using gemini flash

.gitignore

@@ -1,2 +1,4 @@
 .venv
-.DS_Store
+.DS_Store
+services/__pycache__
+.env

.gradio/certificate.pem

@@ -0,0 +1,31 @@
+-----BEGIN CERTIFICATE-----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=
+-----END CERTIFICATE-----

app.py

@@ -1,10 +1,9 @@
 import gradio as gr
 from services.ocr import upload_file, ocr, ocr_languages
-from services.interleave import interleave
+from services.interleave import interleave_text
 from services.translation import translation_languages, translation
-from services.transliteration import transliterate_languages, transliterate
+from services.transliteration import transliterate_languages, indic_transliteration
 
-img_path="bmc/qr-code.png"
 
 html_header = f"""
 <div style="display:flex; align-items:center; justify-content:space-between;">
@@ -23,16 +22,15 @@ with gr.Blocks(title="LMX", analytics_enabled=True) as interface:
         gr.HTML(html_header)
         gr.Button(variant="primary", scale=1, min_width=5, value="☕️ Buy me a coffee", size="lg", link_target="_blank", link="https://buymeacoffee.com/jobvargh")
         gr.Button(variant="primary", scale=1, min_width=5, value="Github", size="lg", link_target="_blank", link="https://github.com/jobinv12/LMX")
-
     
     with gr.Tab("OCR"):
         with gr.Row():
             with gr.Column():
-                ocr_path = gr.Textbox(interactive=False, visible=False)
-                ocr_languages = gr.Dropdown(label="Source language", choices=ocr_languages, filterable=True, info="Select source language")
+                ocr_image_url = gr.Textbox(interactive=False, visible=False)
+                ocr_languages = gr.Dropdown(label="Source language", choices=ocr_languages, filterable=True, info="Select source language", visible=False)
                 ocr_upload_file = gr.UploadButton(label="Upload a Image", file_types=[".png",".jpeg",".webp",".jpg"], file_count="single")
                 ocr_file = gr.Image(height=250, width=700, type="filepath", show_label=False, buttons=[], container=False)
-                ocr_upload_file.upload(upload_file, inputs=ocr_upload_file, outputs=[ocr_file, ocr_path])
+                ocr_upload_file.upload(upload_file, inputs=ocr_upload_file, outputs=[ocr_file, ocr_image_url])
 
             with gr.Column():
                 ocr_output = gr.TextArea(label="Output", lines=15 ,interactive=False)
@@ -41,7 +39,7 @@ with gr.Blocks(title="LMX", analytics_enabled=True) as interface:
             clear_btn = gr.ClearButton(value="Clear", variant="secondary")
             ocr_submit_btn = gr.Button(value="Run", variant="huggingface")
 
-        ocr_submit_btn.click(fn=ocr, inputs=[ocr_path, ocr_languages], outputs=ocr_output)
+        ocr_submit_btn.click(fn=ocr, inputs=[ocr_image_url, ocr_languages], outputs=ocr_output)
 
     with gr.Tab("Transliteration"):
         with gr.Row():
@@ -57,7 +55,7 @@ with gr.Blocks(title="LMX", analytics_enabled=True) as interface:
             clear_btn = gr.ClearButton(value="Clear", variant="secondary")
             transliteration_submit_btn = gr.Button(value="Run", variant="huggingface")
         
-        transliteration_submit_btn.click(fn=transliterate, inputs=[transliteration_src_lang, transliteration_trgt_lang, transliteration_input_area], outputs=transliteration_output_area)
+        transliteration_submit_btn.click(fn=indic_transliteration, inputs=[transliteration_src_lang, transliteration_trgt_lang, transliteration_input_area], outputs=transliteration_output_area)
     
     with gr.Tab("Translation"):
         with gr.Row():
@@ -74,7 +72,6 @@ with gr.Blocks(title="LMX", analytics_enabled=True) as interface:
 
         translation_submit_btn.click(fn=translation, inputs=[translation_src_lang, translation_trgt_lang, translation_input_area], outputs=translation_output_area) 
 
