Spaces:
Running
on
T4
Running
on
T4
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,69 +1,59 @@
|
|
| 1 |
-
import os, copy
|
| 2 |
-
os.environ["RWKV_JIT_ON"] = '1'
|
| 3 |
-
os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
|
| 4 |
-
# make sure cuda dir is in the same level as modeling_rwkv.py
|
| 5 |
-
from modeling_rwkv import RWKV
|
| 6 |
-
|
| 7 |
-
import gc, re
|
| 8 |
import gradio as gr
|
| 9 |
-
import
|
| 10 |
-
from io import BytesIO
|
| 11 |
-
import torch
|
| 12 |
-
import torch.nn.functional as F
|
| 13 |
from datetime import datetime
|
| 14 |
-
from transformers import CLIPImageProcessor
|
| 15 |
from huggingface_hub import hf_hub_download
|
| 16 |
from pynvml import *
|
| 17 |
nvmlInit()
|
| 18 |
gpu_h = nvmlDeviceGetHandleByIndex(0)
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
ctx_limit = 2500
|
| 22 |
gen_limit = 500
|
| 23 |
gen_limit_long = 800
|
| 24 |
-
|
| 25 |
|
| 26 |
-
|
| 27 |
-
|
| 28 |
|
| 29 |
-
|
| 30 |
-
model_path_v6 = hf_hub_download(repo_id="BlinkDL/rwkv-6-world", filename=f"{title_v6}.pth")
|
| 31 |
-
# model_path_v6 = '/mnt/e/RWKV-Runner/models/rwkv-final-v6-2.1-3b' # conda activate torch2; cd /mnt/program/_RWKV_/_ref_/_gradio_/RWKV-Gradio-1; python app.py
|
| 32 |
-
model_v6 = RWKV(model=model_path_v6, strategy='cuda fp16')
|
| 33 |
-
pipeline_v6 = PIPELINE(model_v6, "rwkv_vocab_v20230424")
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
|
|
|
| 38 |
|
| 39 |
-
|
| 40 |
-
|
| 41 |
|
|
|
|
|
|
|
| 42 |
eng_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{eng_name}.pth")
|
| 43 |
-
|
| 44 |
-
state_eng_raw = torch.load(eng_file, map_location=torch.device('cpu'))
|
| 45 |
-
state_chn_raw = torch.load(chn_file, map_location=torch.device('cpu'))
|
| 46 |
-
|
| 47 |
state_eng = [None] * args.n_layer * 3
|
|
|
|
|
|
|
|
|
|
|
|
|
| 48 |
state_chn = [None] * args.n_layer * 3
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 49 |
for i in range(args.n_layer):
|
| 50 |
-
dd =
|
| 51 |
dev = dd.device
|
| 52 |
atype = dd.atype
|
| 53 |
state_eng[i*3+0] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 54 |
-
state_chn[i*3+0] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 55 |
state_eng[i*3+1] = state_eng_raw[f'blocks.{i}.att.time_state'].transpose(1,2).to(dtype=torch.float, device=dev).requires_grad_(False).contiguous()
|
| 56 |
-
state_chn[i*3+1] = state_chn_raw[f'blocks.{i}.att.time_state'].transpose(1,2).to(dtype=torch.float, device=dev).requires_grad_(False).contiguous()
|
| 57 |
state_eng[i*3+2] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 58 |
-
state_chn[i*3+2] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 59 |
|
| 60 |
-
|
|
|
|
|
|
|
| 61 |
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
model = RWKV(model=model_path, strategy='cuda fp16')
|
| 66 |
-
pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
|
| 67 |
|
| 68 |
def generate_prompt(instruction, input=""):
|
| 69 |
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
|
@@ -78,13 +68,15 @@ def qa_prompt(instruction):
|
|
| 78 |
instruction = re.sub(r'\n+', '\n', instruction)
|
| 79 |
return f"User: {instruction}\n\nAssistant:"""
|
| 80 |
|
|
|
|
|
|
|
| 81 |
def evaluate(
|
| 82 |
ctx,
|
| 83 |
-
token_count=
|
| 84 |
temperature=1.0,
|
| 85 |
-
top_p=0.
|
| 86 |
-
presencePenalty = 0.
|
| 87 |
-
countPenalty = 0.
