Update app.py

app.py

@@ -1,398 +1,41 @@
-# pylint: disable=line-too-long, broad-exception-caught, invalid-name, missing-function-docstring, too-many-instance-attributes, missing-class-docstring
-# ruff: noqa: E501
-import gc
-import os
-import platform
-import random
-import time
-from dataclasses import asdict, dataclass
-from pathlib import Path
-from typing import Optional, Sequence
-
-# from types import SimpleNamespace
 import gradio as gr
-import psutil
-from about_time import about_time
-from ctransformers import AutoModelForCausalLM
-from examples_list import examples_list
-from loguru import logger
-
-LLM = None
-gc.collect()
-
-# raise SystemExit(0)
-
-# Prompt template: Guanaco
-# {past_history}
-prompt_template = """You are a helpful assistant. Let's think step by step.
-### Human:
-{question}
-### Assistant:"""
-
-# Prompt template: garage-bAInd/Stable-Platypus2-13B
-prompt_template = """
-### System:
-This is a system prompt, please behave and help the user.
-
-### Instruction:
-
-{question}
-
-### Response:
-"""
-prompt_template = """
-[INST] Write code to solve the following coding problem that obeys the constraints and
-passes the example test cases. Please wrap your code answer using ```:
+from huggingface_hub import hf_hub_download
+from llama_cpp import Llama
 
-{question}
-[/INST]
-"""
+## Download the GGUF model
+model_name = "jackangel/LLama_3_Instruct_SPFx_Docs_Unsloth"
+model_file = "Llama_3_Instruct_SPFx_Docs_Unsloth.Q4_K_M.gguf" # this is the specific model file we'll use in this example. It's a 4-bit quant, but other levels of quantization are available in the model repo if preferred
+model_path = hf_hub_download(model_name, filename=model_file)
 
-human_prefix = "### Instruction"
-ai_prefix = "### Response"
-stop_list = [f"{human_prefix}:"]
 
 _ = psutil.cpu_count(logical=False) - 1
 cpu_count: int = int(_) if _ else 1
 logger.debug(f"{cpu_count=}")
 
