Call via llama

app.py

@@ -1,13 +1,21 @@
+import os
 import gradio as gr
-from huggingface_hub import InferenceClient
+import copy
+from llama_cpp import Llama
+from huggingface_hub import hf_hub_download  
 
-"""
-For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-"""
-client = InferenceClient("TheBloke/phi-2-GGUF")
 
+llm = Llama(
+    model_path=hf_hub_download(
+        repo_id=os.environ.get("REPO_ID", "TheBloke/phi-2-GGUF"),
+        filename=os.environ.get("MODEL_FILE", "phi-2.Q4_K_M.gguf"),
+    ),
+    n_ctx=2048,
+    n_gpu_layers=50, # change n_gpu_layers if you have more or less VRAM 
+) 
 
-def respond(
+
+def generate_text(
     message,
     history: list[tuple[str, str]],
     system_message,
@@ -15,35 +23,51 @@ def respond(
     temperature,
     top_p,
 ):
-    messages = [{"role": "system", "content": system_message}]
-
-    for val in history:
-        if val[0]:
-            messages.append({"role": "user", "content": val[0]})
-        if val[1]:
-            messages.append({"role": "assistant", "content": val[1]})
+    temp = ""
+    input_prompt = f"[INST] <<SYS>>\n{system_message}\n<</SYS>>\n\n "
+    for interaction in history:
+        input_prompt = input_prompt + str(interaction[0]) + " [/INST] " + str(interaction[1]) + " </s><s> [INST] "
 
-    messages.append({"role": "user", "content": message})
+    input_prompt = input_prompt + str(message) + " [/INST] "
 
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
+    output = llm(
+        input_prompt,
         temperature=temperature,
         top_p=top_p,
-    ):
-        token = message.choices[0].delta.content
+        top_k=40,
+        repeat_penalty=1.1,
+        max_tokens=max_tokens,
+        stop=[
+            "<|prompter|>",
+            "<|endoftext|>",
+            "<|endoftext|> \n",
+            "ASSISTANT:",
+            "USER:",
+            "SYSTEM:",
+        ],
+        stream=True,
+    )
+    for out in output:
+        stream = copy.deepcopy(out)
+        temp += stream["choices"][0]["text"]
+        yield temp
 
-        response += token
-        yield response
 
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
 demo = gr.ChatInterface(
-    respond,
+    generate_text,
+    title="llama-cpp-python on GPU",
+    description="Running LLM with https://github.com/abetlen/llama-cpp-python",
+    examples=[
+        ['How to setup a human base on Mars? Give short answer.'],
+        ['Explain theory of relativity to me like I’m 8 years old.'],
+        ['What is 9,000 * 9,000?'],
+        ['Write a pun-filled happy birthday message to my friend Alex.'],
+        ['Justify why a penguin might make a good king of the jungle.']
+    ],
+    cache_examples=False,
+    retry_btn=None,
+    undo_btn="Delete Previous",
+    clear_btn="Clear",
     additional_inputs=[
         gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
@@ -60,4 +84,4 @@ demo = gr.ChatInterface(
 
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()