Add application file

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* 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
+*.spm filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,77 @@
+import io, os, tempfile
+import gradio as gr
+from huggingface_hub import snapshot_download
+import ctranslate2
+import sentencepiece as spm
+import hanlp
+import re
+MODEL_REPO = "ogaith/zhen-ctranslate2"  # your CT2 model repo
+SRC_SPM = "source.spm"
+TGT_SPM = "target.spm"
+
+# Download model once into the Space cache
+MODEL_DIR = snapshot_download(MODEL_REPO)
+
+# Load translators / tokenizers
+translator = ctranslate2.Translator(MODEL_DIR, device="auto")
+sp_src = spm.SentencePieceProcessor(os.path.join(MODEL_DIR, SRC_SPM))
+sp_tgt = spm.SentencePieceProcessor(os.path.join(MODEL_DIR, TGT_SPM))
+hanlp_tok = hanlp.load(hanlp.pretrained.tok.COARSE_ELECTRA_SMALL_ZH)
+
+def preprocess_source(hanlp_tok, text: str) -> str:
+    text = text.strip()
+    if not text:
+        return ""
+    tokens = hanlp_tok(text)
+    text = "".join(tokens)
+
+    text = re.sub(r'([\u4e00-\u9fff])(\d)', r'\1 \2', text)
+    text = re.sub(r'(\d)([\u4e00-\u9fff])', r'\1 \2', text)
+
+    text = re.sub(r'\s+', ' ', text).strip()
+    return text
+
+def translate_file(file_obj, beam_size, max_len, batch_size):
+    if file_obj is None:
+        return None
+    data = file_obj.decode("utf-8").splitlines()
+
+    out_lines = []
+    for i in range(0, len(data), batch_size):
+        chunk = data[i:i+batch_size]
+        pre = [preprocess_source(s) for s in chunk]
+        src_tok = [sp_src.encode(s, out_type=str) for s in pre]
+        results = translator.translate_batch(
+            src_tok,
+            beam_size=int(beam_size),
+            max_decoding_length=int(max_len)
+        )
+        for r in results:
+            out_lines.append(sp_tgt.decode(r.hypotheses[0]))
+
+    # Return a downloadable txt
+    return "\n".join(out_lines) + ("\n" if out_lines else "")
+
+with gr.Blocks() as demo:
+    gr.Markdown("# Chinese ➜ English (CTranslate2, HanLP+SentencePiece)")
+    gr.Markdown("Upload a UTF-8 `.txt` with one sentence per line (Chinese).")
+    with gr.Row():
+        inp = gr.File(label="Upload .txt", file_count="single", type="bytes")
+    with gr.Row():
+        beam = gr.Slider(1, 8, value=4, step=1, label="Beam size")
+        max_len = gr.Slider(16, 512, value=256, step=1, label="Max decoding length")
+        bs = gr.Slider(1, 128, value=32, step=1, label="Batch size")
+    btn = gr.Button("Translate")
+    out = gr.File(label="Download translations (.txt)")
+
+    def _run(file_bytes, beam_size, max_len, batch_size):
+        text = translate_file(file_bytes, beam_size, max_len, batch_size)
+        if text is None:
+            return None
+        tmp = tempfile.NamedTemporaryFile(delete=False, suffix=".txt", mode="w", encoding="utf-8")
+        tmp.write(text); tmp.close()
+        return tmp.name
+
+    btn.click(_run, [inp, beam, max_len, bs], out)
+
+demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+ctranslate2
+sentencepiece
+hanlp
+gradio
+huggingface_hub

source.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09889d7c2bb953f6ad7e99abbf4c6c332f833db32663e2501cc90daf82e6746d
+size 419918

target.spm

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c66a7dc5c34aba632c5ba256d99bb34f4086f61f73038e4978660ae773b2a5cc
+size 799721