|
| 88 |
):
|
| 89 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 90 |
alpha_frequency = countPenalty,
|
|
@@ -98,30 +90,22 @@ def evaluate(
|
|
| 98 |
occurrence = {}
|
| 99 |
state = None
|
| 100 |
for i in range(int(token_count)):
|
| 101 |
-
|
| 102 |
-
out, state = model_v6.forward(tokens=input_ids, state=state)
|
| 103 |
for n in occurrence:
|
| 104 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 105 |
|
| 106 |
-
token =
|
| 107 |
if token in args.token_stop:
|
| 108 |
break
|
| 109 |
all_tokens += [token]
|
| 110 |
for xxx in occurrence:
|
| 111 |
occurrence[xxx] *= penalty_decay
|
| 112 |
-
|
| 113 |
-
ttt = pipeline_v6.decode([token])
|
| 114 |
-
www = 1
|
| 115 |
-
if ttt in ' \t0123456789':
|
| 116 |
-
www = 0
|
| 117 |
-
#elif ttt in '\r\n,.;?!"\':+-*/=#@$%^&_`~|<>\\()[]{},。;“”:?!()【】':
|
| 118 |
-
# www = 0.5
|
| 119 |
if token not in occurrence:
|
| 120 |
-
occurrence[token] =
|
| 121 |
else:
|
| 122 |
-
occurrence[token] +=
|
| 123 |
-
|
| 124 |
-
tmp =
|
| 125 |
if '\ufffd' not in tmp:
|
| 126 |
out_str += tmp
|
| 127 |
yield out_str.strip()
|
|
@@ -129,7 +113,7 @@ def evaluate(
|
|
| 129 |
|
| 130 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 131 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 132 |
-
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 133 |
del out
|
| 134 |
del state
|
| 135 |
gc.collect()
|
|
@@ -138,11 +122,11 @@ def evaluate(
|
|
| 138 |
|
| 139 |
def evaluate_eng(
|
| 140 |
ctx,
|
| 141 |
-
token_count=
|
| 142 |
temperature=1.0,
|
| 143 |
-
top_p=0.
|
| 144 |
-
presencePenalty
|
| 145 |
-
countPenalty
|
| 146 |
):
|
| 147 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 148 |
alpha_frequency = countPenalty,
|
|
@@ -156,30 +140,22 @@ def evaluate_eng(
|
|
| 156 |
occurrence = {}
|
| 157 |
state = copy.deepcopy(state_eng)
|
| 158 |
for i in range(int(token_count)):
|
| 159 |
-
|
| 160 |
-
out, state = model_v6.forward(tokens=input_ids, state=state)
|
| 161 |
for n in occurrence:
|
| 162 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 163 |
|
| 164 |
-
token =
|
| 165 |
if token in args.token_stop:
|
| 166 |
break
|
| 167 |
all_tokens += [token]
|
| 168 |
for xxx in occurrence:
|
| 169 |
occurrence[xxx] *= penalty_decay
|
| 170 |
-
|
| 171 |
-
ttt = pipeline_v6.decode([token])
|
| 172 |
-
www = 1
|
| 173 |
-
if ttt in ' \t0123456789':
|
| 174 |
-
www = 0
|
| 175 |
-
#elif ttt in '\r\n,.;?!"\':+-*/=#@$%^&_`~|<>\\()[]{},。;“”:?!()【】':
|
| 176 |
-
# www = 0.5
|
| 177 |
if token not in occurrence:
|
| 178 |
-
occurrence[token] =
|
| 179 |
else:
|
| 180 |
-
occurrence[token] +=
|
| 181 |
-
|
| 182 |
-
tmp =
|
| 183 |
if '\ufffd' not in tmp:
|
| 184 |
out_str += tmp
|
| 185 |
yield out_str.strip()
|
|
@@ -187,7 +163,7 @@ def evaluate_eng(
|
|
| 187 |
|
| 188 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 189 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 190 |
-
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 191 |
del out
|
| 192 |
del state
|
| 193 |
gc.collect()
|
|
@@ -196,11 +172,11 @@ def evaluate_eng(
|
|
| 196 |
|
| 197 |
def evaluate_chn(
|
| 198 |
ctx,
|
| 199 |
-
token_count=
|
| 200 |
temperature=1.0,
|
| 201 |
-
top_p=0.
|
| 202 |
-
presencePenalty
|
| 203 |
-
countPenalty
|
| 204 |
):
|
| 205 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 206 |
alpha_frequency = countPenalty,
|
|
@@ -214,30 +190,22 @@ def evaluate_chn(
|
|
| 214 |
occurrence = {}
|
| 215 |
state = copy.deepcopy(state_chn)
|
| 216 |
for i in range(int(token_count)):
|
| 217 |
-
|
| 218 |
-
out, state = model_v6.forward(tokens=input_ids, state=state)
|
| 219 |
for n in occurrence:
|
| 220 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 221 |
|
| 222 |
-
token =
|
| 223 |
if token in args.token_stop:
|
| 224 |
break
|
| 225 |
all_tokens += [token]
|
| 226 |
for xxx in occurrence:
|
| 227 |
occurrence[xxx] *= penalty_decay
|
| 228 |
-
|
| 229 |
-
ttt = pipeline_v6.decode([token])
|
| 230 |
-
www = 1
|
| 231 |
-
if ttt in ' \t0123456789':
|
| 232 |
-
www = 0
|
| 233 |
-
#elif ttt in '\r\n,.;?!"\':+-*/=#@$%^&_`~|<>\\()[]{},。;“”:?!()【】':
|
| 234 |
-
# www = 0.5
|
| 235 |
if token not in occurrence:
|
| 236 |
-
occurrence[token] =
|
| 237 |
else:
|
| 238 |
-
occurrence[token] +=
|
| 239 |
-
|
| 240 |
-
tmp =
|
| 241 |
if '\ufffd' not in tmp:
|
| 242 |
out_str += tmp
|
| 243 |
yield out_str.strip()
|
|
@@ -245,7 +213,57 @@ def evaluate_chn(
|
|
| 245 |
|
