr_line[1:]
+            quad_box = [int(i) for i in quad_box]
+            quad_box = self.coordinates_quantizer.dequantize(
+                torch.tensor(np.array(quad_box).reshape(-1, 2)),
+                size=image_size
+            ).reshape(-1).tolist()
+
+            if area_threshold > 0:
+                x_coords = [i for i in quad_box[0::2]]
+                y_coords = [i for i in quad_box[1::2]]
+
+                # apply the Shoelace formula
+                area = 0.5 * abs(sum(x_coords[i] * y_coords[i + 1] - x_coords[i + 1] * y_coords[i] for i in range(4 - 1)))
+
+                if area < (image_width * image_height) * area_threshold:
+                    continue
+
+            bboxes.append(quad_box)
+            labels.append(ocr_content)
+            instances.append({
+                'quad_box': quad_box,
+                'text': ocr_content,
+            })
+        return instances
+
+    def parse_phrase_grounding_from_text_and_spans(self, text, pattern, image_size):
+        # ignore <s> </s> and <pad>
+        cur_span = 0
+        if text.startswith('<s>'):   
+            cur_span += 3
+
+        text = text.replace('<s>', '')
+        text = text.replace('</s>', '')
+        text = text.replace('<pad>', '')
+
+        pattern = r"([^<]+(?:<loc_\d+>){4,})"
+        phrases = re.findall(pattern, text)
+    
+        # pattern should be text pattern and od pattern
+        pattern = r'^\s*(.*?)(?=<od>|</od>|<box>|</box>|<bbox>|</bbox>|<loc_)'
+        box_pattern = r'<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>'
+
+        instances = []
+        for pharse_text in phrases:
+            phrase_text_strip = pharse_text.replace('<ground>', '', 1)
+            phrase_text_strip = pharse_text.replace('<obj>', '', 1)
+
+            if phrase_text_strip == '':
+                cur_span += len(pharse_text)
+                continue
+
+            # Prepare instance.
+            instance = {}
+
+            # parse phrase, get string 
+            phrase = re.search(pattern, phrase_text_strip)
+            if phrase is None:
+                cur_span += len(pharse_text)
+                continue
+
+            # parse bboxes by box_pattern
+            bboxes_parsed = list(re.finditer(box_pattern, pharse_text))
+            if len(bboxes_parsed) == 0:
+                cur_span += len(pharse_text)
+                continue
+
+            phrase = phrase.group()
+            # remove leading and trailing spaces
+            phrase = phrase.strip()
+
+            if phrase in self.black_list_of_phrase_grounding:
+                cur_span += len(pharse_text)
+                continue
+
+            # a list of list 
+            bbox_bins = [[int(_bboxes_parsed.group(j)) for j in range(1, 5)] for _bboxes_parsed in bboxes_parsed]
+            instance['bbox'] = self.box_quantizer.dequantize(
+                boxes=torch.tensor(bbox_bins),
+                size=image_size
+            ).tolist()  
+
+            # exclude non-ascii characters
+            phrase = phrase.encode('ascii',errors='ignore').decode('ascii')
+            instance['cat_name'] = phrase
+
+            instances.append(instance)
+
+        return instances
+
+    def parse_description_with_bboxes_from_text_and_spans(self, text, pattern, image_size, allow_empty_phrase=False):
+        # temporary parse solution, split by '.'
+        # ignore <s> </s> and <pad>
+
+        text = text.replace('<s>', '')
+        text = text.replace('</s>', '')
+        text = text.replace('<pad>', '')
+
+        if allow_empty_phrase:
+            pattern = rf"(?:(?:<loc_\d+>){{4,}})"
+        else:
+            pattern = r"([^<]+(?:<loc_\d+>){4,})"
+        phrases = re.findall(pattern, text)
+    
+        # pattern should be text pattern and od pattern
+        pattern = r'^\s*(.*?)(?=<od>|</od>|<box>|</box>|<bbox>|</bbox>|<loc_)'
+        box_pattern = r'<loc_(\d+)><loc_(\d+)><loc_(\d+)><loc_(\d+)>'
+
+        instances = []
+        for pharse_text in phrases:
+            phrase_text_strip = pharse_text.replace('<ground>', '', 1)
+            phrase_text_strip = pharse_text.replace('<obj>', '', 1)
+
+            if phrase_text_strip == '' and not allow_empty_phrase:
+                continue
+
+            # parse phrase, get string 
+            phrase = re.search(pattern, phrase_text_strip)
+            if phrase is None:
+                continue
+
+            phrase = phrase.group()
+            # remove leading and trailing spaces
+            phrase = phrase.strip()
+
+            # parse bboxes by box_pattern
+            bboxes_parsed = list(re.finditer(box_pattern, pharse_text))
+            if len(bboxes_parsed) == 0:
+                continue
+
+            # a list of list 
+            bbox_bins = [[int(_bboxes_parsed.group(j)) for j in range(1, 5)] for _bboxes_parsed in bboxes_parsed]
+
+            bboxes = self.box_quantizer.dequantize(
+                boxes=torch.tensor(bbox_bins),
+                size=image_size
+            ).tolist()  
+
+            phrase = phrase.encode('ascii',errors='ignore').decode('ascii')
+            for _bboxes in bboxes:
+                # Prepare instance.