-LLM = AutoModelForCausalLM.from_pretrained(
-    "jackangel/LLama_3_Instruct_SPFx_Docs_Unsloth",
-    model_type="llama",
-    threads=cpu_count,
-)
-
-os.environ["TZ"] = "Asia/Shanghai"
-try:
-    time.tzset()  # type: ignore # pylint: disable=no-member
-except Exception:
-    # Windows
-    logger.warning("Windows, cant run time.tzset()")
-
-
-# ctransformers.Config() default
-# Config(top_k=40, top_p=0.95, temperature=0.8,
-# repetition_penalty=1.1, last_n_tokens=64, seed=-1,
-# batch_size=8, threads=-1, max_new_tokens=256,
-# stop=None, stream=False, reset=True,
-# context_length=-1, gpu_layers=0)
-@dataclass
-class GenerationConfig:
-    temperature: float = 0.7
-    top_k: int = 50
-    top_p: float = 0.9
-    repetition_penalty: float = 1.0
-    max_new_tokens: int = 512
-    seed: int = 42
-    reset: bool = False
-    stream: bool = True
-    threads: int = cpu_count
-    # stop: list[str] = field(default_factory=lambda: stop_list)
-
-# ctransformers\llm.py
-@dataclass
-class Config:
-    # sample
-    top_k: int = 40
-    top_p: float = 0.95
-    temperature: float = 0.8
-    repetition_penalty: float = 1.1
-    last_n_tokens: int = 64
-    seed: int = -1
-
-    # eval
-    batch_size: int = 8
-    threads: int = -1
-
-    # generate
-    max_new_tokens: int = 512  # 256
-    stop: Optional[Sequence[str]] = None
-    stream: bool = True  # False
-    reset: bool = False  # True
-
-    # model
-    # context_length: int = -1
-    # gpu_layers: int = 0
-
-
-def generate(
-    question: str,
-    llm=LLM,
-    # config: GenerationConfig = GenerationConfig(),
-    config: Config = Config(),
-):
-    """Run model inference, will return a Generator if streaming is true."""
-    # _ = prompt_template.format(question=question)
-    # print(_)
-
-    prompt = prompt_template.format(question=question)
-
-    return llm(
-        prompt,
-        **asdict(config),
-        # **vars(config),
-    )
-
-
-# logger.debug(f"{asdict(GenerationConfig())=}")
-logger.debug(f"{Config(stream=True)=}")
-logger.debug(f"{vars(Config(stream=True))=}")
-
-
-def user(user_message, history):
-    # return user_message, history + [[user_message, None]]
-    if history is None:
-        history = []
-    history.append([user_message, None])
-    return user_message, history  # keep user_message
-
-
-def user1(user_message, history):
-    # return user_message, history + [[user_message, None]]
-    if history is None:
-        history = []
-    history.append([user_message, None])
-    return "", history  # clear user_message
-
-
-def bot_(history):
-    user_message = history[-1][0]
-    resp = random.choice(["How are you?", "I love you", "I'm very hungry"])
-    bot_message = user_message + ": " + resp
-    history[-1][1] = ""
-    for character in bot_message:
-        history[-1][1] += character
-        time.sleep(0.02)
-        yield history
-
-    history[-1][1] = resp
-    yield history
-
-
-def bot(history):
-    user_message = ""
-    try:
-        user_message = history[-1][0]
-    except Exception as exc:
-        logger.error(exc)
-    response = []
-
-    logger.debug(f"{user_message=}")
-
-    with about_time() as atime:  # type: ignore
-        flag = 1
-        prefix = ""
-        then = time.time()
-
-        logger.debug("about to generate")
-
-        config = GenerationConfig(reset=True)
-        for elm in generate(user_message, config=config):
-            if flag == 1:
-                logger.debug("in the loop")
-                prefix = f"({time.time() - then:.2f}s) "
-                flag = 0
-                print(prefix, end="", flush=True)
-                logger.debug(f"{prefix=}")
-            print(elm, end="", flush=True)
-            # logger.debug(f"{elm}")
-
-            response.append(elm)
-            history[-1][1] = prefix + "".join(response)
-            yield history
-
-    _ = (
-        f"(time elapsed: {atime.duration_human}, "  # type: ignore
-        f"{atime.duration/len(''.join(response)):.2f}s/char)"  # type: ignore
-    )
-
-    history[-1][1] = "".join(response) + f"\n{_}"
-    yield history
-
-
-def predict_api(prompt):
-    logger.debug(f"{prompt=}")
-    try:
-        # user_prompt = prompt
-        config = GenerationConfig(
-            temperature=0.2,
-            top_k=10,
-            top_p=0.9,
-            repetition_penalty=1.0,
-            max_new_tokens=512,  # adjust as needed
-            seed=42,
-            reset=True,  # reset history (cache)
-            stream=False,
-            # threads=cpu_count,
-            # stop=prompt_prefix[1:2],
-        )
-
-        response = generate(
-            prompt,
-            config=config,
-        )
-
-        logger.debug(f"api: {response=}")
-    except Exception as exc:
-        logger.error(exc)
-        response = f"{exc=}"
-    # bot = {"inputs": [response]}
-    # bot = [(prompt, response)]
-
-    return response
-
-
-css = """
-    .importantButton {
-        background: linear-gradient(45deg, #7e0570,#5d1c99, #6e00ff) !important;
-        border: none !important;
-    }
-    .importantButton:hover {
-        background: linear-gradient(45deg, #ff00e0,#8500ff, #6e00ff) !important;
-        border: none !important;
-    }
-    .disclaimer {font-variant-caps: all-small-caps; font-size: xx-small;}
-    .xsmall {font-size: x-small;}
-"""
-
-logger.info("start block")
-
-with gr.Blocks(
-    title=f"{Path(model_loc).name}",
-    # theme=gr.themes.Soft(text_size="sm", spacing_size="sm"),
-    theme=gr.themes.Glass(text_size="sm", spacing_size="sm"),
-    css=css,
-) as block:
-    # buff_var = gr.State("")