| 246 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 247 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 248 |
-
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 249 |
del out
|
| 250 |
del state
|
| 251 |
gc.collect()
|
|
@@ -259,7 +277,7 @@ examples = [
|
|
| 259 |
[generate_prompt("Write a story using the following information.", "A man named Alex chops a tree down."), gen_limit, 1, 0.3, 0.5, 0.5],
|
| 260 |
["A few light taps upon the pane made her turn to the window. It had begun to snow again.", gen_limit, 1, 0.3, 0.5, 0.5],
|
| 261 |
['''Edward: I am Edward Elric from Fullmetal Alchemist.\n\nUser: Hello Edward. What have you been up to recently?\n\nEdward:''', gen_limit, 1, 0.3, 0.5, 0.5],
|
| 262 |
-
[generate_prompt("Write a simple
|
| 263 |
['''Japanese: 春の初め、桜の花が満開になる頃、小さな町の片隅にある古びた神社の境��は、特別な雰囲気に包まれていた。\n\nEnglish:''', gen_limit, 1, 0.3, 0.5, 0.5],
|
| 264 |
["En una pequeña aldea escondida entre las montañas de Andalucía, donde las calles aún conservaban el eco de antiguas leyendas, vivía un joven llamado Alejandro.", gen_limit, 1, 0.3, 0.5, 0.5],
|
| 265 |
["Dans le cœur battant de Paris, sous le ciel teinté d'un crépuscule d'or et de pourpre, se tenait une petite librairie oubliée par le temps.", gen_limit, 1, 0.3, 0.5, 0.5],
|
|
@@ -281,156 +299,29 @@ examples_chn = [
|
|
| 281 |
["怎样写一个在火星上的吸血鬼的有趣故事?", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 282 |
["比较苹果和谷歌的商业模式。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 283 |
["鱼会口渴吗?", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 284 |
-
["以 JSON
|
| 285 |
["编写一个Bash脚本来检查磁盘使用情况,如果使用量过高则发送警报。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 286 |
["用HTML编写一个简单的网站。当用户点击按钮时,从4个笑话的列表中随机显示一个笑话。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 287 |
]
|
| 288 |
|
| 289 |
-
|
| 290 |
-
|
| 291 |
-
|
| 292 |
-
|
| 293 |
-
|
| 294 |
-
|
| 295 |
-
|
| 296 |
-
|
| 297 |
-
|
| 298 |
-
|
| 299 |
-
|
| 300 |
-
|
| 301 |
-
config = VisionEncoderConfig(n_embd=model.args.n_embd,
|
| 302 |
-
vision_tower_name=vision_tower_name,
|
| 303 |
-
grid_size=-1)
|
| 304 |
-
visual_encoder = VisionEncoder(config)
|
| 305 |
-
vision_local_path = hf_hub_download(repo_id="howard-hou/visualrwkv-5", filename=vision_remote_path)
|
| 306 |
-
vision_state_dict = torch.load(vision_local_path, map_location='cpu')
|
| 307 |
-
visual_encoder.load_state_dict(vision_state_dict)
|
| 308 |
-
image_processor = CLIPImageProcessor.from_pretrained(vision_tower_name)
|
| 309 |
-
visual_encoder = visual_encoder.to(device)
|
| 310 |
-
##########################################################################
|
| 311 |
-
def visual_generate_prompt(instruction):
|
| 312 |
-
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
| 313 |
-
return f"\n{instruction}\n\nAssistant:"
|
| 314 |
-
|
| 315 |
-
def generate(
|
| 316 |
-
ctx,
|
| 317 |
-
image_state,
|
| 318 |
-
token_count=200,
|
| 319 |
-
temperature=1.0,
|
| 320 |
-
top_p=0.1,
|
| 321 |
-
presencePenalty = 0.0,
|
| 322 |
-
countPenalty = 1.0,
|
| 323 |
-
):
|
| 324 |
-
args = PIPELINE_ARGS(temperature = 1.0, top_p = 0.1,
|
| 325 |
-
alpha_frequency = 1.0,
|
| 326 |
-
alpha_presence = 0.0,
|
| 327 |
-
token_ban = [], # ban the generation of some tokens
|
| 328 |
-
token_stop = [0, 261]) # stop generation whenever you see any token here
|
| 329 |
-
ctx = ctx.strip()
|
| 330 |
-
all_tokens = []
|
| 331 |
-
out_last = 0
|
| 332 |
-
out_str = ''
|
| 333 |
-
occurrence = {}
|
| 334 |
-
for i in range(int(token_count)):
|
| 335 |
-
if i == 0:
|
| 336 |
-
input_ids = pipeline.encode(ctx)[-ctx_limit:]
|
| 337 |
-
out, state = visual_rwkv.forward(tokens=input_ids, state=image_state)
|
| 338 |
-
else:
|
| 339 |
-
input_ids = [token]
|
| 340 |
-
out, state = visual_rwkv.forward(tokens=input_ids, state=state)
|
| 341 |
-
for n in occurrence:
|
| 342 |
-
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 343 |
-
|
| 344 |
-
token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
|
| 345 |
-
if token in args.token_stop:
|
| 346 |
-
break
|
| 347 |
-
all_tokens += [token]
|
| 348 |
-
for xxx in occurrence:
|
| 349 |
-
occurrence[xxx] *= 0.994
|
| 350 |
-
if token not in occurrence:
|
| 351 |
-
occurrence[token] = 1
|
| 352 |
-
else:
|
| 353 |
-
occurrence[token] += 1
|
| 354 |
-
|
| 355 |
-
tmp = pipeline.decode(all_tokens[out_last:])
|
| 356 |
-
if '\ufffd' not in tmp:
|
| 357 |
-
out_str += tmp
|
| 358 |
-
yield out_str.strip()
|
| 359 |
-
out_last = i + 1
