Upload folder using huggingface_hub

README.md

@@ -7,7 +7,7 @@ sdk: gradio
 sdk_version: 6.30.0
 app_file: app.py
 python_version: "3.12"
-short_description: Terjemahan ID-EN dan eksplorasi model CALI
+short_description: Indonesian-English translation and CALI text generation
 license: apache-2.0
 models:
   - Sandroeth/cali-id-en-translate
@@ -16,22 +16,27 @@ models:
 
 # cali
 
-Demo dua model karya Sandroeth: terjemahan Indonesia–Inggris dua arah dan kelanjutan teks dengan CALI-0.1B.
+Explore two lightweight language models by Sandroeth in an English-language interface.
 
-## Inferensi
+- **Translation:** translate everyday sentences between Indonesian and English.
+- **Explore CALI:** continue a text prompt with the experimental CALI-0.1B base model.
 
-Tab terjemahan mengikuti [README model](https://huggingface.co/Sandroeth/cali-id-en-translate/blob/main/README.md): prompt `[id→en]\n{text}\n→` atau `[en→id]\n{text}\n→`, `do_sample=False`, `repetition_penalty=1.5`, `no_repeat_ngram_size=3`, `use_cache=False`, dan default 64 token. Opsi kalimat pertama aktif secara default sesuai contoh resmi; nonaktifkan untuk melihat keluaran lengkap.
+## Inference
 
-Tab model dasar memakai prompt teks biasa tanpa chat template. Model eksperimental ini dilatih dari awal dengan data terbatas; kualitas keluaran bisa bervariasi.
+Translation follows the [official model README](https://huggingface.co/Sandroeth/cali-id-en-translate/blob/main/README.md): `[id→en]\n{text}\n→` or `[en→id]\n{text}\n→`, deterministic decoding, `repetition_penalty=1.5`, `no_repeat_ngram_size=3`, `use_cache=False`, and 64 output tokens by default. The first-sentence option matches the official example; disable it to inspect the complete generated output.
 
-Model dimuat sekali dengan kode arsitektur kustom dari revisi yang dikunci dalam `app.py`. Float32 dipertahankan karena kode RoPE penerjemah menghasilkan tensor float32. Transformers 5.9.0 mengikuti versi pada konfigurasi model. Total input dan keluaran dibatasi 1024 token; maksimal keluaran 256 token.
+The base model receives a plain text prompt without a chat template. Both models use their custom architecture code at the revisions pinned in `app.py`. Float32 preserves compatibility with the translation model's RoPE implementation. Transformers 5.9.0 matches the translation checkpoint configuration.
 
-## Hosting
+The combined input and output budget is 1024 tokens, with up to 256 output tokens. Translation inputs are limited to 200 words.
 
-Dirancang untuk Gradio ZeroGPU; pilih hardware ZeroGPU lewat pengaturan Space atau CLI (`--flavor zero-a10g`). Kedua handler memakai `@spaces.GPU`. CPU juga didukung jika tersedia pada akun. `gradio`, `spaces`, `torch`, dan `huggingface_hub` disediakan runtime Spaces.
+## Hosting and API
 
-API Gradio: `/translate` dan `/complete`. MCP diaktifkan saat aplikasi diluncurkan. Contoh di-cache secara lazy setelah dipanggil. Statistik menampilkan token dan waktu inferensi.
+This Gradio Space uses ZeroGPU. Both models are loaded once at startup, and the inference handlers use `@spaces.GPU`. CPU execution is also supported when available. Spaces manages Gradio, PyTorch, `spaces`, and `huggingface_hub`.
 
-## Keterbatasan
+Gradio API endpoints: `/translate` and `/complete`. MCP is enabled. Examples are cached lazily after first use. Each response includes token counts and inference time.
 
-Terjemahan paling sesuai untuk kalimat sehari-hari, maksimal 200 kata per permintaan. Input sangat pendek dan topik teknis bisa tidak akurat. Model dasar adalah model pelengkap teks eksperimental.
+## Limitations
+
+Translation works best on everyday sentences. Very short inputs, technical topics, and long passages may be unreliable. CALI-0.1B is an experimental base model trained on limited data; continuations can repeat, drift off-topic, or be inconsistent.
+
+Both models are released under Apache 2.0 by Sandroeth.