-
     with gr.Tab("Interleave"):
         with gr.Row():
             interleave_input_area = gr.TextArea(label="First Input", type="text")
@@ -86,7 +83,7 @@ with gr.Blocks(title="LMX", analytics_enabled=True) as interface:
 
         with gr.Row():
             interleave_output_area = gr.TextArea(label="Output", interactive=False, type="text")
-        interleave_submit_btn.click(fn=interleave, inputs=[interleave_input_area, interleave_input2_area], outputs=[interleave_output_area])
+        interleave_submit_btn.click(fn=interleave_text, inputs=[interleave_input_area, interleave_input2_area], outputs=[interleave_output_area])
     
     with gr.Row():
         gr.HTML(html_template=html_footer)

config.py

@@ -0,0 +1,17 @@
+from dotenv import load_dotenv
+from langchain_google_genai import ChatGoogleGenerativeAI
+import os
+
+load_dotenv()
+
+model = ChatGoogleGenerativeAI(
+    model = "gemini-2.5-flash",
+    temperature = 1.0,
+    max_tokens = None,
+    timeout = 10,
+    max_retries=1
+)
+
+HF_BUCKET_NAME = os.environ.get("HF_BUCKET_NAME")
+HF_BUCKET_URL = os.environ.get("HF_BUCKET_URL")
+HF_BUCKET_TOKEN = os.environ.get("HF_BUCKET_TOKEN")

requirements.txt

@@ -1,28 +1,21 @@
 aiofiles==24.1.0
-aiohappyeyeballs==2.6.1
-aiohttp==3.13.3
-aiosignal==1.4.0
-alabaster==1.0.0
 annotated-doc==0.0.4
 annotated-types==0.7.0
-anyio==4.12.1
-attrs==26.1.0
-babel==2.18.0
+anyio==4.13.0
 brotli==1.2.0
 certifi==2026.2.25
 cffi==2.0.0
 charset-normalizer==3.4.6
 click==8.3.1
-colorama==0.4.6
-cython==3.2.4
-datasets==4.8.2
-dill==0.4.1
-docutils==0.22.4
-fastapi==0.135.1
+cryptography==46.0.5
+distro==1.9.0
+fastapi==0.135.2
 ffmpy==1.0.0
 filelock==3.25.2
-frozenlist==1.8.0
+filetype==1.2.0
 fsspec==2026.2.0
+google-auth==2.49.1
+google-genai==1.68.0
 gradio==6.9.0
 gradio-client==2.3.0
 groovy==0.1.2
@@ -30,82 +23,52 @@ h11==0.16.0
 hf-xet==1.4.2
 httpcore==1.0.9
 httpx==0.28.1
-huggingface-hub==0.36.2
+huggingface-hub==1.7.2
 idna==3.11
-imagesize==2.0.0
-indic-nlp-library-itt==0.1.1
-indictranstoolkit==1.1.1
 jinja2==3.1.6
-joblib==1.5.3
-lxml==6.0.2
+jsonpatch==1.33
+jsonpointer==3.1.1
+langchain-core==1.2.21
+langchain-google-genai==4.2.1
+langsmith==0.7.22
 markdown-it-py==4.0.0
 markupsafe==3.0.3
 mdurl==0.1.2
-mlx==0.31.1
-mlx-lm==0.31.1
-mlx-vlm==0.4.0
-morfessor==2.0.6
-mpmath==1.3.0
-multidict==6.7.1
-multiprocess==0.70.19
-networkx==3.6.1
 numpy==2.4.3
-opencv-python==4.13.0.92
 orjson==3.11.7
 packaging==26.0
 pandas==3.0.1
 pillow==12.1.1
-portalocker==3.2.0
-propcache==0.4.1
-protobuf==7.34.0
-pyarrow==23.0.1
+pyasn1==0.6.3
+pyasn1-modules==0.4.2
 pycparser==3.0
 pydantic==2.12.5
 pydantic-core==2.41.5
 pydub==0.25.1
 pygments==2.19.2
 python-dateutil==2.9.0.post0
+python-dotenv==1.2.2
 python-multipart==0.0.22
 pytz==2026.1.post1
 pyyaml==6.0.3
-regex==2026.2.28
 requests==2.32.5
+requests-toolbelt==1.0.0
 rich==14.3.3
-roman-numerals==4.1.0
-sacrebleu==2.6.0
-sacremoses==0.1.1
 safehttpx==0.1.7
-safetensors==0.7.0
 semantic-version==2.10.0
-sentencepiece==0.2.1
-setuptools==82.0.1
 shellingham==1.5.4
 six==1.17.0
-snowballstemmer==3.0.1
-soundfile==0.13.1
-sphinx==9.1.0
-sphinx-argparse==0.5.2
-sphinx-rtd-theme==3.1.0
-sphinxcontrib-applehelp==2.0.0
-sphinxcontrib-devhelp==2.0.0
-sphinxcontrib-htmlhelp==2.1.0
-sphinxcontrib-jquery==4.1
-sphinxcontrib-jsmath==1.0.1
-sphinxcontrib-qthelp==2.0.0
-sphinxcontrib-serializinghtml==2.0.0
+sniffio==1.3.1
 starlette==0.52.1
-sympy==1.14.0
-tabulate==0.10.0
-tokenizers==0.21.4
+tenacity==9.1.4
 tomlkit==0.13.3
-torch==2.10.0
-torchvision==0.25.0
 tqdm==4.67.3
-transformers==4.53.2
 typer==0.24.1
 typing-extensions==4.15.0
 typing-inspection==0.4.2
 urllib3==2.6.3
+uuid-utils==0.14.1
 uvicorn==0.42.0
+websockets==16.0
 xxhash==3.6.0
-yarl==1.23.0
+zstandard==0.25.0