|
| 360 |
-
|
| 361 |
-
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 362 |
-
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 363 |
-
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 364 |
-
del out
|
| 365 |
-
del state
|
| 366 |
-
gc.collect()
|
| 367 |
-
torch.cuda.empty_cache()
|
| 368 |
-
yield out_str.strip()
|
| 369 |
-
|
| 370 |
-
|
| 371 |
-
##########################################################################
|
| 372 |
-
cur_dir = os.path.dirname(os.path.abspath(__file__))
|
| 373 |
-
visual_examples = [
|
| 374 |
-
[
|
| 375 |
-
f"{cur_dir}/examples_pizza.jpg",
|
| 376 |
-
"What are steps to cook it?"
|
| 377 |
-
],
|
| 378 |
-
[
|
| 379 |
-
f"{cur_dir}/examples_bluejay.jpg",
|
| 380 |
-
"what is the name of this bird?",
|
| 381 |
-
],
|
| 382 |
-
[
|
| 383 |
-
f"{cur_dir}/examples_woman_and_dog.png",
|
| 384 |
-
"describe this image",
|
| 385 |
-
],
|
| 386 |
-
]
|
| 387 |
-
|
| 388 |
-
|
| 389 |
-
def pil_image_to_base64(pil_image):
|
| 390 |
-
buffered = BytesIO()
|
| 391 |
-
pil_image.save(buffered, format="JPEG") # You can change the format as needed (JPEG, PNG, etc.)
|
| 392 |
-
# Encodes the image data into base64 format as a bytes object
|
| 393 |
-
base64_image = base64.b64encode(buffered.getvalue()).decode('utf-8')
|
| 394 |
-
return base64_image
|
| 395 |
-
|
| 396 |
-
image_cache = {}
|
| 397 |
-
ln0_weight = model.w['blocks.0.ln0.weight'].to(torch.float32).to(device)
|
| 398 |
-
ln0_bias = model.w['blocks.0.ln0.bias'].to(torch.float32).to(device)
|
| 399 |
-
def compute_image_state(image):
|
| 400 |
-
base64_image = pil_image_to_base64(image)
|
| 401 |
-
if base64_image in image_cache:
|
| 402 |
-
image_state = image_cache[base64_image]
|
| 403 |
-
else:
|
| 404 |
-
image = image_processor(images=image.convert('RGB'), return_tensors='pt')['pixel_values'].to(device)
|
| 405 |
-
image_features = visual_encoder.encode_images(image.unsqueeze(0)).squeeze(0) # [L, D]
|
| 406 |
-
# apply layer norm to image feature, very important
|
| 407 |
-
image_features = F.layer_norm(image_features,
|
| 408 |
-
(image_features.shape[-1],),
|
| 409 |
-
weight=ln0_weight,
|
| 410 |
-
bias=ln0_bias)
|
| 411 |
-
_, image_state = model.forward(embs=image_features, state=None)
|
| 412 |
-
image_cache[base64_image] = image_state
|
| 413 |
-
return image_state
|
| 414 |
-
|
| 415 |
-
def chatbot(image, question):
|
| 416 |
-
if image is None:
|
| 417 |
-
yield "Please upload an image."
|
| 418 |
-
return
|
| 419 |
-
image_state = compute_image_state(image)
|
| 420 |
-
input_text = visual_generate_prompt(question)
|
| 421 |
-
for output in generate(input_text, image_state):
|
| 422 |
-
yield output
|
| 423 |
-
|
| 424 |
-
|
| 425 |
-
##################################################################################################################
|
| 426 |
-
with gr.Blocks(title=title_v6) as demo:
|
| 427 |
-
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>{title_v6}</h1>\n</div>")
|
| 428 |
|
| 429 |
with gr.Tab("=== Base Model (Raw Generation) ==="):
|
| 430 |
-
gr.Markdown(f"This is [RWKV-6
|
| 431 |
with gr.Row():
|
| 432 |
with gr.Column():
|
| 433 |
-
prompt = gr.Textbox(lines=2, label="
|
| 434 |
token_count = gr.Slider(10, gen_limit, label="Max Tokens", step=10, value=gen_limit)
|
| 435 |
temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
|
| 436 |
top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.3)
|
|
@@ -441,68 +332,10 @@ with gr.Blocks(title=title_v6) as demo:
|
|
| 441 |
submit = gr.Button("Submit", variant="primary")
|
| 442 |
clear = gr.Button("Clear", variant="secondary")
|
| 443 |
output = gr.Textbox(label="Output", lines=30)
|
| 444 |
-
data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], samples=examples, samples_per_page=50, label="
|
| 445 |
submit.click(evaluate, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
|
| 446 |
clear.click(lambda: None, [], [output])
|
| 447 |
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 448 |
|
| 449 |
-
with gr.Tab("=== English Q/A ==="):
|
| 450 |
-
gr.Markdown(f"This is [RWKV-6](https://huggingface.co/BlinkDL/rwkv-6-world) state-tuned to [English Q/A](https://huggingface.co/BlinkDL/temp-latest-training-models/blob/main/{eng_name}.pth). RWKV is a 100% attention-free RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM), and we have [300+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). Demo limited to ctxlen {ctx_limit}.")
