update_model_init_fp16 (#53)

- update model init with float16 (93d5b247d87e5e0afeafe5b49c4a796aa5ba9e7c)
- update model init with float16 (bab1b3cbc340fac15205568422af11476c232442)

Co-authored-by: Haiping Wu <haipingwu@users.noreply.huggingface.co>

.gitattributes

@@ -34,4 +34,3 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 *bin filter=lfs diff=lfs merge=lfs -text
-pytorch_model.bin filter=lfs diff=lfs merge=lfs -text

config.json

@@ -7,11 +7,11 @@
     "AutoConfig": "configuration_florence2.Florence2Config",
     "AutoModelForCausalLM": "modeling_florence2.Florence2ForConditionalGeneration"
   },
-  "bos_token_id": 2,
+  "bos_token_id": 0,
-  "eos_token_id": 1,
+  "eos_token_id": 2,
   "ignore_index": -100,
   "model_type": "florence2",
-  "pad_token_id": 0,
+  "pad_token_id": 1,
   "projection_dim": 1024,
   "text_config": {
       "vocab_size": 51289,
@@ -79,7 +79,7 @@
     "image_feature_source": ["spatial_avg_pool", "temporal_avg_pool"]
   },
   "vocab_size": 51289,
-  "torch_dtype": "float32",
+  "torch_dtype": "float16",
   "transformers_version": "4.41.0.dev0",
   "is_encoder_decoder": true
 }

modeling_florence2.py

@@ -2240,6 +2240,10 @@ class Florence2Seq2SeqLMOutput(ModelOutput):
     decoding.
 
     Args:
+        loss (`torch.FloatTensor` of shape `(1,)`, *optional*, returned when `labels` is provided):
+            Language modeling loss.
+        logits (`torch.FloatTensor` of shape `(batch_size, sequence_length, config.vocab_size)`):
+            Prediction scores of the language modeling head (scores for each vocabulary token before SoftMax).
         last_hidden_state (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`):
             Sequence of hidden-states at the output of the last layer of the decoder of the model.
 
@@ -2288,7 +2292,8 @@ class Florence2Seq2SeqLMOutput(ModelOutput):
 
             image_hidden_states of the model produced by the vision encoder
     """
-
+    loss: Optional[torch.FloatTensor] = None
+    logits: torch.FloatTensor = None
     last_hidden_state: torch.FloatTensor = None
     past_key_values: Optional[Tuple[Tuple[torch.FloatTensor]]] = None
     decoder_hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
@@ -2297,6 +2302,7 @@ class Florence2Seq2SeqLMOutput(ModelOutput):
     encoder_last_hidden_state: Optional[torch.FloatTensor] = None
     encoder_hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
     encoder_attentions: Optional[Tuple[torch.FloatTensor, ...]] = None
+    image_hidden_states: Optional[Tuple[torch.FloatTensor, ...]] = None
 
 
 FLORENCE2_START_DOCSTRING = r"""
@@ -2527,7 +2533,6 @@ class Florence2ForConditionalGeneration(Florence2PreTrainedModel):
     def __init__(self, config: Florence2Config):
         super().__init__(config)
         assert config.vision_config.model_type == 'davit', 'only DaViT is supported for now'
-        del config.vision_config.model_type
         self.vision_tower = DaViT.from_config(config=config.vision_config)
         # remove unused layers 
         del self.vision_tower.head
@@ -2731,6 +2736,7 @@ class Florence2ForConditionalGeneration(Florence2PreTrainedModel):
                 image_features = self._encode_image(pixel_values)
                 inputs_embeds, attention_mask = self._merge_input_ids_with_image_features(image_features, inputs_embeds)
 
+        if inputs_embeds is not None:
             attention_mask = attention_mask.to(inputs_embeds.dtype)
         outputs = self.language_model(
             attention_mask=attention_mask,

processing_florence2.py

@@ -596,7 +596,7 @@ class Florence2PostProcesser(object):
                 {
                     '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
+                    'AREA_THRESHOLD': 0.00
                 },
                 {
                     'TASK_NAME': 'phrase_grounding',
@@ -1025,7 +1025,7 @@ class Florence2PostProcesser(object):
                     text,
                     pattern=pattern,
                     image_size=image_size,
-                    area_threshold=self.parse_tasks_configs[task].get('AREA_THRESHOLD', 0.01),
+                    area_threshold=self.parse_tasks_configs[task].get('AREA_THRESHOLD', 0.0),
                 )
                 parsed_dict['ocr'] = instances
             elif task == 'phrase_grounding':