services/interleave.py

@@ -1,11 +1,29 @@
-def interleave(first_input_text:str, second_input_text:str) -> str:
-    first_input = [
-        line.strip() for line in first_input_text.splitlines() if line.strip()
-    ]
-    second_input = [
-        line.strip() for line in second_input_text.splitlines() if line.strip()
+from config import model
+
+# def interleave_text(first_input_text:str, second_input_text:str) -> str:
+#     first_input = [
+#         line.strip() for line in first_input_text.splitlines() if line.strip()
+#     ]
+#     second_input = [
+#         line.strip() for line in second_input_text.splitlines() if line.strip()
+#     ]
+
+#     result = "\n\n".join(f"{x}\n{y}" for x, y in zip(first_input, second_input))
+
+#     return result
+
+
+def interleave_text(first_input_text:str, second_input_text:str) -> str:
+
+    messages = [
+        (
+            "system", "You are a helpful assistant that interleave text between to inputs. Interleave the user inputs."
+        ),
+        (
+            "human", f"{first_input_text} {second_input_text}"
+        )
     ]
 
-    result = "\n\n".join(f"{x}\n{y}" for x, y in zip(first_input, second_input))
+    response = model.invoke(messages)
 
-    return result
+    return response.content

services/ocr.py

@@ -1,10 +1,9 @@
 import os
 from datetime import datetime
 import shutil
-import mlx.core as mx
-from mlx_vlm import load, generate
-from mlx_vlm.prompt_utils import apply_chat_template
-from mlx_vlm.utils import load_config
+from huggingface_hub import batch_bucket_files
+from langchain_core.messages import HumanMessage
+from config import model, HF_BUCKET_NAME, HF_BUCKET_URL, HF_BUCKET_TOKEN
 
 ocr_languages = [
     "English", "Afrikaans", "Amharic", "Arabic", "Assamese", "Azerbaijani", 
@@ -34,16 +33,6 @@ ocr_languages = [
     "Vietnamese", "Yiddish", "Yoruba"
 ]
 
-model_id = "mlx-community/olmOCR-2-7B-1025-8bit"
-
-try:
-    # Load the model
-    model, processor = load(model_id, trust_remote_code=True)
-    config = load_config(model_id)
-    print("olmocr2 model loaded successfully")
-except Exception as e:
-    print(f"Failed to load deepseekocr2 model {e}")
-
 def upload_file(file):
 
     if file is None:
@@ -61,14 +50,32 @@ def upload_file(file):
 
     shutil.copy(file, dest_path)
 
-    return dest_path, dest_path
-    
-def ocr(image_path:str, language:str) -> str:
+    store_file = batch_bucket_files(
+        bucket_id=HF_BUCKET_NAME,
+        add=[
+            (dest_path, new_filename)
+        ],
+        token=HF_BUCKET_TOKEN
+    )
 
-    prompt = f"Convert the {language} document to markdown."
+    file_url = f"{HF_BUCKET_URL}/{new_filename}"
+
+    return dest_path, file_url
+    
+def ocr(image_url:str, language:str) -> str:
     
-    formatted_output = apply_chat_template(processor, config, prompt, num_images=1)
+    messages = [
+        HumanMessage(
+       content=[
+           {"type": "text", "text": "Extract the content of this document into well-formatted markdown. No Intro. Extract the user file. "},
+           {
+               "type": "image_url",
+               "image_url": image_url
+           }
+       ]
+    )
+    ]
 
-    output = generate(model, processor, formatted_output, [image_path], verbose=False, max_tokens=2000)
+    response = model.invoke(messages)
 
-    return output.text
+    return response.content

services/translation.py

@@ -1,5 +1,5 @@
-import torch
-from transformers import AutoModelForImageTextToText, AutoProcessor
+from config import model
+
 
 translation_languages: list = ['English', 'Abkhaz', 'Acehnese', 'Acholi', 
                          'Afar', 'Afrikaans', 'Albanian', 'Alur', 
@@ -310,14 +310,6 @@ languages_iso_codes: dict = {
     "Zulu": "zu"
 }
 