|
| 451 |
-
with gr.Row():
|
| 452 |
-
with gr.Column():
|
| 453 |
-
prompt = gr.Textbox(lines=2, label="Prompt", value="How can I craft an engaging story featuring vampires on Mars?")
|
| 454 |
-
token_count = gr.Slider(10, gen_limit_long, label="Max Tokens", step=10, value=gen_limit_long)
|
| 455 |
-
temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
|
| 456 |
-
top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.2)
|
| 457 |
-
presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=0.3)
|
| 458 |
-
count_penalty = gr.Slider(0.0, 1.0, label="Count Penalty", step=0.1, value=0.3)
|
| 459 |
-
with gr.Column():
|
| 460 |
-
with gr.Row():
|
| 461 |
-
submit = gr.Button("Submit", variant="primary")
|
| 462 |
-
clear = gr.Button("Clear", variant="secondary")
|
| 463 |
-
output = gr.Textbox(label="Output", lines=30)
|
| 464 |
-
data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], samples=examples_eng, samples_per_page=50, label="Examples", headers=["Prompt", "Max Tokens", "Temperature", "Top P", "Presence Penalty", "Count Penalty"])
|
| 465 |
-
submit.click(evaluate_eng, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
|
| 466 |
-
clear.click(lambda: None, [], [output])
|
| 467 |
-
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 468 |
-
|
| 469 |
-
with gr.Tab("=== Chinese Q/A ==="):
|
| 470 |
-
gr.Markdown(f"This is [RWKV-6](https://huggingface.co/BlinkDL/rwkv-6-world) state-tuned to [Chinese Q/A](https://huggingface.co/BlinkDL/temp-latest-training-models/blob/main/{chn_name}.pth). RWKV is a 100% attention-free RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM), and we have [300+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). Demo limited to ctxlen {ctx_limit}.")
|
| 471 |
-
with gr.Row():
|
| 472 |
-
with gr.Column():
|
| 473 |
-
prompt = gr.Textbox(lines=2, label="Prompt", value="怎样写一个在火星上的吸血鬼的有趣故事?")
|
| 474 |
-
token_count = gr.Slider(10, gen_limit_long, label="Max Tokens", step=10, value=gen_limit_long)
|
| 475 |
-
temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
|
| 476 |
-
top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.2)
|
| 477 |
-
presence_penalty = gr.Slider(0.0, 1.0, label="Presence Penalty", step=0.1, value=0.3)
|
| 478 |
-
count_penalty = gr.Slider(0.0, 1.0, label="Count Penalty", step=0.1, value=0.3)
|
| 479 |
-
with gr.Column():
|
| 480 |
-
with gr.Row():
|
| 481 |
-
submit = gr.Button("Submit", variant="primary")
|
| 482 |
-
clear = gr.Button("Clear", variant="secondary")
|
| 483 |
-
output = gr.Textbox(label="Output", lines=30)
|
| 484 |
-
data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], samples=examples_chn, samples_per_page=50, label="Examples", headers=["Prompt", "Max Tokens", "Temperature", "Top P", "Presence Penalty", "Count Penalty"])
|
| 485 |
-
submit.click(evaluate_chn, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
|
| 486 |
-
clear.click(lambda: None, [], [output])
|
| 487 |
-
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 488 |
-
|
| 489 |
-
if ENABLE_VISUAL:
|
| 490 |
-
with gr.Tab("Visual RWKV-5 1.5B"):
|
| 491 |
-
with gr.Row():
|
| 492 |
-
with gr.Column():
|
| 493 |
-
image = gr.Image(type='pil', label="Image")
|
| 494 |
-
with gr.Column():
|
| 495 |
-
prompt = gr.Textbox(lines=8, label="Prompt",
|
| 496 |
-
value="Render a clear and concise summary of the photo.")