+                instance = {}
+                instance['bbox'] = _bboxes
+                # exclude non-ascii characters
+                instance['cat_name'] = phrase
+                instances.append(instance)
+
+        return instances
+
+    def parse_description_with_polygons_from_text_and_spans(self, text, pattern, image_size, 
+                                                            allow_empty_phrase=False,
+                                                            polygon_sep_token='<sep>',
+                                                            polygon_start_token='<poly>',
+                                                            polygon_end_token='</poly>',
+                                                            with_box_at_start=False,
+                                                            ):
+        
+        # ref_seg format: '<expression><x1><y1><x2><y2><><><sep><><><><>'
+        # ignore <s> </s> and <pad>
+
+        text = text.replace('<s>', '')
+        text = text.replace('</s>', '')
+        text = text.replace('<pad>', '')
+
+        if allow_empty_phrase:
+            pattern = rf"(?:(?:<loc_\d+>|{re.escape(polygon_sep_token)}|{re.escape(polygon_start_token)}|{re.escape(polygon_end_token)}){{4,}})"
+        else:
+            # [^<]+: This part matches one or more characters that are not the < symbol. 
+            # The ^ inside the square brackets [] is a negation, meaning it matches anything except <.
+            #
+            pattern = rf"([^<]+(?:<loc_\d+>|{re.escape(polygon_sep_token)}|{re.escape(polygon_start_token)}|{re.escape(polygon_end_token)}){{4,}})"
+        phrases = re.findall(pattern, text)
+
+        phrase_string_pattern = r'^\s*(.*?)(?=<od>|</od>|<box>|</box>|<bbox>|</bbox>|<loc_|<poly>)'
+        box_pattern =  rf'((?:<loc_\d+>)+)(?:{re.escape(polygon_sep_token)}|$)'
+
+        # one polygons instance is separated by polygon_start_token and polygon_end_token
+        polygons_instance_pattern = rf'{re.escape(polygon_start_token)}(.*?){re.escape(polygon_end_token)}'
+    
+        instances = []
+        for phrase_text in phrases:
+
+            # exclude loc_\d+>
+            # need to get span if want to include category score
+            phrase_text_strip = re.sub(r'^loc_\d+>', '', phrase_text, count=1)
+
+            # phrase = phrase.replace('<poly>', '')
+            # phrase = phrase.replace('poly>', '')
+
+            if phrase_text_strip == '' and not allow_empty_phrase:
+                continue
+
+
+            # parse phrase, get string 
+            phrase = re.search(phrase_string_pattern, phrase_text_strip)
+            if phrase is None:
+                continue
+            phrase = phrase.group()
+            # remove leading and trailing spaces
+            phrase = phrase.strip()
+
+            # parse bboxes by box_pattern
+
+            # split by polygon_start_token and polygon_end_token first using polygons_instance_pattern
+            if polygon_start_token in phrase_text and polygon_end_token in phrase_text:
+                polygons_instances_parsed = list(re.finditer(polygons_instance_pattern, phrase_text))
+            else:
+                polygons_instances_parsed = [phrase_text]
+
+            for _polygons_instances_parsed in polygons_instances_parsed:
+                # Prepare instance.