app.py

@@ -27,7 +27,7 @@ def run_model(key: str, prompt: str, max_new_tokens: int, **options) -> tuple[st
     inputs = tokenizer(prompt, return_tensors="pt", return_token_type_ids=False).to(DEVICE)
     input_length = inputs["input_ids"].shape[1]
     if input_length + max_new_tokens > model.config.seq_len:
-        raise gr.Error("Teks terlalu panjang. Pendekkan input atau kurangi jumlah token keluaran (batas total 1024 token).")
+        raise gr.Error("Text is too long. Shorten the input or reduce the output limit (1024 tokens total).")
     start = time.perf_counter()
     with torch.inference_mode():
         output = model.generate(
@@ -38,23 +38,23 @@ def run_model(key: str, prompt: str, max_new_tokens: int, **options) -> tuple[st
     text = tokenizer.decode(generated, skip_special_tokens=True).strip()
     elapsed = time.perf_counter() - start
     print(f"{key}: input={input_length}, output={len(generated)}, seconds={elapsed:.2f}", flush=True)
-    return text, f"{input_length} token input · {len(generated)} token keluaran · {elapsed:.2f} detik"
+    return text, f"{input_length} input tokens · {len(generated)} output tokens · {elapsed:.2f} seconds"
 
 
 @spaces.GPU(duration=60)
-def translate(text: str, direction: str = "Indonesia → English", max_new_tokens: int = 64,
+def translate(text: str, direction: str = "Indonesian → English", max_new_tokens: int = 64,
               first_sentence: bool = True) -> tuple[str, str]:
     """Translate a short sentence between Indonesian and English using CALI's trained prompt format."""
     if not text or not text.strip():
-        raise gr.Error("Masukkan teks yang ingin diterjemahkan.")
-    if direction not in ("Indonesia → English", "English → Indonesia"):
-        raise gr.Error("Arah terjemahan tidak valid.")
+        raise gr.Error("Enter the text you want to translate.")
+    if direction not in ("Indonesian → English", "English → Indonesian"):
+        raise gr.Error("Invalid translation direction.")
     max_new_tokens = int(max_new_tokens)
     if not 16 <= max_new_tokens <= 256:
-        raise gr.Error("Pilih 16–256 token keluaran.")
+        raise gr.Error("Choose between 16 and 256 output tokens.")
     if len(text.split()) > 200:
-        raise gr.Error("Gunakan maksimal 200 kata; model ini ditujukan untuk teks pendek.")
-    tag = "[id→en]" if direction == "Indonesia → English" else "[en→id]"
+        raise gr.Error("Use up to 200 words; this model is designed for short texts.")
+    tag = "[id→en]" if direction == "Indonesian → English" else "[en→id]"
     result, stats = run_model("translate", f"{tag}\n{text.strip()}\n→", max_new_tokens,
                               do_sample=False, repetition_penalty=1.5, no_repeat_ngram_size=3)
     if first_sentence:
@@ -67,10 +67,10 @@ def complete(prompt: str, max_new_tokens: int = 96, temperature: float = 0.7,
              top_p: float = 0.9) -> tuple[str, str]:
     """Continue an Indonesian or English text prompt with the experimental CALI base language model."""
     if not prompt or not prompt.strip():
-        raise gr.Error("Masukkan awal teks untuk dilanjutkan.")
+        raise gr.Error("Enter a text prompt to continue.")
     max_new_tokens = int(max_new_tokens)
     if not 16 <= max_new_tokens <= 256 or not 0 <= temperature <= 1.5 or not 0.1 <= top_p <= 1:
-        raise gr.Error("Parameter generasi di luar rentang yang didukung.")
+        raise gr.Error("Generation settings are outside the supported range.")
     options = {"do_sample": temperature > 0, "repetition_penalty": 1.1}
     if temperature > 0:
         options.update(temperature=float(temperature), top_p=float(top_p))
@@ -78,41 +78,41 @@ def complete(prompt: str, max_new_tokens: int = 96, temperature: float = 0.7,
 
