Update files for compatibility with transformers

chat_template.jinja

@@ -0,0 +1,15 @@
+{%- for message in messages %}
+    {%- if message['content'] is string %}
+        {{- message['content'] }}
+    {%- else %}
+        {%- for content in message['content'] | selectattr('type', 'equalto', 'image') %}
+            {{- '<image>' }}
+        {%- endfor %}
+        {%- for content in message['content'] | selectattr('type', 'equalto', 'text') %}
+            {{- content['text'] }}
+        {%- endfor %}
+    {%- endif %}
+    {%- if message['role'] == 'assistant' %}
+        {{- eos_token }}
+    {%- endif %}
+{%- endfor %}

config.json

@@ -51,7 +51,7 @@
     "vocab_size": 129280,
     "sliding_window_size": 128
   },
-  "model_type": "unlimited-ocr",
+  "model_type": "unlimited_ocr",
   "projector_config": {
     "input_dim": 2048,
     "model_type": "mlp_projector",

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "no_repeat_ngram_size": 35,
+  "no_repeat_ngram_window_size": 128,
+  "max_new_tokens": 32768,
+  "transformers_version": "5.14.0.dev0",
+  "use_cache": true
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2bc48a7a110061ea58fff65d3169367eebe3aee371ca6968dc2219c1b2855fc6
+size 6672547120

modeling_unlimitedocr.py

@@ -154,7 +154,6 @@ def process_image_with_refs(image, ref_texts, output_path, image_prefix=''):
 
 
 
-
 def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
     best_ratio_diff = float('inf')
     best_ratio = (1, 1)
@@ -426,7 +425,7 @@ class TPSTextStreamer(TextStreamer):
 
 
 class UnlimitedOCRConfig(DeepseekV2Config):
-    model_type = "unlimited-ocr"
+    model_type = "unlimited_ocr"
 
 class UnlimitedOCRModel(DeepseekV2Model):
     config_class = UnlimitedOCRConfig
@@ -879,10 +878,6 @@ class UnlimitedOCRForCausalLM(DeepseekV2ForCausalLM):
                 # elif base_size == 640:
                 #     valid_img_tokens += int(100 * ratio)
                 
-
-
-
-                
                 images_list.append(image_transform(global_view).to(torch.bfloat16))
 
                 # global_view_tensor = image_transform(global_view).to(torch.bfloat16)
@@ -1296,4 +1291,4 @@ class UnlimitedOCRForCausalLM(DeepseekV2ForCausalLM):
             with open(f'{output_path}/result.md', 'w', encoding='utf-8') as afile:
                 afile.write(outputs)
 
-        return outputs, output_tokens
+        return outputs, output_tokens

preprocessor_config.json

@@ -0,0 +1,20 @@
+{
+  "image_processor_type": "UnlimitedOcrImageProcessor",
+  "do_convert_rgb": true,
+  "do_resize": true,
+  "resample": 3,
+  "do_rescale": true,
+  "rescale_factor": 0.00392156862745098,
+  "do_normalize": true,
+  "crop_to_patches": true,
+  "min_patches": 2,
+  "max_patches": 32,
+  "tile_size": 640,
+  "background_color": [127, 127, 127],
+  "size": {
+    "height": 1024,
+    "width": 1024
+  },
+  "image_mean": [0.5, 0.5, 0.5],
+  "image_std": [0.5, 0.5, 0.5]
+}

processor_config.json

@@ -21,8 +21,8 @@
   "image_token": "<image>",
   "mask_prompt": false,
   "normalize": true,
-  "pad_token": "<\uff5c\u2581pad\u2581\uff5c>",
+  "pad_token": "<｜▁pad▁｜>",
   "patch_size": 16,
-  "processor_class": "UnlimitedOCRHFProcessor",
+  "processor_class": "UnlimitedOcrProcessor",
   "sft_format": "unlimitedocr"
 }