Initial commit

app.py

@@ -0,0 +1,35 @@
+import gradio as gr
+from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
+
+model_path = "anilbhatt1/phi2-oasst-guanaco-bf16-custom"
+model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)
+tokenizer = AutoTokenizer.from_pretrained(model_path)
+
+def generate_text(prompt, response_length):   
+            
+    prompt = str(prompt)
+    max_len = int(response_length)    
+
+    gen = pipeline('text-generation', model=model, tokenizer=tokenizer, max_length=max_len)
+    result = gen(prompt)
+    output_msg = result[0]['generated_text']
+    return output_msg
+
+def gradio_fn(prompt, response_length):
+    output_txt_msg = generate_text(prompt, response_length)    
+    return output_txt_msg
+
+markdown_description = """
+- This is a Gradio app that answers the query you ask it
+- Uses **microsoft/phi-2 qlora** optimized model finetuned on **timdettmers/openassistant-guanaco** dataset 
+"""
+demo = gr.Interface(fn=gradio_fn,
+                    inputs=[gr.Textbox(info="How may I help you ? please enter your prompt here..."),
+                            gr.Slider(value=50, minimum=50, maximum=200, \
+                                      info="Choose a response length min chars=50, max=200")],                       
+                    outputs=gr.Textbox(),
+                    title="phi2 - Dialog Partner",
+                    description=markdown_description,
+                    article=" **Credits** : https://github.com/mshumer/gpt-llm-trainer ")
+
+demo.queue().launch(share=True)

requirements.txt

@@ -0,0 +1,5 @@
+gradio==3.50.2
+torch>=2.1.0
+transformers
+tokenizers
+einops