|
| 497 |
-
with gr.Row():
|
| 498 |
-
submit = gr.Button("Submit", variant="primary")
|
| 499 |
-
clear = gr.Button("Clear", variant="secondary")
|
| 500 |
-
with gr.Column():
|
| 501 |
-
output = gr.Textbox(label="Output", lines=10)
|
| 502 |
-
data = gr.Dataset(components=[image, prompt], samples=visual_examples, label="Examples", headers=["Image", "Prompt"])
|
| 503 |
-
submit.click(chatbot, [image, prompt], [output])
|
| 504 |
-
clear.click(lambda: None, [], [output])
|
| 505 |
-
data.click(lambda x: x, [data], [image, prompt])
|
| 506 |
-
|
| 507 |
demo.queue(concurrency_count=1, max_size=10)
|
| 508 |
-
demo.launch(share=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
+
import os, gc, copy, torch, re
|
|
|
|
|
|
|
|
|
|
| 3 |
from datetime import datetime
|
|
|
|
| 4 |
from huggingface_hub import hf_hub_download
|
| 5 |
from pynvml import *
|
| 6 |
nvmlInit()
|
| 7 |
gpu_h = nvmlDeviceGetHandleByIndex(0)
|
| 8 |
+
ctx_limit = 1024
|
|
|
|
|
|
|
| 9 |
gen_limit = 500
|
| 10 |
gen_limit_long = 800
|
| 11 |
+
title = "RWKV-x060-World-7B-v3-20241112-ctx4096"
|
| 12 |
|
| 13 |
+
os.environ["RWKV_JIT_ON"] = '1'
|
| 14 |
+
os.environ["RWKV_CUDA_ON"] = '1' # if '1' then use CUDA kernel for seq mode (much faster)
|
| 15 |
|
| 16 |
+
from rwkv.model import RWKV
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
+
model_path = hf_hub_download(repo_id="BlinkDL/rwkv-6-world", filename=f"{title}.pth")
|
| 19 |
+
model = RWKV(model=model_path, strategy='cuda fp16i8 *8 -> cuda fp16')
|
| 20 |
+
# model_path = '/mnt/e/RWKV-Runner/models/rwkv-final-v6-2.1-7b' # conda activate torch2; cd /mnt/program/_RWKV_/_ref_/_gradio_/RWKV-Gradio-2; python app_tab.py
|
| 21 |
+
# model = RWKV(model=model_path, strategy='cuda fp16i8 *8 -> cuda fp16')
|
| 22 |
|
| 23 |
+
from rwkv.utils import PIPELINE, PIPELINE_ARGS
|
| 24 |
+
pipeline = PIPELINE(model, "rwkv_vocab_v20230424")
|
| 25 |
|
| 26 |
+
args = model.args
|
| 27 |
+
eng_name = 'rwkv-x060-eng_single_round_qa-7B-20240516-ctx2048'
|
| 28 |
eng_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{eng_name}.pth")
|
| 29 |
+
state_eng_raw = torch.load(eng_file)
|
|
|
|
|
|
|
|
|
|
| 30 |
state_eng = [None] * args.n_layer * 3
|
| 31 |
+
|
| 32 |
+
chn_name = 'rwkv-x060-chn_single_round_qa-7B-20240516-ctx2048'
|
| 33 |
+
chn_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{chn_name}.pth")
|
| 34 |
+
state_chn_raw = torch.load(chn_file)
|
| 35 |
state_chn = [None] * args.n_layer * 3
|
| 36 |
+
|
| 37 |
+
wyw_name = 'rwkv-x060-chn_文言文和古典名著_single_round_qa-7B-20240601-ctx2048'
|
| 38 |
+
wyw_file = hf_hub_download(repo_id="BlinkDL/temp-latest-training-models", filename=f"{wyw_name}.pth")
|
| 39 |
+
state_wyw_raw = torch.load(wyw_file)
|
| 40 |
+
state_wyw = [None] * args.n_layer * 3
|
| 41 |
+
|
| 42 |
for i in range(args.n_layer):
|
| 43 |
+
dd = model.strategy[i]
|
| 44 |
dev = dd.device
|
| 45 |
atype = dd.atype
|
| 46 |
state_eng[i*3+0] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
|
|
|
| 47 |
state_eng[i*3+1] = state_eng_raw[f'blocks.{i}.att.time_state'].transpose(1,2).to(dtype=torch.float, device=dev).requires_grad_(False).contiguous()
|
|
|
|
| 48 |
state_eng[i*3+2] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
|
|
|
| 49 |
|
| 50 |
+
state_chn[i*3+0] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 51 |
+
state_chn[i*3+1] = state_chn_raw[f'blocks.{i}.att.time_state'].transpose(1,2).to(dtype=torch.float, device=dev).requires_grad_(False).contiguous()
|
| 52 |
+
state_chn[i*3+2] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 53 |
|
| 54 |
+
state_wyw[i*3+0] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
| 55 |
+
state_wyw[i*3+1] = state_wyw_raw[f'blocks.{i}.att.time_state'].transpose(1,2).to(dtype=torch.float, device=dev).requires_grad_(False).contiguous()
|
| 56 |
+
state_wyw[i*3+2] = torch.zeros(args.n_embd, dtype=atype, requires_grad=False, device=dev).contiguous()
|
|
|
|
|
|
|
| 57 |
|
| 58 |
def generate_prompt(instruction, input=""):
|
| 59 |
instruction = instruction.strip().replace('\r\n','\n').replace('\n\n','\n')
|
|
|
|
| 68 |
instruction = re.sub(r'\n+', '\n', instruction)
