SatyaDev13 commited on
Commit
cd6ccaa
·
verified ·
1 Parent(s): 5ca6095

Upload app.py

Browse files
Files changed (1) hide show
  1. app.py +134 -94
app.py CHANGED
@@ -5,10 +5,13 @@ from typing import Optional
5
 
6
  os.environ["GRADIO_SSR_MODE"] = "false"
7
 
8
- import gradio as gr
9
- import spaces
10
  import torch
11
  from transformers import AutoTokenizer, AutoModelForCausalLM
 
 
 
 
 
12
 
13
  logging.basicConfig(
14
  level=logging.INFO,
@@ -51,17 +54,13 @@ def build_prompt(system_prompt: str, user_prompt: str) -> str:
51
  {"role": "system", "content": system_prompt},
52
  {"role": "user", "content": user_prompt},
53
  ]
54
-
55
  if hasattr(tokenizer, "apply_chat_template"):
56
  try:
57
  return tokenizer.apply_chat_template(
58
- messages,
59
- tokenize=False,
60
- add_generation_prompt=True,
61
  )
62
  except Exception:
63
  pass
64
-
65
  return f"System: {system_prompt}\nUser: {user_prompt}\nAssistant:"
66
 
67
 
@@ -69,7 +68,6 @@ def build_prompt(system_prompt: str, user_prompt: str) -> str:
69
  def generate_reply(prompt: str, system_prompt: str, max_tokens: int, temperature: float):
70
  text = build_prompt(system_prompt, prompt)
71
  inputs = tokenizer(text, return_tensors="pt")
72
-
73
  device = next(model.parameters()).device
74
  inputs = {k: v.to(device) for k, v in inputs.items()}
75
 
@@ -88,53 +86,36 @@ def generate_reply(prompt: str, system_prompt: str, max_tokens: int, temperature
88
  return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
89
 
90
 
91
- def chat_fn(prompt, system_prompt, max_tokens, temperature):
92
- prompt = (prompt or "").strip()
93
- system_prompt = (system_prompt or DEFAULT_SYSTEM_PROMPT).strip()
 
 
 
 
 
 
 
94
 
95
- if not prompt:
96
- return "Prompt kosong."
97
-
98
- try:
99
- return generate_reply(
100
- prompt=prompt,
101
- system_prompt=system_prompt,
102
- max_tokens=max_tokens,
103
- temperature=temperature,
104
- )
105
- except Exception as e:
106
- logger.exception("Gagal generate")
107
- return f"Error: {e}"
108
 
 
 
 
109
 
110
- # ============================================================
111
- # Fungsi untuk endpoint REST tambahan (dipanggil lewat gr.api di bawah)
112
- # ============================================================
113
- def api_generate(
114
- prompt: str,
115
- system_prompt: str = DEFAULT_SYSTEM_PROMPT,
116
- max_tokens: int = 1024,
117
- temperature: float = 0.7,
118
- ) -> str:
119
- """Generate teks dari model. Endpoint: /generate (POST lalu GET hasil)."""
120
- prompt = (prompt or "").strip()
121
- if not prompt:
122
- raise gr.Error("Field 'prompt' tidak boleh kosong.")
123
 
124
- try:
125
- return generate_reply(
126
- prompt=prompt,
127
- system_prompt=(system_prompt or DEFAULT_SYSTEM_PROMPT).strip(),
128
- max_tokens=int(max_tokens),
129
- temperature=float(temperature),
130
- )
131
- except Exception as e:
132
- logger.exception("Gagal generate (API)")
133
- raise gr.Error(f"Gagal generate: {e}")
134
 
135
 
136
- def api_health() -> dict:
137
- """Cek status service. Endpoint: /health (POST lalu GET hasil)."""
 
138
  return {
139
  "status": "ok",
140
  "model": MODEL_ID,
@@ -142,56 +123,115 @@ def api_health() -> dict:
142
  }
143
 
144
 
145
- # ============================================================
146
- # UI Gradio + pendaftaran endpoint API tambahan
147
- # ============================================================
148
- with gr.Blocks(title="ZeroGPU Chat API") as demo:
149
- gr.Markdown("# ZeroGPU Chat API")
150
- gr.Markdown(
151
- "UI ini juga punya endpoint API bawaan (`/chat`), "
152
- "plus dua endpoint REST tambahan: `/generate` dan `/health` "
153
- "(lihat tombol 'Use via API' di bagian bawah halaman untuk detail lengkap)."
154
- )
155
 
