first finance bot

app.py

@@ -0,0 +1,54 @@
+import transformers
+import gradio as gr
+import warnings
+import torch
+warnings.simplefilter('ignore')
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+tokenizer = transformers.GPT2Tokenizer.from_pretrained('gpt2')
+#add padding token, beginstring and endstring tokens
+tokenizer.add_special_tokens(
+{
+    "pad_token":"<pad>",
+    "bos_token":"<startstring>",
+    "eos_token":"<endstring>"
+})
+#add bot token since it is not a special token
+tokenizer.add_tokens(["<bot>:"])
+
+model = transformers.GPT2LMHeadModel.from_pretrained('gpt2')
+model.resize_token_embeddings(len(tokenizer))
+model.load_state_dict(torch.load('gpt2talk.pt', map_location=torch.device('cpu')))
+
+model.eval()
+def inference(quiz):
+    quiz1 = quiz
+    quiz = "<startstring>"+quiz+" <bot>:"
+    
+    quiztoken = tokenizer(quiz,
+                          return_tensors='pt'
+                         )
+    
+    answer = model.generate(**quiztoken, max_length=200, top_k=0.7,top_p=0.1)[0]
+    answer = tokenizer.decode(answer, skip_special_tokens=True)
+    answer = answer.replace(" <bot>:","").replace(quiz1,"") + '.'
+    return answer
+
+def chatbot(input_text):
+    response = inference(input_text)
+    return response
+
+# Create the Gradio interface
+iface = gr.Interface(
+    fn=chatbot,
+    inputs=gr.Textbox(),
+    outputs=gr.Textbox(),
+    live=False, #set false to avoid caching
+    interpretation="chat",
+    title="ChatFinance",
+    description="Ask the a question and see its response!",
+)
+
+# Launch the Gradio interface
+iface.launch()

gpt2talk.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbb4f3318512a3c112ad2fc12db6a7fb41b1beb98d4ead0825728a228705fb66
+size 497826671

requirements.txt

@@ -0,0 +1,4 @@
+torch==2.0.1
+transformers==4.31.0
+gradio==3.44.4
+gradio_client==0.5.1