Demo for prompt-tuned opt-2.7b

app.py

@@ -0,0 +1,26 @@
+import gradio as gr
+import torch
+
+from transformers import pipeline, AutoTokenizer, AutoModel
+
+#pipe = pipeline("text-generation", model="furquan/opt_2_7_b_prompt_tuned_sentiment_analysis", trust_remote_code=True, cache_dir="/local/home/furquanh/myProjects/week12/").to('cuda')
+
+tokenizer = AutoTokenizer.from_pretrained("furquan/opt_2_7_b_prompt_tuned_sentiment_analysis",cache_dir="/local/home/furquanh/myProjects/week12/", trust_remote_code=True)
+model = AutoModel.from_pretrained("furquan/opt_2_7_b_prompt_tuned_sentiment_analysis",cache_dir="/local/home/furquanh/myProjects/week12/", trust_remote_code=True)
+
+title = "OPT-2.7B"
+description = "This demo uses meta's opt-2.7b model prompt tuned on the Stanford Sentiment Treebank-5 way dataset to only output the sentiment of a given text."
+article = "<p style='text-align: center'><a href='https://arxiv.org/pdf/2104.08691.pdf' target='_blank'>The Power of Scale for Parameter-Efficient Prompt Tuning</a></p>"
+
+
+def sentiment(text):
+    tokenized = tokenizer(text, return_tensors='pt')
+    with torch.no_grad():
+        outputs = model.generate(
+        input_ids=tokenized["input_ids"], attention_mask=tokenized["attention_mask"]
+        )
+    return f"text: {text} Sentiment: {tokenizer.decode(outputs[0][-3:], skip_special_tokens=True)}"
+
+iface = gr.Interface(fn=sentiment, inputs="text", outputs="text", title=title,
+    description=description, article=article)
+iface.launch()