156
- prompt = gr.Textbox(
157
- label="Prompt",
158
- lines=6,
159
- placeholder="Tulis pesan di sini...",
160
- )
161
- system_prompt = gr.Textbox(
162
- label="System prompt",
163
- value=DEFAULT_SYSTEM_PROMPT,
164
- lines=3,
165
- )
166
- with gr.Row():
167
- max_tokens = gr.Slider(1, 4096, value=1024, step=1, label="Max tokens")
168
- temperature = gr.Slider(0.0, 2.0, value=0.7, step=0.1, label="Temperature")
169
-
170
- output = gr.Textbox(label="Jawaban", lines=10)
171
- btn = gr.Button("Kirim")
172
-
173
- btn.click(
174
- fn=chat_fn,
175
- inputs=[prompt, system_prompt, max_tokens, temperature],
176
- outputs=output,
177
- api_name="chat",
178
- )
179
 
180
- gr.Examples(
181
- examples=[
182
- ["Jelaskan black hole dengan bahasa sederhana.", "Anda adalah asisten AI yang jelas dan ringkas.", 512, 0.7],
183
- ["Buatkan ringkasan sejarah internet.", "Anda adalah asisten AI yang membantu.", 512, 0.6],
184
- ],
185
- inputs=[prompt, system_prompt, max_tokens, temperature],
186
  )
 
 
187
 
188
- # Endpoint tambahan yang TIDAK terikat komponen UI manapun.
189
- # Tetap lewat sistem queue & launch() Gradio, jadi ZeroGPU tetap terdeteksi normal.
190
- gr.api(api_generate, api_name="generate")
191
- gr.api(api_health, api_name="health")
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
192
 
193
- demo.queue(max_size=20, default_concurrency_limit=1)
194
 
195
  if __name__ == "__main__":
196
- demo.launch()
197
 
 
5
 
6
  os.environ["GRADIO_SSR_MODE"] = "false"
7
 
 
 
8
  import torch
9
  from transformers import AutoTokenizer, AutoModelForCausalLM
10
+ import spaces
11
+ from gradio import Server
12
+ from fastapi.middleware.cors import CORSMiddleware
13
+ from fastapi.responses import HTMLResponse
14
+ from pydantic import BaseModel, Field
15
 
16
  logging.basicConfig(
17
  level=logging.INFO,
 
54
  {"role": "system", "content": system_prompt},
55
  {"role": "user", "content": user_prompt},
56
  ]
 
57
  if hasattr(tokenizer, "apply_chat_template"):
58
  try:
59
  return tokenizer.apply_chat_template(
60
+ messages, tokenize=False, add_generation_prompt=True,
 
 
61
  )
62
  except Exception:
63
  pass
 
64
  return f"System: {system_prompt}\nUser: {user_prompt}\nAssistant:"
65
 
66
 
 
68
  def generate_reply(prompt: str, system_prompt: str, max_tokens: int, temperature: float):
69
  text = build_prompt(system_prompt, prompt)
70
  inputs = tokenizer(text, return_tensors="pt")
 
71
  device = next(model.parameters()).device
72
  inputs = {k: v.to(device) for k, v in inputs.items()}
73
 
 
86
  return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
87
 
88
 
89
+ # ============================================================
90
+ # Server (FastAPI asli dari Gradio) — mendukung route POST/GET
91
+ # biasa yang langsung balas hasil dalam satu kali panggilan,
92
+ # SEKALIGUS tetap terdeteksi ZeroGPU karena tetap pakai app.launch().
93
+ # ============================================================
94
+ class GenerateRequest(BaseModel):
95
+ prompt: str = Field(..., description="Pertanyaan / instruksi untuk model")
96
+ system_prompt: Optional[str] = Field(default=DEFAULT_SYSTEM_PROMPT)
97
+ max_tokens: int = Field(default=1024, ge=1, le=4096)
98
+ temperature: float = Field(default=0.7, ge=0.0, le=2.0)
99
 
 
 
 
 
 
 
 
 
 
 
 
 
 
100
 
101
+ class GenerateResponse(BaseModel):
102
+ response: str
103
+ model: str = MODEL_ID
104
 
 
 
 
 
 
 
 
 
 
 
 
 
 
105
 
106
+ app = Server(title="ZeroGPU Chat API")
107
+
108
+ app.add_middleware(
109
+ CORSMiddleware,
110
+ allow_origins=["*"],
111
+ allow_methods=["*"],
112
+ allow_headers=["*"],
113
+ )
 
 
114
 
115
 
116
+ @app.get("/api/health")
117
+ def health():
118
+ """GET /api/health — cek status, tanpa GPU, langsung balas."""
119
  return {
120
  "status": "ok",
121
  "model": MODEL_ID,
 
123
  }
124
 
125
 
126
+ @app.post("/api/generate", response_model=GenerateResponse)
127
+ def api_generate_post(body: GenerateRequest):
128
+ """
129
+ POST /api/generate — kirim JSON, langsung dapat jawaban di response
130
+ yang SAMA (tidak perlu ambil event_id / panggilan kedua).
 