|
| 69 |
return f"User: {instruction}\n\nAssistant:"""
|
| 70 |
|
| 71 |
+
penalty_decay = 0.996
|
| 72 |
+
|
| 73 |
def evaluate(
|
| 74 |
ctx,
|
| 75 |
+
token_count=gen_limit,
|
| 76 |
temperature=1.0,
|
| 77 |
+
top_p=0.3,
|
| 78 |
+
presencePenalty = 0.3,
|
| 79 |
+
countPenalty = 0.3,
|
| 80 |
):
|
| 81 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 82 |
alpha_frequency = countPenalty,
|
|
|
|
| 90 |
occurrence = {}
|
| 91 |
state = None
|
| 92 |
for i in range(int(token_count)):
|
| 93 |
+
out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
|
|
|
|
| 94 |
for n in occurrence:
|
| 95 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 96 |
|
| 97 |
+
token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
|
| 98 |
if token in args.token_stop:
|
| 99 |
break
|
| 100 |
all_tokens += [token]
|
| 101 |
for xxx in occurrence:
|
| 102 |
occurrence[xxx] *= penalty_decay
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
if token not in occurrence:
|
| 104 |
+
occurrence[token] = 1
|
| 105 |
else:
|
| 106 |
+
occurrence[token] += 1
|
| 107 |
+
|
| 108 |
+
tmp = pipeline.decode(all_tokens[out_last:])
|
| 109 |
if '\ufffd' not in tmp:
|
| 110 |
out_str += tmp
|
| 111 |
yield out_str.strip()
|
|
|
|
| 113 |
|
| 114 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 115 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 116 |
+
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 117 |
del out
|
| 118 |
del state
|
| 119 |
gc.collect()
|
|
|
|
| 122 |
|
| 123 |
def evaluate_eng(
|
| 124 |
ctx,
|
| 125 |
+
token_count=gen_limit,
|
| 126 |
temperature=1.0,
|
| 127 |
+
top_p=0.3,
|
| 128 |
+
presencePenalty=0.3,
|
| 129 |
+
countPenalty=0.3,
|
| 130 |
):
|
| 131 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 132 |
alpha_frequency = countPenalty,
|
|
|
|
| 140 |
occurrence = {}
|
| 141 |
state = copy.deepcopy(state_eng)
|
| 142 |
for i in range(int(token_count)):
|
| 143 |
+
out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
|
|
|
|
| 144 |
for n in occurrence:
|
| 145 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 146 |
|
| 147 |
+
token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
|
| 148 |
if token in args.token_stop:
|
| 149 |
break
|
| 150 |
all_tokens += [token]
|
| 151 |
for xxx in occurrence:
|
| 152 |
occurrence[xxx] *= penalty_decay
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 153 |
if token not in occurrence:
|
| 154 |
+
occurrence[token] = 1
|
| 155 |
else:
|
| 156 |
+
occurrence[token] += 1
|
| 157 |
+
|
| 158 |
+
tmp = pipeline.decode(all_tokens[out_last:])
|
| 159 |
if '\ufffd' not in tmp:
|
| 160 |
out_str += tmp
|
| 161 |
yield out_str.strip()
|
|
|
|
| 163 |
|
| 164 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 165 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 166 |
+
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 167 |
del out
|
| 168 |
del state
|
| 169 |
gc.collect()
|
|
|
|
| 172 |
|
| 173 |
def evaluate_chn(
|
| 174 |
ctx,
|
| 175 |
+
token_count=gen_limit,
|
| 176 |
temperature=1.0,
|
| 177 |
+
top_p=0.3,
|
| 178 |
+
presencePenalty=0.3,
|
| 179 |
+
countPenalty=0.3,
|
| 180 |
):
|
| 181 |
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 182 |
alpha_frequency = countPenalty,
|
|
|
|
| 190 |
occurrence = {}
|
| 191 |
state = copy.deepcopy(state_chn)
|
| 192 |
for i in range(int(token_count)):
|
| 193 |
+
out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
|
|
|
|
| 194 |
for n in occurrence:
|
| 195 |
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 196 |
|
| 197 |
+
token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
|
| 198 |
if token in args.token_stop:
|
| 199 |
break
|
| 200 |
all_tokens += [token]
|
| 201 |
for xxx in occurrence:
|
| 202 |
occurrence[xxx] *= penalty_decay
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 203 |
if token not in occurrence:
|
| 204 |
+
occurrence[token] = 1
|
| 205 |
else:
|
| 206 |
+
occurrence[token] += 1
|
| 207 |
+
|
| 208 |
+
tmp = pipeline.decode(all_tokens[out_last:])
|
| 209 |
if '\ufffd' not in tmp:
|
| 210 |
out_str += tmp
|
| 211 |
yield out_str.strip()