+                instance = {}
+
+                # polygons_parsed= list(re.finditer(box_pattern, phrase_text))
+                if isinstance(_polygons_instances_parsed, str): 
+                    polygons_parsed= list(re.finditer(box_pattern, _polygons_instances_parsed))
+                else:
+                    polygons_parsed= list(re.finditer(box_pattern, _polygons_instances_parsed.group(1)))
+                if len(polygons_parsed) == 0:
+                    continue
+
+                # a list of list (polygon)
+                bbox = []
+                polygons = []
+                for _polygon_parsed in polygons_parsed:
+                    # group 1: whole <loc_\d+>...</loc_\d+>
+                    _polygon = _polygon_parsed.group(1)
+                    # parse into list of int
+                    _polygon = [int(_loc_parsed.group(1)) for _loc_parsed in re.finditer(r'<loc_(\d+)>', _polygon)]
+                    if with_box_at_start and len(bbox) == 0:
+                        if len(_polygon) > 4:
+                            # no valid bbox prediction
+                            bbox = _polygon[:4]
+                            _polygon = _polygon[4:]
+                        else:
+                            bbox = [0, 0, 0, 0]
+                    # abandon last element if is not paired 
+                    if len(_polygon) % 2 == 1:
+                        _polygon = _polygon[:-1]
+                    
+                    # reshape into (n, 2)
+                    _polygon = self.coordinates_quantizer.dequantize(
+                        torch.tensor(np.array(_polygon).reshape(-1, 2)),
+                        size=image_size
+                    ).reshape(-1).tolist()
+                    # reshape back
+                    polygons.append(_polygon)
+
+                instance['cat_name'] = phrase
+                instance['polygons'] = polygons
+                if len(bbox) != 0:
+                    instance['bbox'] = self.box_quantizer.dequantize(
+                        boxes=torch.tensor([bbox]),
+                        size=image_size
+                    ).tolist()[0]  
+
+                instances.append(instance)
+
+        return instances
+
+    def __call__(
+        self,
+        text=None,
+        image_size=None,
+        parse_tasks=None,
+    ):
+        """
+        Args:
+            text: model outputs
+            image_size: (width, height)
+            parse_tasks: a list of tasks to parse, if None, parse all tasks.
+
+        """
+        if parse_tasks is not None:
+            if isinstance(parse_tasks, str):
+                parse_tasks = [parse_tasks]
+            for _parse_task in parse_tasks:
+                assert _parse_task in self.parse_tasks, f'parse task {_parse_task} not supported'
+        
+        # sequence or text should be provided 
+        assert text is not None, 'text should be provided'
+
+        parsed_dict = {
+            'text': text
+        }
+
+        for task in self.parse_tasks:
+            if parse_tasks is not None and task not in parse_tasks:
+                continue
+
+            pattern = self.parse_tasks_configs[task].get('PATTERN', None)
+
+            if task == 'ocr':
+                instances = self.parse_ocr_from_text_and_spans(
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                    area_threshold=self.parse_tasks_configs[task].get('AREA_THRESHOLD', 0.01),
+                )
+                parsed_dict['ocr'] = instances
+            elif task == 'phrase_grounding':
+                instances = self.parse_phrase_grounding_from_text_and_spans( 
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                )
+                parsed_dict['phrase_grounding'] = instances
+            elif task == 'pure_text':
+                parsed_dict['pure_text'] = text 
+            elif task == 'description_with_bboxes':
+                instances = self.parse_description_with_bboxes_from_text_and_spans( 
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                )
+                parsed_dict['description_with_bboxes'] = instances
+            elif task == 'description_with_polygons':
+                instances = self.parse_description_with_polygons_from_text_and_spans( 
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                )
+                parsed_dict['description_with_polygons'] = instances
+            elif task == 'polygons':
+                instances = self.parse_description_with_polygons_from_text_and_spans( 
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                    allow_empty_phrase=True,
+                )
+                parsed_dict['polygons'] = instances
+            elif task == 'bboxes':
+                instances = self.parse_description_with_bboxes_from_text_and_spans( 
+                    text,
+                    pattern=pattern,
+                    image_size=image_size,
+                    allow_empty_phrase=True,
+                )
+                parsed_dict['bboxes'] = instances
+            elif task == 'description_with_bboxes_or_polygons':
+                if '<poly>' in text:
+                    # only support either polygons or bboxes, not both at the same time
+                    instances = self.parse_description_with_polygons_from_text_and_spans( 
+                        text,
+                        pattern=pattern,
+                        image_size=image_size,
+                    )
+                else:
+                    instances = self.parse_description_with_bboxes_from_text_and_spans( 
+                        text,
+                        pattern=pattern,
+                        image_size=image_size,
+                    )
+                parsed_dict['description_with_bboxes_or_polygons'] = instances
+            else:
+                raise ValueError("task {} is not supported".format(task))
+
+        return parsed_dict

processor_config.json

@@ -0,0 +1,6 @@
+{
+  "auto_map": {
+    "AutoProcessor": "processing_florence2.Florence2Processor"
+  },
+  "processor_class": "Florence2Processor"
+}