Spaces:
Running on Zero
Running on Zero
Upload app.py
Browse files
app.py
CHANGED
|
@@ -30,10 +30,14 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
| 30 |
if tokenizer.pad_token is None:
|
| 31 |
tokenizer.pad_token = tokenizer.eos_token
|
| 32 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 33 |
logger.info(f"Memuat model: {MODEL_ID}")
|
| 34 |
model = AutoModelForCausalLM.from_pretrained(
|
| 35 |
MODEL_ID,
|
| 36 |
-
torch_dtype=
|
| 37 |
device_map="auto",
|
| 38 |
low_cpu_mem_usage=True,
|
| 39 |
token=HF_TOKEN,
|
|
@@ -69,9 +73,9 @@ def generate_reply(prompt: str, system_prompt: str, max_tokens: int, temperature
|
|
| 69 |
with torch.no_grad():
|
| 70 |
outputs = model.generate(
|
| 71 |
**inputs,
|
| 72 |
-
max_new_tokens=max_tokens,
|
| 73 |
do_sample=True,
|
| 74 |
-
temperature=temperature,
|
| 75 |
top_p=0.9,
|
| 76 |
pad_token_id=tokenizer.eos_token_id,
|
| 77 |
)
|
|
@@ -81,43 +85,49 @@ def generate_reply(prompt: str, system_prompt: str, max_tokens: int, temperature
|
|
| 81 |
return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
|
| 82 |
|
| 83 |
def chat_fn(prompt, system_prompt, max_tokens, temperature):
|
| 84 |
-
|
|
|
|
|
|
|
|
|
|
| 85 |
return "Prompt kosong."
|
|
|
|
| 86 |
try:
|
| 87 |
return generate_reply(
|
| 88 |
-
prompt=prompt
|
| 89 |
-
system_prompt=system_prompt
|
| 90 |
-
max_tokens=
|
| 91 |
-
temperature=
|
| 92 |
)
|
| 93 |
except Exception as e:
|
| 94 |
logger.exception("Gagal generate")
|
| 95 |
return f"Error: {e}"
|
| 96 |
|
| 97 |
-
with gr.Blocks(title="ZeroGPU Chat") as demo:
|
| 98 |
-
gr.Markdown("# ZeroGPU Chat")
|
| 99 |
-
gr.Markdown("
|
| 100 |
-
|
| 101 |
-
with gr.Row():
|
| 102 |
-
prompt = gr.Textbox(label="Prompt", lines=6, placeholder="Tulis pesan di sini...")
|
| 103 |
-
output = gr.Textbox(label="Jawaban", lines=10)
|
| 104 |
-
|
| 105 |
-
with gr.Row():
|
| 106 |
-
system_prompt = gr.Textbox(
|
| 107 |
-
label="System prompt",
|
| 108 |
-
value="Anda adalah asisten AI yang membantu.",
|
| 109 |
-
lines=3,
|
| 110 |
-
)
|
| 111 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 112 |
with gr.Row():
|
| 113 |
max_tokens = gr.Slider(1, 4096, value=1024, step=1, label="Max tokens")
|
| 114 |
temperature = gr.Slider(0.0, 2.0, value=0.7, step=0.1, label="Temperature")
|
| 115 |
|
|
|
|
| 116 |
btn = gr.Button("Kirim")
|
|
|
|
| 117 |
btn.click(
|
| 118 |
fn=chat_fn,
|
| 119 |
inputs=[prompt, system_prompt, max_tokens, temperature],
|
| 120 |
outputs=output,
|
|
|
|
| 121 |
)
|
| 122 |
|
| 123 |
gr.Examples(
|
|
|
|
| 30 |
if tokenizer.pad_token is None:
|
| 31 |
tokenizer.pad_token = tokenizer.eos_token
|
| 32 |
|
| 33 |
+
dtype = torch.bfloat16 if torch.cuda.is_available() and torch.cuda.is_bf16_supported() else (
|
| 34 |
+
torch.float16 if torch.cuda.is_available() else torch.float32
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
logger.info(f"Memuat model: {MODEL_ID}")
|
| 38 |
model = AutoModelForCausalLM.from_pretrained(
|
| 39 |
MODEL_ID,
|
| 40 |
+
torch_dtype=dtype,
|
| 41 |
device_map="auto",
|
| 42 |
low_cpu_mem_usage=True,
|
| 43 |
token=HF_TOKEN,
|
|
|
|
| 73 |
with torch.no_grad():
|
| 74 |
outputs = model.generate(
|
| 75 |
**inputs,
|
| 76 |
+
max_new_tokens=int(max_tokens),
|
| 77 |
do_sample=True,
|
| 78 |
+
temperature=float(temperature),
|
| 79 |
top_p=0.9,
|
| 80 |
pad_token_id=tokenizer.eos_token_id,
|
| 81 |
)
|
|
|
|
| 85 |
return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
|
| 86 |
|
| 87 |
def chat_fn(prompt, system_prompt, max_tokens, temperature):
|
| 88 |
+
prompt = (prompt or "").strip()
|
| 89 |
+
system_prompt = (system_prompt or "Anda adalah asisten AI yang membantu.").strip()
|
| 90 |
+
|
| 91 |
+
if not prompt:
|
| 92 |
return "Prompt kosong."
|
| 93 |
+
|
| 94 |
try:
|
| 95 |
return generate_reply(
|
| 96 |
+
prompt=prompt,
|
| 97 |
+
system_prompt=system_prompt,
|
| 98 |
+
max_tokens=max_tokens,
|
| 99 |
+
temperature=temperature,
|
| 100 |
)
|
| 101 |
except Exception as e:
|
| 102 |
logger.exception("Gagal generate")
|
| 103 |
return f"Error: {e}"
|
| 104 |
|
| 105 |
+
with gr.Blocks(title="ZeroGPU Chat API") as demo:
|
| 106 |
+
gr.Markdown("# ZeroGPU Chat API")
|
| 107 |
+
gr.Markdown("UI ini juga punya endpoint API bawaan yang bisa dipanggil via curl/requests.")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
|
| 109 |
+
prompt = gr.Textbox(
|
| 110 |
+
label="Prompt",
|
| 111 |
+
lines=6,
|
| 112 |
+
placeholder="Tulis pesan di sini...",
|
| 113 |
+
)
|
| 114 |
+
system_prompt = gr.Textbox(
|
| 115 |
+
label="System prompt",
|
| 116 |
+
value="Anda adalah asisten AI yang membantu.",
|
| 117 |
+
lines=3,
|
| 118 |
+
)
|
| 119 |
with gr.Row():
|
| 120 |
max_tokens = gr.Slider(1, 4096, value=1024, step=1, label="Max tokens")
|
| 121 |
temperature = gr.Slider(0.0, 2.0, value=0.7, step=0.1, label="Temperature")
|
| 122 |
|
| 123 |
+
output = gr.Textbox(label="Jawaban", lines=10)
|
| 124 |
btn = gr.Button("Kirim")
|
| 125 |
+
|
| 126 |
btn.click(
|
| 127 |
fn=chat_fn,
|
| 128 |
inputs=[prompt, system_prompt, max_tokens, temperature],
|
| 129 |
outputs=output,
|
| 130 |
+
api_name="chat",
|
| 131 |
)
|
| 132 |
|
| 133 |
gr.Examples(
|