Sync

README.md

@@ -8,27 +8,15 @@ sdk_version: 5.38.0
 pinned: false
 ---
 
+Translate from English to Ukrainian.
+
 ## Install
 
-```shell
+```bash
 uv venv --python 3.10
 
 source .venv/bin/activate
 
 uv pip install -r requirements.txt
-
-# in development mode
 uv pip install -r requirements-dev.txt
 ```
-
-## Build image
-
-```shell
-docker build -t en-uk-translator .
-```
-
-## Run
-
-```shell
-docker run -it --rm -p 8888:7860 en-uk-translator
-```

app.py

@@ -1,10 +1,6 @@
 import sys
 import time
 
-from importlib.metadata import version
-from gradio.utils import is_zero_gpu_space
-from gradio.themes import Base
-
 try:
     import spaces
 except ImportError:
@@ -14,6 +10,11 @@ import torch
 import torchaudio
 import gradio as gr
 import torchaudio.transforms as T
+import polars as pl
+
+from importlib.metadata import version
+from gradio.utils import is_zero_gpu_space
+from gradio.themes import Base
 
 from transformers import (
     AutoModelForCausalLM,
@@ -122,18 +123,6 @@ This space translates your text, audio, image from English to Ukrainian. Also, c
 """.strip()
 
 
-translated_text_value = """
-Choose an example below the Translate button or type your text.
-""".strip()
-
-translated_audio_value = """
-Choose an example below the Translate button or upload your audio.
-""".strip()
-
-translated_image_value = """
-Choose an example below the Translate button or upload your image.
-""".strip()
-
 tech_env = f"""
 #### Environment
 
@@ -155,6 +144,39 @@ tech_libraries = f"""
 """.strip()
 
 
+def translate(text: str) -> str:
+    prompt = "Translate the text to Ukrainian:\n" + text
+
+    input_ids = tokenizer.apply_chat_template(
+        [{"role": "user", "content": prompt}],
+        add_generation_prompt=True,
+        return_tensors="pt",
+        tokenize=True,
+    ).to(model.device)
+
+    output = model.generate(
+        input_ids,
+        max_new_tokens=2048,
+        # Greedy Search
+        do_sample=False,
+        repetition_penalty=1.05,
+        # Sampling
+        # do_sample=True,
+        # temperature=0.1,
+        # # top_k=1,
+        # min_p=0.9,
+        # repetition_penalty=1.05,
+    )
+
+    prompt_len = input_ids.shape[1]
+    generated_tokens = output[:, prompt_len:]
+    translated_text = tokenizer.batch_decode(
+        generated_tokens, skip_special_tokens=True
+    )[0]
+
+    return translated_text.strip()
+
+
 @spaces.GPU
 def inference_text(text, progress=gr.Progress()):
     if not text:
@@ -176,36 +198,7 @@ def inference_text(text, progress=gr.Progress()):
         non_empty_sentences, desc="Translating...", unit="sentence"
     ):
         t0 = time.time()
-
-        prompt = "Translate the text to Ukrainian:\n" + sentence
-
-        input_ids = tokenizer.apply_chat_template(
-            [{"role": "user", "content": prompt}],
-            add_generation_prompt=True,
-            return_tensors="pt",
-            tokenize=True,
-        ).to(model.device)
-
-        output = model.generate(
-            input_ids,
-            max_new_tokens=2048,
-            # Greedy Search
-            do_sample=False,
-            repetition_penalty=1.05,
-            # Sampling
-            # do_sample=True,
-            # temperature=0.1,
-            # # top_k=1,
-            # min_p=0.9,
-            # repetition_penalty=1.05,
-        )
-
-        prompt_len = input_ids.shape[1]
-        generated_tokens = output[:, prompt_len:]
-        translated_text = tokenizer.batch_decode(
-            generated_tokens, skip_special_tokens=True
-        )[0]
-
+        translated_text = translate(sentence)
         elapsed_time = round(time.time() - t0, 2)
 
         translated_text = translated_text.strip()
@@ -219,15 +212,7 @@ def inference_text(text, progress=gr.Progress()):
 
     gr.Info("Finished!", duration=2)
 
-    result_texts = []
-
-    for result in results:
-        result_texts.append(f"{result['translated_text']}\n")
-
-    sum_elapsed_text = sum([result["elapsed_time"] for result in results])
-    print(f"Elapsed time: {round(sum_elapsed_text, 4)} seconds")
-
-    return "\n".join(result_texts)
+    return pl.DataFrame(results)
 
 
 @spaces.GPU
@@ -287,39 +272,9 @@ def inference_audio(audio, progress=gr.Progress()):
         non_empty_sentences, desc="Translating...", unit="sentence"
     ):
         t0 = time.time()
-
-        prompt = "Translate the text to Ukrainian:\n" + sentence
-
-        input_ids = tokenizer.apply_chat_template(
-            [{"role": "user", "content": prompt}],
-            add_generation_prompt=True,
-            return_tensors="pt",
-            tokenize=True,