-model_id = "google/translategemma-4b-it"
-
-try:
-   processor = AutoProcessor.from_pretrained(model_id)
-   model = AutoModelForImageTextToText.from_pretrained(model_id, device_map="auto")
-   print("Translation model loaded successfully")
-except Exception as e:
-    print(f"Failed to load translation model {e}")
 
 def translation(src_lang:str, trgt_lang:str, text:str) -> str:
 
@@ -325,28 +317,16 @@ def translation(src_lang:str, trgt_lang:str, text:str) -> str:
     trgt_lang_iso = languages_iso_codes[trgt_lang]
 
     messages = [
-        {
-            "role": "user",
-            "content": [
-                {
-                    "type": "text",
-                    "source_lang_code": src_lang_iso,
-                    "target_lang_code": trgt_lang_iso,
-                    "text": text,
-                }
-            ],
-        }
+        (
+            "system", f"You are a helpful assistant that translates {src_lang_iso} to {trgt_lang_iso}. Translate the user sentence."
+        ),
+        (
+            "human", f"{text}"
+        )
     ]
 
-    inputs = processor.apply_chat_template(
-        messages, tokenize=True, add_generation_prompt=True, return_dict=True, return_tensors="pt"
-    ).to(model.device, dtype=torch.bfloat16)
-    input_len = len(inputs['input_ids'][0])
-
-    with torch.inference_mode():
-        generation = model.generate(**inputs, do_sample=False, max_new_tokens=2000)
+    response = model.invoke(messages)
 
-    generation = generation[0][input_len:]
-    decoded = processor.decode(generation, skip_special_tokens=True)
+    return response.content
 
-    return decoded
+    

services/transliteration.py

@@ -1,9 +1,4 @@
-import torch
-from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
-from IndicTransToolkit.processor import IndicProcessor
-
-DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-
+from config import model
 
 transliterate_languages: list = [
     "English",
@@ -63,43 +58,20 @@ transliteration_iso_codes: dict = {
     "Urdu": "urd_Arab"
 }
 
-def indic_transliteration(model_name:str, src_lang:str, trgt_lang:str, text:str) -> str:
-
-    model_id = model_name
-    try:
-        tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
-        model = AutoModelForSeq2SeqLM.from_pretrained(model_id,trust_remote_code=True,torch_dtype=torch.bfloat16).to(DEVICE)
-        print(f"{model_id} model loaded sucessfully.")
-    except Exception as e:
-        print(f"Failed to load {model_id} model.")
-
-    ip = IndicProcessor(inference=True)
-
-    batch = ip.preprocess_batch(
-        [text],
-        src_lang=src_lang,
-        tgt_lang=trgt_lang
-    )
-
-    inputs = tokenizer(batch, truncation=True, padding="longest", return_tensors="pt", return_attention_mask=True).to(DEVICE)
-
-    with torch.no_grad():
-        generated_tokens = model.generate(**inputs, use_cache=True, min_length=0, max_length=2048, num_beams=5, num_return_sequences=1)
-    
-    generated_tokens = tokenizer.batch_decode(generated_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=True)
-
-    translations = ip.postprocess_batch(generated_tokens, lang=trgt_lang)
+def indic_transliteration(src_lang:str, trgt_lang:str, text:str) -> str:
 
-    return "\n".join(translations)
+    src_lang_iso = transliteration_iso_codes[src_lang]
+    trgt_lang_iso = transliteration_iso_codes[trgt_lang]
 
-def transliterate(src_lang:str, trgt_lang:str, text:str) -> str:
+    messages = [
+        (
+            "system", f"You are a helpful assistant that transliterates {src_lang_iso} to {trgt_lang_iso}. Transliterate the user sentence."
+        ),
+        (
+            "human", f"{text}"
+        )
+    ]
 
-    src_lang_iso:str = transliteration_iso_codes.get(src_lang)
-    trgt_lang_iso:str = transliteration_iso_codes.get(trgt_lang)
+    response = model.invoke(messages)
 
-    if "eng_Latn" in src_lang_iso and "eng_Latn" not in trgt_lang_iso:
-        return indic_transliteration("ai4bharat/indictrans2-en-indic-1B", src_lang_iso, trgt_lang_iso, text)
-    elif "eng_Latn" not in src_lang_iso and "eng_Latn" in trgt_lang_iso:
-        return indic_transliteration("ai4bharat/indictrans2-indic-en-1B", src_lang_iso, trgt_lang_iso, text)
-    else:
-        return indic_transliteration("ai4bharat/indictrans2-indic-indic-1B", src_lang_iso, trgt_lang_iso, text)
+    return response.content