ZeroGPU fast chat for Diba v0.1

README.md

@@ -1,13 +1,17 @@
 ---
-title: Diba Chat Gpu
-emoji: 💻
+title: Diba Chat GPU
+emoji: ⚡
 colorFrom: indigo
-colorTo: indigo
+colorTo: pink
 sdk: gradio
-sdk_version: 6.27.0
-python_version: '3.12'
+sdk_version: "5.49.1"
+python_version: "3.10"
 app_file: app.py
 pinned: false
+license: apache-2.0
+short_description: دیبا، تست سریع روی GPU
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# دیبا · تست سریع
+
+نسخه‌ی آزمایشی دیبا روی ZeroGPU، ساخته‌ی [دیباچین](https://dibachain.ir).

app.py

@@ -0,0 +1,76 @@
+"""
+Diba fast chat (ZeroGPU). Loads the merged v0.1 weights from the private staging repo with transformers
+and streams replies on a ZeroGPU slice. Thinking is always off.
+"""
+import os
+from threading import Thread
+
+import gradio as gr
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+
+MODEL_ID = os.environ.get("DIBA_MODEL", "DibaAi/Diba-Base-staging-private")
+TOKEN = os.environ.get("HF_TOKEN")
+
+tok = AutoTokenizer.from_pretrained(MODEL_ID, token=TOKEN)
+model = AutoModelForCausalLM.from_pretrained(MODEL_ID, token=TOKEN, dtype=torch.bfloat16)
+model.to("cuda")  # ZeroGPU attaches the GPU only inside @spaces.GPU calls
+model.eval()
+
+
+def build_messages(message, history, system):
+    msgs = [{"role": "system", "content": system}] if system.strip() else []
+    for turn in history:
+        if isinstance(turn, dict):
+            msgs.append({"role": turn["role"], "content": turn["content"]})
+        else:
+            user, bot = turn
+            msgs += [{"role": "user", "content": user}, {"role": "assistant", "content": bot}]
+    msgs.append({"role": "user", "content": message})
+    return msgs
+
+
+@spaces.GPU(duration=90)
+def chat(message, history, system, temperature, max_new_tokens):
+    msgs = build_messages(message, history, system)
+    inputs = tok.apply_chat_template(msgs, add_generation_prompt=True, enable_thinking=False,
+                                     return_tensors="pt", return_dict=True).to("cuda")
+    streamer = TextIteratorStreamer(tok, skip_prompt=True, skip_special_tokens=True)
+    kwargs = dict(**inputs, streamer=streamer, max_new_tokens=int(max_new_tokens),
+                  do_sample=temperature > 0, temperature=max(float(temperature), 1e-5), top_p=0.9,
+                  repetition_penalty=1.05)
+    Thread(target=model.generate, kwargs=kwargs).start()
+    out = ""
+    for piece in streamer:
+        out += piece
+        yield out
+
+
+CSS = """
+.gradio-container {direction: rtl; font-family: Vazirmatn, Tahoma, sans-serif;}
+.message, .prose {text-align: right;}
+pre, code {direction: ltr; text-align: left;}
+"""
+
+demo = gr.ChatInterface(
+    chat,
+    type="messages",
+    title="دیبا · تست سریع",
+    description="نسخه‌ی آزمایشی ۰.۱ روی GPU. تاریخ ایران بپرس یا درخواست کد بده.",
+    additional_inputs=[
+        gr.Textbox(value="", label="پیام سیستم (اختیاری)"),
+        gr.Slider(0.0, 1.2, value=0.4, step=0.1, label="دما"),
+        gr.Slider(64, 2048, value=768, step=32, label="حداکثر طول پاسخ"),
+    ],
+    examples=[
+        ["سلام، خودت رو معرفی کن.", "", 0.4, 256],
+        ["کوروش بزرگ کی بود و چه کارهایی کرد؟", "", 0.4, 512],
+        ["یه تابع جاوااسکریپت به اسم toPersianDigits بنویس که اعداد انگلیسی رشته رو فارسی کنه.", "", 0.2, 512],
+        ["یه صفحه‌ی ورود راست‌چین با HTML و CSS بساز.", "", 0.4, 1536],
+    ],
+    css=CSS,
+)
+
+if __name__ == "__main__":
+    demo.launch()

requirements.txt

@@ -0,0 +1,5 @@
+transformers>=5.0
+accelerate>=1.0
+huggingface_hub>=1.0
+flash-linear-attention
+spaces