-        ).to(model.device)
-
-        output = model.generate(
-            input_ids,
-            max_new_tokens=2048,
-            # Greedy Search
-            do_sample=False,
-            repetition_penalty=1.05,
-            # Sampling
-            # do_sample=True,
-            # temperature=0.1,
-            # # top_k=1,
-            # min_p=0.9,
-            # repetition_penalty=1.05,
-        )
-
-        prompt_len = input_ids.shape[1]
-        generated_tokens = output[:, prompt_len:]
-        translated_text = tokenizer.batch_decode(
-            generated_tokens, skip_special_tokens=True
-        )[0]
-
+        translated_text = translate(sentence)
         elapsed_time = round(time.time() - t0, 2)
 
-        translated_text = translated_text.strip()
         results.append(
             {
                 "sentence": sentence,
@@ -330,15 +285,7 @@ def inference_audio(audio, progress=gr.Progress()):
 
     gr.Info("Finished!", duration=2)
 
-    result_texts = []
-
-    for result in results:
-        result_texts.append(f"{result['sentence']}: {result['translated_text']}\n")
-
-    sum_elapsed_text = sum([result["elapsed_time"] for result in results])
-    print(f"Elapsed time: {round(sum_elapsed_text, 4)} seconds")
-
-    return "\n".join(result_texts)
+    return pl.DataFrame(results)
 
 
 @spaces.GPU
@@ -365,39 +312,9 @@ def inference_image(image, progress=gr.Progress()):
 
     for sentence in progress.tqdm(sentences, desc="Translating...", unit="sentence"):
         t0 = time.time()
-
-        prompt = "Translate the text to Ukrainian:\n" + sentence
-
-        input_ids = tokenizer.apply_chat_template(
-            [{"role": "user", "content": prompt}],
-            add_generation_prompt=True,
-            return_tensors="pt",
-            tokenize=True,
-        ).to(model.device)
-
-        output = model.generate(
-            input_ids,
-            max_new_tokens=2048,
-            # Greedy Search
-            do_sample=False,
-            repetition_penalty=1.05,
-            # Sampling
-            # do_sample=True,
-            # temperature=0.1,
-            # # top_k=1,
-            # min_p=0.9,
-            # repetition_penalty=1.05,
-        )
-
-        prompt_len = input_ids.shape[1]
-        generated_tokens = output[:, prompt_len:]
-        translated_text = tokenizer.batch_decode(
-            generated_tokens, skip_special_tokens=True
-        )[0]
-
+        translated_text = translate(sentence)
         elapsed_time = round(time.time() - t0, 2)
 
-        translated_text = translated_text.strip()
         results.append(
             {
                 "sentence": sentence,
@@ -408,15 +325,7 @@ def inference_image(image, progress=gr.Progress()):
 
     gr.Info("Finished!", duration=2)
 
-    result_texts = []
-
-    for result in results:
-        result_texts.append(f"> {result['sentence']}: {result['translated_text']}\n")
-
-    sum_elapsed_text = sum([result["elapsed_time"] for result in results])
-    print(f"Elapsed time: {round(sum_elapsed_text, 4)} seconds")
-
-    return "\n".join(result_texts)
+    return pl.DataFrame(results)
 
 
 def create_app():
@@ -430,11 +339,8 @@ def create_app():
         gr.Markdown(description_head)
         gr.Markdown("## Usage")
 
-        translated_text = gr.Textbox(
+        translated_text = gr.DataFrame(
             label="Translated text",
-            placeholder=translated_text_value,
-            show_copy_button=True,
-            lines=5,
         )
 
         text = gr.Textbox(label="Text", autofocus=True, lines=5)
@@ -457,11 +363,8 @@ def create_audio_app():
         gr.Markdown(description_head)
         gr.Markdown("## Usage")
 
-        translated_text = gr.Textbox(
+        translated_text = gr.DataFrame(
             label="Translated text",
-            placeholder=translated_audio_value,
-            show_copy_button=True,
-            lines=5,
         )
 
         audio = gr.Audio(label="Audio file", sources="upload", type="filepath")
@@ -486,11 +389,8 @@ def create_image_app():
         gr.Markdown(description_head)
         gr.Markdown("## Usage")
 
-        translated_text = gr.Textbox(
+        translated_text = gr.DataFrame(
             label="Translated text",
-            placeholder=translated_image_value,
-            show_copy_button=True,
-            lines=5,
         )
 
         image = gr.Image(label="Image file", sources="upload", type="filepath")

justfile

@@ -0,0 +1,3 @@
+default:
+    ruff check
+    ruff format

requirements.txt

@@ -8,3 +8,5 @@ torchaudio
 accelerate
 
 python-doctr
+
+polars