feat: setting up working fastapi application

README.md

@@ -11,3 +11,4 @@ license: mit
 ---
 
 
+

app.py

@@ -4,13 +4,10 @@
 import gradio as gr
 import markdown
 
-
 # internal imports
 from model import mistral
-from explanation import interpret as shap
-
 
-# function to load md files in pthon as a string
+# function to load md files in python as a string
 def load_md(path):
 
     # credit: official python-markdown documentation (https://python-markdown.github.io/reference/)
@@ -35,7 +32,7 @@ with gr.Blocks() as ui:
             gr.Markdown(
                 """
                 ### ChatBot Demo
-                Mitral AI 7B Model fine-tuned for instruction and fully open source (see at [HGF](https://huggingface.co/mistralai/Mistral-7B-v0.1))
+                Mitral AI 7B notebooks fine-tuned for instruction and fully open source (see at [HGF](https://huggingface.co/mistralai/Mistral-7B-v0.1))
                 """)
         # row with chatbot ui displaying "conversation" with the model (see documentation: https://www.gradio.app/docs/chatbot)
         with gr.Row():
@@ -89,7 +86,7 @@ with gr.Blocks() as ui:
                 """)
 
     # model overview tab for transparency
-    with gr.Tab("Model Overview"):
+    with gr.Tab("notebooks Overview"):
         with gr.Tab("Mistral 7B Instruct"):
             gr.Markdown(value=load_md("./model/mistral.md"))
         with gr.Tab("LlaMa 2 7B Chat"):
@@ -98,8 +95,8 @@ with gr.Blocks() as ui:
     # final row to show legal information - credits, data protection and link to the LICENSE on GitHub
     with gr.Row():
         with gr.Accordion("Credits, Data Protection and License", open=False):
-            gr.Markdown(value=load_md("./public/credits_dataprotection_license.md"))
+            gr.Markdown(value=load_md("public/credits_dataprotection_license.md"))
 
 # launch function for Gradio Interface
 if __name__ == "__main__":
-    ui.launch(debug=True)
+    ui.launch(share=False)

main.py

@@ -0,0 +1,7 @@
+from fastapi import FastAPI
+import gradio as gr
+import app as gradio_app
+
+app = FastAPI()
+
+app = gr.mount_gradio_app(app, gradio_app.ui, path="/")

model/llama2.py

@@ -1,66 +0,0 @@
-from huggingface_hub import InferenceClient
-import os
-
-# huggingface token used to load closed off models
-token = os.environ.get("HGFTOKEN")
-
-# interference client created from mistral 7b instruction fine tuned model
-# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
-interference = InferenceClient(
-    "mistralai/Mistral-7B-Instruct-v0.1"
-)
-
-# default model settings
-model_temperature = 0.7
-model_max_new_tokens = 320
-model_top_p = 0.95
-model_repetition_penalty = 1.1
-
-# chat function - basically the main function calling other functions and returning a response to showcase in chatbot ui
-def chat (prompt, history,):
-
-    # creating formatted prompt and calling for an answer from the model
-    formatted_prompt = format_prompt(prompt, history)
-    answer=respond(formatted_prompt)
-
-    # updating the chat history with the new answer
-    history.append((prompt, answer))
-
-    # returning the chat history to be displayed in the chatbot ui
-    return "",history
-
-# function to format prompt in a way that is understandable for the text generation model
-# credit: copied 1:1 from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
-def format_prompt(message, history):
-    prompt = "<s>"
-
-    # labeling each message in the history as bot or user
-    for user_prompt, bot_response in history:
-        prompt += f"[INST] {user_prompt} [/INST]"
-        prompt += f" {bot_response}</s> "
-    prompt += f"[INST] {message} [/INST]"
-    return prompt
-
-# function to get the response
-# credit: minimally changed from Hugging Face, Inc/ Omar Sanseviero (see https://huggingface.co/spaces/osanseviero/mistral-super-fast/)
-def respond(formatted_prompt):
-
-    # setting model temperature and
-    temperature = float(model_temperature)
-    if temperature < 1e-2:
-        temperature = 1e-2
-    top_p = float(model_top_p)
-
-    # creating model arguments/settings
-    generate_kwargs = dict(
-        temperature=temperature,
-        max_new_tokens=model_max_new_tokens,
-        top_p=top_p,
-        repetition_penalty=model_repetition_penalty,
-        do_sample=True,
-        seed=42,
-    )
-
-    # calling for model output and returning it
-    output = interference.text_generation(formatted_prompt, **generate_kwargs, stream=False, details=True, return_full_text=False).generated_text
-    return output

requirements.txt

@@ -4,4 +4,6 @@ torch~=2.1.1
 shap~=0.43.0
 accelerate~=0.24.1
 markdown~=3.5.1
-huggingface_hub~=0.19.4
+huggingface_hub~=0.19.4
+fastapi~=0.104.1
+uvicorn~=0.24.0