 
 with gr.Blocks(title="cali") as demo:
-    gr.Markdown("# cali\nDua model ringan karya **Sandroeth** untuk menjelajahi bahasa Indonesia dan Inggris.")
-    with gr.Tab("Terjemahan"):
-        gr.Markdown("### Dari satu bahasa ke bahasa lain\nPaling cocok untuk kalimat sehari-hari. Input 1–2 kata, istilah teknis, dan teks panjang bisa menghasilkan terjemahan yang kurang akurat.")
-        direction = gr.Radio(["Indonesia → English", "English → Indonesia"], value="Indonesia → English", label="Arah terjemahan")
+    gr.Markdown("# cali\nTwo lightweight models by **Sandroeth** for exploring Indonesian and English.")
+    with gr.Tab("Translation"):
+        gr.Markdown("### From one language to another\nBest for everyday sentences. Very short inputs (1–2 words), technical terms, and long passages may produce less accurate translations.")
+        direction = gr.Radio(["Indonesian → English", "English → Indonesian"], value="Indonesian → English", label="Translation direction")
         with gr.Row():
-            source = gr.Textbox(label="Teks sumber", lines=5, placeholder="Dia pergi ke pasar setiap pagi.", max_length=6000)
-            translated = gr.Textbox(label="Hasil terjemahan", lines=5, interactive=False)
-        with gr.Accordion("Pengaturan", open=False):
-            tokens = gr.Slider(16, 256, value=64, step=1, label="Maksimal token keluaran")
-            first = gr.Checkbox(value=True, label="Ambil kalimat pertama (sesuai contoh resmi)")
-        translate_button = gr.Button("Terjemahkan", variant="primary")
-        translation_stats = gr.Textbox(label="Statistik", interactive=False)
+            source = gr.Textbox(label="Source text", lines=5, placeholder="Dia pergi ke pasar setiap pagi.", max_length=6000)
+            translated = gr.Textbox(label="Translation", lines=5, interactive=False)
+        with gr.Accordion("Settings", open=False):
+            tokens = gr.Slider(16, 256, value=64, step=1, label="Maximum output tokens")
+            first = gr.Checkbox(value=True, label="Return the first sentence only (matches the official example)")
+        translate_button = gr.Button("Translate", variant="primary")
+        translation_stats = gr.Textbox(label="Stats", interactive=False)
         translate_button.click(translate, [source, direction, tokens, first], [translated, translation_stats], api_name="translate")
         gr.Examples([
-            ["Dia pergi ke pasar setiap pagi.", "Indonesia → English"],
-            ["Saya makan nasi.", "Indonesia → English"],
-            ["The weather is very cold today.", "English → Indonesia"],
-            ["She is happy.", "English → Indonesia"],
+            ["Dia pergi ke pasar setiap pagi.", "Indonesian → English"],
+            ["Saya makan nasi.", "Indonesian → English"],
+            ["The weather is very cold today.", "English → Indonesian"],
+            ["She is happy.", "English → Indonesian"],
         ], inputs=[source, direction], outputs=[translated, translation_stats], fn=translate,
             cache_examples=True, cache_mode="lazy")
-    with gr.Tab("Eksplorasi CALI"):
-        gr.Markdown("### Lanjutkan sebuah teks\nCALI-0.1B adalah model dasar eksperimental. Berikan awal kalimat atau paragraf untuk diteruskan. Hasil dapat berulang atau tidak konsisten.")
-        prompt = gr.Textbox(label="Awal teks", lines=4, placeholder="Indonesia adalah negara yang")
-        continuation = gr.Textbox(label="Kelanjutan yang dihasilkan", lines=6, interactive=False)
-        with gr.Accordion("Pengaturan", open=False):
-            base_tokens = gr.Slider(16, 256, value=96, step=1, label="Maksimal token keluaran")
-            temperature = gr.Slider(0, 1.5, value=0.7, step=0.05, label="Temperature (0 = deterministik)")
+    with gr.Tab("Explore CALI"):
+        gr.Markdown("### Continue a piece of text\nCALI-0.1B is an experimental base model. Start a sentence or paragraph and let it continue. Outputs may repeat or become inconsistent.")
+        prompt = gr.Textbox(label="Prompt", lines=4, placeholder="Indonesia adalah negara yang")
+        continuation = gr.Textbox(label="Generated continuation", lines=6, interactive=False)
+        with gr.Accordion("Settings", open=False):
+            base_tokens = gr.Slider(16, 256, value=96, step=1, label="Maximum output tokens")
+            temperature = gr.Slider(0, 1.5, value=0.7, step=0.05, label="Temperature (0 = deterministic)")
             top_p = gr.Slider(0.1, 1, value=0.9, step=0.05, label="Top-p")
-        generate_button = gr.Button("Lanjutkan teks", variant="primary")
-        base_stats = gr.Textbox(label="Statistik", interactive=False)
+        generate_button = gr.Button("Generate continuation", variant="primary")
+        base_stats = gr.Textbox(label="Stats", interactive=False)
         generate_button.click(complete, [prompt, base_tokens, temperature, top_p], [continuation, base_stats], api_name="complete")
         gr.Examples([["Indonesia adalah negara yang"], ["The sun rises in the"]],
                     inputs=[prompt], outputs=[continuation, base_stats], fn=complete,
                     cache_examples=True, cache_mode="lazy")
-    gr.Markdown("Model: [CALI Translate](https://huggingface.co/Sandroeth/cali-id-en-translate) · [CALI-0.1B](https://huggingface.co/Sandroeth/cali-0.1B) · Lisensi Apache 2.0")
+    gr.Markdown("Model: [CALI Translate](https://huggingface.co/Sandroeth/cali-id-en-translate) · [CALI-0.1B](https://huggingface.co/Sandroeth/cali-0.1B) · Apache 2.0 license")
 
 if __name__ == "__main__":
     demo.queue(max_size=20, default_concurrency_limit=1).launch(mcp_server=True)