-    with gr.Accordion("🎈 Info", open=False):
-        gr.Markdown(
-            f"""<h5><center>{Path(model_loc).name}</center></h4>
-            Doesn't quite work -- no output or run forever.""",
-            elem_classes="xsmall",
-        )
-
-    # chatbot = gr.Chatbot().style(height=700)  # 500
-    chatbot = gr.Chatbot(height=500)
-
-    # buff = gr.Textbox(show_label=False, visible=True)
-
-    with gr.Row():
-        with gr.Column(scale=5):
-            msg = gr.Textbox(
-                label="Chat Message Box",
-                placeholder="Ask me anything (press Shift+Enter or click Submit to send)",
-                show_label=False,
-                # container=False,
-                lines=6,
-                max_lines=30,
-                show_copy_button=True,
-                # ).style(container=False)
-            )
-        with gr.Column(scale=1, min_width=50):
-            with gr.Row():
-                submit = gr.Button("Submit", elem_classes="xsmall")
-                stop = gr.Button("Stop", visible=True)
-                clear = gr.Button("Clear History", visible=True)
-    with gr.Row(visible=False):
-        with gr.Accordion("Advanced Options:", open=False):
-            with gr.Row():
-                with gr.Column(scale=2):
-                    system = gr.Textbox(
-                        label="System Prompt",
-                        value=prompt_template,
-                        show_label=False,
-                        container=False,
-                        # ).style(container=False)
-                    )
-                with gr.Column():
-                    with gr.Row():
-                        change = gr.Button("Change System Prompt")
-                        reset = gr.Button("Reset System Prompt")
-
-    with gr.Accordion("Example Inputs", open=True):
-        examples = gr.Examples(
-            examples=examples_list,
-            inputs=[msg],
-            examples_per_page=40,
-        )
-
-    # with gr.Row():
-    with gr.Accordion("Disclaimer", open=False):
-        _ = Path(model_loc).name
-        gr.Markdown(
-            f"Disclaimer: {_} can produce factually incorrect output, and should not be relied on to produce "
-            f"factually accurate information. {_} was trained on various public datasets; while great efforts "
-            "have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
-            "biased, or otherwise offensive outputs.",
-            elem_classes=["disclaimer"],
-        )
-
-    msg_submit_event = msg.submit(
-        # fn=conversation.user_turn,
-        fn=user,
-        inputs=[msg, chatbot],
-        outputs=[msg, chatbot],
-        queue=True,
-        show_progress="full",
-        # api_name=None,
-    ).then(bot, chatbot, chatbot, queue=True)
-    submit_click_event = submit.click(
-        # fn=lambda x, y: ("",) + user(x, y)[1:],  # clear msg
-        fn=user1,  # clear msg
-        inputs=[msg, chatbot],
-        outputs=[msg, chatbot],
-        queue=True,
-        # queue=False,
-        show_progress="full",
-        # api_name=None,
-    ).then(bot, chatbot, chatbot, queue=True)
-    stop.click(
-        fn=None,
-        inputs=None,
-        outputs=None,
-        cancels=[msg_submit_event, submit_click_event],
-        queue=False,
-    )
-    clear.click(lambda: None, None, chatbot, queue=False)
-
-    with gr.Accordion("For Chat/Translation API", open=False, visible=False):
-        input_text = gr.Text()
-        api_btn = gr.Button("Go", variant="primary")
-        out_text = gr.Text()
-
-    api_btn.click(
-        predict_api,
-        input_text,
-        out_text,
-        api_name="api",
-    )
-
-    # block.load(update_buff, [], buff, every=1)
-    # block.load(update_buff, [buff_var], [buff_var, buff], every=1)
-
-# concurrency_count=5, max_size=20
-# max_size=36, concurrency_count=14
-# CPU cpu_count=2 16G, model 7G
-# CPU UPGRADE cpu_count=8 32G, model 7G
-
-# does not work
-_ = """
-# _ = int(psutil.virtual_memory().total / 10**9 // file_size - 1)
-# concurrency_count = max(_, 1)
-if psutil.cpu_count(logical=False) >= 8:
-    # concurrency_count = max(int(32 / file_size) - 1, 1)
-else:
-    # concurrency_count = max(int(16 / file_size) - 1, 1)
-# """
-
-# default concurrency_count = 1
-# block.queue(concurrency_count=concurrency_count, max_size=5).launch(debug=True)
-
-server_port = 7860
-if "forindo" in platform.node():
-    server_port = 7861
-block.queue(max_size=5).launch(
-    debug=True, server_name="0.0.0.0", server_port=server_port
+## Instantiate model from downloaded file
+llm = Llama(
+    model_path=model_path,
+    n_ctx=1024,  # Context length to use
+    n_threads=cpu_count,            # Number of CPU threads to use
+    n_gpu_layers=0        # Number of model layers to offload to GPU
 )
 
-# block.queue(max_size=5).launch(debug=True, server_name="0.0.0.0")
+## Generation kwargs
+generation_kwargs = {
+    "max_tokens":1000,
+    "stop":["</s>"],
+    "echo":False, # Echo the prompt in the output
+    "top_k":1 # This is essentially greedy decoding, since the model will always return the highest-probability token. Set this value > 1 for sampling decoding
+}
+
+
+def chatbot(message, history):
+    prompt = "INSTRUCTION: You are a helpful assistant\nINPUT: " + message + "\nOUTPUT:"
+    airemember =  ""
+    for human,assistant in history:
+
+        airemember += "USER: " + human + "\nASSISTANT:" + assistant+"\n\n"
+    sendtoai = airemember + prompt
+    result = llm(sendtoai, **generation_kwargs)
+    return result
+app = gr.ChatInterface(chatbot)