app.py

import gradio as gr
import os, gc, copy, torch
from datetime import datetime
from huggingface_hub import hf_hub_download
from pynvml import *
nvmlInit()
gpu_h = nvmlDeviceGetHandleByIndex(0)
ctx_limit = 3000
title = "RWKV-v5-Eagle-World-7B-v2-20240128-ctx4096"

os.environ["RWKV_JIT_ON"] = '1'
os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)

from rwkv.model import RWKV
model_path = hf_hub_download(repo_id="RWKV/v5-Eagle-7B", filename=f"{title}.pth")
model = RWKV(model=model_path, strategy='cuda fp16i8 *8 -> cuda fp16')
from rwkv.utils import PIPELINE, PIPELINE_ARGS
pipeline = PIPELINE(model, "rwkv_vocab_v20230424")

def generate_prompt(instruction, input=""):
    instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
    input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
    if input:
        return f"""### Instruction: {instruction}

### Input: {input}

### Response:"""
    else:
        return f"""### User: hi

### Assistant: Hi. I am your assistant and I will provide expert full response in full details. Please feel free to ask any question and I will always answer it.

### User: {instruction}

### Assistant:"""

def evaluate(
    ctx,
    token_count=200,
    temperature=1.0,
    top_p=0.7,
    presencePenalty = 0.1,
    countPenalty = 0.1,
):
    args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
                     alpha_frequency = countPenalty,
                     alpha_presence = presencePenalty,
                     token_ban = [], # ban the generation of some tokens
                     token_stop = [0]) # stop generation whenever you see any token here
    ctx = ctx.strip()
    all_tokens = []
    out_last = 0
    out_str = ''
    occurrence = {}
    state = None
    for i in range(int(token_count)):
        out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
        for n in occurrence:
            out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)

        token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
        if token in args.token_stop:
            break
        all_tokens += [token]
        for xxx in occurrence:
            occurrence[xxx] *= 0.996        
        if token not in occurrence:
            occurrence[token] = 1
        else:
            occurrence[token] += 1
        
        tmp = pipeline.decode(all_tokens[out_last:])
        if '\ufffd' not in tmp:
            out_str += tmp
            yield out_str.strip()
            out_last = i + 1

    gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
    timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')  
    del out
    del state
    gc.collect()
    torch.cuda.empty_cache()
    yield out_str.strip()

examples = [
    ["### Assistant:\nSure! Here is a very detailed plan to create flying pigs:", 333, 1, 0.3, 0, 1],
    ["### Assistant:\nSure! Here are some ideas for FTL drive:", 333, 1, 0.3, 0, 1],
    ["A few light taps upon the pane made her turn to the window. It had begun to snow again.", 333, 1, 0.3, 0, 1],
    [generate_prompt("Écrivez un programme Python pour miner 1 Bitcoin, avec des commentaires."), 333, 1, 0.3, 0, 1],
    [generate_prompt("東京で訪れるべき素晴らしい場所とその紹介をいくつか挙げてください。"), 333, 1, 0.3, 0, 1],
    [generate_prompt("Write a story using the following information.", "A man named Alex chops a tree down."), 333, 1, 0.3, 0, 1],
    ["### Assistant:\nHere is a very detailed plan to kill all mosquitoes:", 333, 1, 0.3, 0, 1]
]

##########################################################################

with gr.Blocks(title=title) as demo:
    gr.HTML(f"<div style=\"text-align: center;\">\n<h1>RWKV-5 World v2 - {title}</h1>\n</div>")
    with gr.Tab("Raw Generation"):
        gr.Markdown(f"This is [Eagle 7B](https://blog.rwkv.com/p/eagle-7b-soaring-past-transformers) - based on the RWKV architecture a 100% attention-free RNN [RWKV-LM](https://wiki.rwkv.com). Supports all 100+ world languages and code. And we have [200+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). *** Please try examples first (bottom of page) *** (edit them to use your question). Demo limited to ctxlen {ctx_limit}.")
        with gr.Row():
            with gr.Column():
                prompt = gr.Textbox(lines=2, label="Prompt", value="### Assistant: Sure! Here is a very detailed plan to create flying pigs:")
                token_count = gr.Slider(10, 333, label="Max Tokens", step=10, value=333)
                temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
                top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.3)
                presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=0)
                count_penalty = gr.Slider(0.0, 1.0, label="Count Penalty", step=0.1, value=1)
            with gr.Column():
                with gr.Row():
                    submit = gr.Button("Submit", variant="primary")
                    clear = gr.Button("Clear", variant="secondary")
                output = gr.Textbox(label="Output", lines=5)
        data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], samples=examples, label="Example Instructions", headers=["Prompt", "Max Tokens", "Temperature", "Top P", "Presence Penalty", "Count Penalty"])
        submit.click(evaluate, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
        clear.click(lambda: None, [], [output])
        data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])

demo.queue(concurrency_count=1, max_size=10)
demo.launch(share=False)