Update app.py

app.py

@@ -1,126 +1,143 @@
-#!/usr/bin/env python
+import spaces
+import json
+import subprocess
+from llama_cpp import Llama
+from llama_cpp_agent import LlamaCppAgent, MessagesFormatterType
+from llama_cpp_agent.providers import LlamaCppPythonProvider
+from llama_cpp_agent.chat_history import BasicChatHistory
+from llama_cpp_agent.chat_history.messages import Roles
+import gradio as gr
+from huggingface_hub import hf_hub_download
 
-import os
-from collections.abc import Iterator
-from threading import Thread
+hf_hub_download(
+    repo_id="fluently-lm/FluentlyLM-Prinum-Q4_K_M-GGUF",
+    filename="fluentlylm-prinum-q4_k_m.gguf",
+    local_dir="./models"
+)
 
-import gradio as gr
-import spaces
-import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+@spaces.GPU(duration=90)
+def respond(
+    message,
+    history: list[tuple[str, str]],
+    model,
+    system_message,
+    max_tokens,
+    temperature,
+    top_p,
+    top_k,
+    repeat_penalty,
+):
+    chat_template = MessagesFormatterType.GEMMA_2
 
-DESCRIPTION = "# FluentlyLM Prinum"
+    llm = Llama(
+        model_path=f"models/{model}",
+        flash_attn=True,
+        n_gpu_layers=81,
+        n_batch=1024,
+        n_ctx=8192,
+    )
+    provider = LlamaCppPythonProvider(llm)
 
-MAX_MAX_NEW_TOKENS = 2048
-DEFAULT_MAX_NEW_TOKENS = 1024
-MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
+    agent = LlamaCppAgent(
+        provider,
+        system_prompt=f"{system_message}",
+        predefined_messages_formatter_type=chat_template,
+        debug_output=True
+    )
 
-device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
+    settings = provider.get_provider_default_settings()
+    settings.temperature = temperature
+    settings.top_k = top_k
+    settings.top_p = top_p
+    settings.max_tokens = max_tokens
+    settings.repeat_penalty = repeat_penalty
+    settings.stream = True
 
-model_id = "fluently-lm/FluentlyLM-Prinum"
-tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    device_map="auto",
-    torch_dtype=torch.bfloat16,
-)
-model.eval()
+    messages = BasicChatHistory()
 
+    for msn in history:
+        user = {
+            'role': Roles.user,
+            'content': msn[0]
+        }
+        assistant = {
+            'role': Roles.assistant,
+            'content': msn[1]
+        }
+        messages.add_message(user)
+        messages.add_message(assistant)
 
-@spaces.GPU(duration=120)
-def generate(
-    message: str,
-    chat_history: list[dict],
-    system_prompt: str = "",
-    max_new_tokens: int = 1024,
-    temperature: float = 0.7,
-    top_p: float = 0.8,
-    top_k: int = 20,
-    repetition_penalty: float = 1.05,
-) -> Iterator[str]:
-    messages = []
-    if system_prompt:
-        messages.append({"role": "system", "content": system_prompt})
-    messages.extend(chat_history.copy())
-    messages.append({"role": "user", "content": message})
+    stream = agent.get_chat_response(
+        message,
+        llm_sampling_settings=settings,
+        chat_history=messages,
+        returns_streaming_generator=True,
+        print_output=False
+    )
 
-    input_ids = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
-    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
-        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
-        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
-    input_ids = input_ids.to(model.device)
+    outputs = ""
+    for output in stream:
+        outputs += output
+        yield outputs
 
-    streamer = TextIteratorStreamer(tokenizer, timeout=30.0, skip_prompt=True, skip_special_tokens=True)
-    generate_kwargs = dict(
-        {"input_ids": input_ids},
-        streamer=streamer,
-        max_new_tokens=max_new_tokens,
-        do_sample=True,
-        top_p=top_p,
-        top_k=top_k,
-        temperature=temperature,
-        num_beams=1,
-        repetition_penalty=repetition_penalty,
+def create_interface(model_name, description):
+    return gr.ChatInterface(
+        respond,
+        additional_inputs=[
+            gr.Textbox(value=model_name, label="Model", interactive=False),
+            gr.Textbox(value="You are Fluently, a helpful assistant. You should think step-by-step. You should answer all question by prioritizing the principle of openness", label="System message"),
+            gr.Slider(minimum=1, maximum=4096, value=2048, step=1, label="Max tokens"),
+            gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
+            gr.Slider(
+                minimum=0.1,
+                maximum=1.0,
+                value=0.95,
+                step=0.05,
+                label="Top-p",
+            ),
+            gr.Slider(
+                minimum=0,
+                maximum=100,
+                value=40,
+                step=1,
+                label="Top-k",
+            ),
+            gr.Slider(
+                minimum=0.0,
+                maximum=2.0,
+                value=1.1,
+                step=0.1,
+                label="Repetition penalty",
+            ),
+        ],
+        retry_btn="Retry",
+        undo_btn="Undo",
+        clear_btn="Clear",
+        submit_btn="Send",
+        title=f"{model_name}",
+        description=description,
+        examples=[
+        ["Hi! How are you?",
+         "Write a short story about a scary island.",
+         "Prove that the force of gravity applies to all bodies in the Universe.",
+         "Give examples of how a quantum computer works."
+        ],
+        ],
+        chatbot=gr.Chatbot(
+            label=None,
+            scale=1,
+            likeable=True,
+            show_copy_button=True
+        )
     )
-    t = Thread(target=model.generate, kwargs=generate_kwargs)
-    t.start()
 
-    outputs = []
-    for text in streamer:
-        outputs.append(text)
-        yield "".join(outputs)
+description = """<h2 align="center"<bold>FluentlyLM Prinum</bold> Demo</h2>"""
+interface = create_interface('fluentlylm-prinum-q4_k_m.gguf', description)
 
+demo = gr.Blocks()
 
-demo = gr.ChatInterface(
-    fn=generate,
-    additional_inputs=[
-        gr.Textbox(label="System Prompt", value="You are FluentlyLM, created by Project Fluently. You are a helpful assistant."),
-        gr.Slider(
-            label="Max new tokens",
-            minimum=1,
-            maximum=MAX_MAX_NEW_TOKENS,
-            step=1,
-            value=DEFAULT_MAX_NEW_TOKENS,
-        ),
-        gr.Slider(
-            label="Temperature",
-            minimum=0.1,
-            maximum=4.0,
-            step=0.1,
-            value=0.65,
-        ),
-        gr.Slider(
-            label="Top-p (nucleus sampling)",
-            minimum=0.05,
-            maximum=1.0,
-            step=0.05,
-            value=0.8,
-        ),
-        gr.Slider(
-            label="Top-k",
-            minimum=1,
-            maximum=1000,
-            step=1,
-            value=20,
-        ),
-        gr.Slider(
-            label="Repetition penalty",
-            minimum=1.0,
-            maximum=2.0,
-            step=0.05,
-            value=1.05,
-        ),
-    ],
-    stop_btn=None,
-    examples=[
-        ["Hi! How are you?"],
-    ],
-    cache_examples=False,
-    type="messages",
-    description=DESCRIPTION,
-    css_paths="style.css",
-    fill_height=True,
-)
+with demo:
+    interface.render()
 
 if __name__ == "__main__":
     demo.launch()