Spaces:
Build error
Build error
Commit
·
59618c7
1
Parent(s):
f092f6f
Update app.py
Browse files
app.py
CHANGED
|
@@ -3,20 +3,42 @@ import os, gc, copy, torch
|
|
| 3 |
from datetime import datetime
|
| 4 |
from huggingface_hub import hf_hub_download
|
| 5 |
from pynvml import *
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
| 8 |
ctx_limit = 2000
|
| 9 |
title = "RWKV-5-World-1B5-v2-20231025-ctx4096"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
| 11 |
os.environ["RWKV_JIT_ON"] = '1'
|
| 12 |
os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
|
| 13 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
from rwkv.model import RWKV
|
| 15 |
-
model_path = hf_hub_download(repo_id="BlinkDL/rwkv-5-world", filename=f"{
|
| 16 |
-
model = RWKV(model=model_path, strategy=
|
| 17 |
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
| 18 |
pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
|
| 19 |
|
|
|
|
| 20 |
def generate_prompt(instruction, input=""):
|
| 21 |
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
| 22 |
input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
|
|
@@ -35,6 +57,7 @@ User: {instruction}
|
|
| 35 |
|
| 36 |
Assistant:"""
|
| 37 |
|
|
|
|
| 38 |
def evaluate(
|
| 39 |
ctx,
|
| 40 |
token_count=200,
|
|
@@ -84,6 +107,7 @@ def evaluate(
|
|
| 84 |
torch.cuda.empty_cache()
|
| 85 |
yield out_str.strip()
|
| 86 |
|
|
|
|
| 87 |
examples = [
|
| 88 |
["Assistant: Sure! Here is a very detailed plan to create flying pigs:", 333, 1, 0.3, 0, 1],
|
| 89 |
["Assistant: Sure! Here are some ideas for FTL drive:", 333, 1, 0.3, 0, 1],
|
|
@@ -108,6 +132,7 @@ Edward:''', 333, 1, 0.3, 0, 1],
|
|
| 108 |
|
| 109 |
##########################################################################
|
| 110 |
|
|
|
|
| 111 |
with gr.Blocks(title=title) as demo:
|
| 112 |
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>RWKV-5 World v2 - {title}</h1>\n</div>")
|
| 113 |
with gr.Tab("Raw Generation"):
|
|
@@ -130,5 +155,6 @@ with gr.Blocks(title=title) as demo:
|
|
| 130 |
clear.click(lambda: None, [], [output])
|
| 131 |
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 132 |
|
|
|
|
| 133 |
demo.queue(concurrency_count=1, max_size=10)
|
| 134 |
demo.launch(share=False)
|
|
|
|
| 3 |
from datetime import datetime
|
| 4 |
from huggingface_hub import hf_hub_download
|
| 5 |
from pynvml import *
|
| 6 |
+
|
| 7 |
+
# Flag to check if GPU is present
|
| 8 |
+
HAS_GPU = False
|
| 9 |
+
|
| 10 |
+
# Model title and context size limit
|
| 11 |
ctx_limit = 2000
|
| 12 |
title = "RWKV-5-World-1B5-v2-20231025-ctx4096"
|
| 13 |
+
model_file = "RWKV-5-World-1B5-v2-20231025-ctx4096"
|
| 14 |
+
|
| 15 |
+
# Get the GPU count
|
| 16 |
+
try:
|
| 17 |
+
nvmlInit()
|
| 18 |
+
GPU_COUNT = nvmlDeviceGetCount()
|
| 19 |
+
if GPU_COUNT > 0:
|
| 20 |
+
HAS_GPU = True
|
| 21 |
+
gpu_h = nvmlDeviceGetHandleByIndex(0)
|
| 22 |
+
except NVMLError as error:
|
| 23 |
+
print(error)
|
| 24 |
+
|
| 25 |
|
| 26 |
os.environ["RWKV_JIT_ON"] = '1'
|
| 27 |
os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
|
| 28 |
|
| 29 |
+
# Model strat to use
|
| 30 |
+
MODEL_STRAT="cuda fp16"
|
| 31 |
+
if HAS_GPU == False :
|
| 32 |
+
"cpu bf16"
|
| 33 |
+
|
| 34 |
+
# Load the model accordingly
|
| 35 |
from rwkv.model import RWKV
|
| 36 |
+
model_path = hf_hub_download(repo_id="BlinkDL/rwkv-5-world", filename=f"{model_file}.pth")
|
| 37 |
+
model = RWKV(model=model_path, strategy=MODEL_STRAT)
|
| 38 |
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
| 39 |
pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
|
| 40 |
|
| 41 |
+
# Prompt generation
|
| 42 |
def generate_prompt(instruction, input=""):
|
| 43 |
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
| 44 |
input = input.strip().replace('\r\n','\n').replace('\n\n','\n')
|
|
|
|
| 57 |
|
| 58 |
Assistant:"""
|
| 59 |
|
| 60 |
+
# Evaluation logic
|
| 61 |
def evaluate(
|
| 62 |
ctx,
|
| 63 |
token_count=200,
|
|
|
|
| 107 |
torch.cuda.empty_cache()
|
| 108 |
yield out_str.strip()
|
| 109 |
|
| 110 |
+
# Examples and gradio blocks
|
| 111 |
examples = [
|
| 112 |
["Assistant: Sure! Here is a very detailed plan to create flying pigs:", 333, 1, 0.3, 0, 1],
|
| 113 |
["Assistant: Sure! Here are some ideas for FTL drive:", 333, 1, 0.3, 0, 1],
|
|
|
|
| 132 |
|
| 133 |
##########################################################################
|
| 134 |
|
| 135 |
+
# Gradio blocks
|
| 136 |
with gr.Blocks(title=title) as demo:
|
| 137 |
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>RWKV-5 World v2 - {title}</h1>\n</div>")
|
| 138 |
with gr.Tab("Raw Generation"):
|
|
|
|
| 155 |
clear.click(lambda: None, [], [output])
|
| 156 |
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 157 |
|
| 158 |
+
# Gradio launch
|
| 159 |
demo.queue(concurrency_count=1, max_size=10)
|
| 160 |
demo.launch(share=False)
|