init

app.py

@@ -0,0 +1,46 @@
+import os
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
+import gradio as gr
+import spaces
+
+huggingface_token = os.getenv("HUGGINGFACE_TOKEN")
+if not huggingface_token:
+    raise ValueError("HUGGINGFACE_TOKEN environment variable is not set")
+
+model_id = "meta-llama/Meta-Llama-3.1-8B-Instruct"
+device = "cuda"
+dtype = torch.bfloat16
+
+tokenizer = AutoTokenizer.from_pretrained(model_id, token=huggingface_token)
+
+
+@spaces.GPU
+def generate_text(prompt, system_message="You are a helpful assistant."):
+    model = AutoModelForCausalLM.from_pretrained(
+        model_id, torch_dtype=dtype, device_map=device, token=huggingface_token
+    )
+    text_generator = pipeline("text-generation", model=model, tokenizer=tokenizer, torch_dtype=dtype, device_map=device)
+
+    messages = [
+        {"role": "system", "content": system_message},
+        {"role": "user", "content": prompt},
+    ]
+
+    result = text_generator(messages, max_new_tokens=256, do_sample=True, temperature=0.7)
+    return result[0]["generated_text"]
+
+
+iface = gr.Interface(
+    fn=generate_text,
+    inputs=[
+        gr.Textbox(lines=3, label="Input Prompt"),
+        gr.Textbox(lines=2, label="System Message", value="You are a helpful assistant."),
+    ],
+    outputs=gr.Textbox(label="Generated Text"),
+    title="Llama 3.1 8B Instruct Text Generation",
+    description="Enter a prompt and optional system message to generate text using the Llama 3.1 8B Instruct model.",
+)
+
+if __name__ == "__main__":
+    iface.launch()

requirements.txt

@@ -0,0 +1,6 @@
+numpy
+torch
+spaces
+bitsandbytes
+accelerate
+transformers