 
 
 
 
131
 
132
+ curl -X POST https://<space-anda>.hf.space/api/generate \\
133
+ -H "Content-Type: application/json" \\
134
+ -d '{"prompt": "Halo, siapa kamu?", "max_tokens": 256, "temperature": 0.7}'
135
+ """
136
+ prompt = (body.prompt or "").strip()
137
+ if not prompt:
138
+ return {"response": "Error: prompt tidak boleh kosong.", "model": MODEL_ID}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
139
 
140
+ text = generate_reply(
141
+ prompt=prompt,
142
+ system_prompt=(body.system_prompt or DEFAULT_SYSTEM_PROMPT).strip(),
143
+ max_tokens=body.max_tokens,
144
+ temperature=body.temperature,
 
145
  )
146
+ return GenerateResponse(response=text)
147
+
148
 
149
+ @app.get("/api/generate", response_model=GenerateResponse)
150
+ def api_generate_get(
151
+ prompt: str,
152
+ system_prompt: str = DEFAULT_SYSTEM_PROMPT,
153
+ max_tokens: int = 1024,
154
+ temperature: float = 0.7,
155
+ ):
156
+ """
157
+ GET /api/generate?prompt=...&max_tokens=...&temperature=...
158
+ Langsung balas jawaban di response yang sama. Praktis untuk tes via browser/curl.
159
+
160
+ curl "https://<space-anda>.hf.space/api/generate?prompt=Halo&max_tokens=256"
161
+ """
162
+ prompt = (prompt or "").strip()
163
+ if not prompt:
164
+ return {"response": "Error: prompt tidak boleh kosong.", "model": MODEL_ID}
165
+
166
+ text = generate_reply(
167
+ prompt=prompt,
168
+ system_prompt=(system_prompt or DEFAULT_SYSTEM_PROMPT).strip(),
169
+ max_tokens=max_tokens,
170
+ temperature=temperature,
171
+ )
172
+ return GenerateResponse(response=text)
173
+
174
+
175
+ # UI sederhana di "/" (custom HTML, panggil /api/generate lewat fetch biasa)
176
+ @app.get("/", response_class=HTMLResponse)
177
+ async def homepage():
178
+ return """
179
+ <!DOCTYPE html>
180
+ <html lang="id">
181
+ <head>
182
+ <meta charset="UTF-8">
183
+ <title>ZeroGPU Chat API</title>
184
+ <style>
185
+ body { background:#0d0d14; color:#e6e6f0; font-family: sans-serif; max-width:720px; margin:40px auto; padding:0 16px; }
186
+ h1 { color:#b98cff; }
187
+ textarea, input { width:100%; background:#1a1a26; color:#e6e6f0; border:1px solid #3a3a55; border-radius:8px; padding:10px; margin:6px 0; box-sizing:border-box; font-family:inherit; }
188
+ button { background:#7c4dff; color:white; border:none; border-radius:8px; padding:10px 20px; cursor:pointer; font-weight:bold; }
189
+ button:disabled { opacity:0.5; cursor:not-allowed; }
190
+ #out { white-space:pre-wrap; background:#14141f; border:1px solid #3a3a55; border-radius:8px; padding:12px; min-height:120px; margin-top:10px; }
191
+ label { font-size:0.85em; color:#9a9ac0; }
192
+ </style>
193
+ </head>
194
+ <body>
195
+ <h1>ZeroGPU Chat API</h1>
196
+ <p>UI ini memanggil <code>POST /api/generate</code> langsung (satu kali panggilan).</p>
197
+ <label>Prompt</label>
198
+ <textarea id="prompt" rows="4" placeholder="Tulis pesan di sini..."></textarea>
199
+ <label>System prompt</label>
200
+ <input id="system_prompt" value="Anda adalah asisten AI yang membantu.">
201
+ <button id="btn" onclick="kirim()">Kirim</button>
202
+ <div id="out"></div>
203
+
204
+ <script>
205
+ async function kirim() {
206
+ const btn = document.getElementById('btn');
207
+ const out = document.getElementById('out');
208
+ btn.disabled = true;
209
+ out.textContent = "Memproses...";
210
+ try {
211
+ const res = await fetch('/api/generate', {
212
+ method: 'POST',
213
+ headers: {'Content-Type': 'application/json'},
214
+ body: JSON.stringify({
215
+ prompt: document.getElementById('prompt').value,
216
+ system_prompt: document.getElementById('system_prompt').value,
217
+ max_tokens: 512,
218
+ temperature: 0.7
219
+ })
220
+ });
221
+ const data = await res.json();
222
+ out.textContent = data.response;
223
+ } catch (e) {
224
+ out.textContent = "Error: " + e;
225
+ } finally {
226
+ btn.disabled = false;
227
+ }
228
+ }
229
+ </script>
230
+ </body>
231
+ </html>
232
+ """
233
 
 
234
 
235
  if __name__ == "__main__":
236
+ app.launch(show_error=True)
237