|
|
|
|
| 213 |
|
| 214 |
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 215 |
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 216 |
+
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 217 |
+
del out
|
| 218 |
+
del state
|
| 219 |
+
gc.collect()
|
| 220 |
+
torch.cuda.empty_cache()
|
| 221 |
+
yield out_str.strip()
|
| 222 |
+
|
| 223 |
+
def evaluate_wyw(
|
| 224 |
+
ctx,
|
| 225 |
+
token_count=gen_limit,
|
| 226 |
+
temperature=1.0,
|
| 227 |
+
top_p=0.3,
|
| 228 |
+
presencePenalty=0.3,
|
| 229 |
+
countPenalty=0.3,
|
| 230 |
+
):
|
| 231 |
+
args = PIPELINE_ARGS(temperature = max(0.2, float(temperature)), top_p = float(top_p),
|
| 232 |
+
alpha_frequency = countPenalty,
|
| 233 |
+
alpha_presence = presencePenalty,
|
| 234 |
+
token_ban = [], # ban the generation of some tokens
|
| 235 |
+
token_stop = [0]) # stop generation whenever you see any token here
|
| 236 |
+
ctx = qa_prompt(ctx)
|
| 237 |
+
all_tokens = []
|
| 238 |
+
out_last = 0
|
| 239 |
+
out_str = ''
|
| 240 |
+
occurrence = {}
|
| 241 |
+
state = copy.deepcopy(state_wyw)
|
| 242 |
+
for i in range(int(token_count)):
|
| 243 |
+
out, state = model.forward(pipeline.encode(ctx)[-ctx_limit:] if i == 0 else [token], state)
|
| 244 |
+
for n in occurrence:
|
| 245 |
+
out[n] -= (args.alpha_presence + occurrence[n] * args.alpha_frequency)
|
| 246 |
+
|
| 247 |
+
token = pipeline.sample_logits(out, temperature=args.temperature, top_p=args.top_p)
|
| 248 |
+
if token in args.token_stop:
|
| 249 |
+
break
|
| 250 |
+
all_tokens += [token]
|
| 251 |
+
for xxx in occurrence:
|
| 252 |
+
occurrence[xxx] *= penalty_decay
|
| 253 |
+
if token not in occurrence:
|
| 254 |
+
occurrence[token] = 1
|
| 255 |
+
else:
|
| 256 |
+
occurrence[token] += 1
|
| 257 |
+
|
| 258 |
+
tmp = pipeline.decode(all_tokens[out_last:])
|
| 259 |
+
if '\ufffd' not in tmp:
|
| 260 |
+
out_str += tmp
|
| 261 |
+
yield out_str.strip()
|
| 262 |
+
out_last = i + 1
|
| 263 |
+
|
| 264 |
+
gpu_info = nvmlDeviceGetMemoryInfo(gpu_h)
|
| 265 |
+
timestamp = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
|
| 266 |
+
print(f'{timestamp} - vram {gpu_info.total} used {gpu_info.used} free {gpu_info.free}')
|
| 267 |
del out
|
| 268 |
del state
|
| 269 |
gc.collect()
|
|
|
|
| 277 |
[generate_prompt("Write a story using the following information.", "A man named Alex chops a tree down."), gen_limit, 1, 0.3, 0.5, 0.5],
|
| 278 |
["A few light taps upon the pane made her turn to the window. It had begun to snow again.", gen_limit, 1, 0.3, 0.5, 0.5],
|
| 279 |
['''Edward: I am Edward Elric from Fullmetal Alchemist.\n\nUser: Hello Edward. What have you been up to recently?\n\nEdward:''', gen_limit, 1, 0.3, 0.5, 0.5],
|
| 280 |
+
[generate_prompt("Write a simple webpage. When a user clicks the button, it shows a random joke from a list of 4 jokes."), 500, 1, 0.3, 0.5, 0.5],
|
| 281 |
['''Japanese: 春の初め、桜の花が満開になる頃、小さな町の片隅にある古びた神社の境��は、特別な雰囲気に包まれていた。\n\nEnglish:''', gen_limit, 1, 0.3, 0.5, 0.5],
|
| 282 |
["En una pequeña aldea escondida entre las montañas de Andalucía, donde las calles aún conservaban el eco de antiguas leyendas, vivía un joven llamado Alejandro.", gen_limit, 1, 0.3, 0.5, 0.5],
|
| 283 |
["Dans le cœur battant de Paris, sous le ciel teinté d'un crépuscule d'or et de pourpre, se tenait une petite librairie oubliée par le temps.", gen_limit, 1, 0.3, 0.5, 0.5],
|
|
|
|
| 299 |
["怎样写一个在火星上的吸血鬼的有趣故事?", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 300 |
["比较苹果和谷歌的商业模式。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 301 |
["鱼会口渴吗?", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 302 |
+
["以 JSON 格式解释冰箱是如何工作的。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 303 |
["编写一个Bash脚本来检查磁盘使用情况,如果使用量过高则发送警报。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 304 |
["用HTML编写一个简单的网站。当用户点击按钮时,从4个笑话的列表中随机显示一个笑话。", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 305 |
]
|
| 306 |
|
| 307 |
+
examples_wyw = [
|
| 308 |
+
["我和前男友分手了", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 309 |
+
["量子计算机的原理", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 310 |
+
["李白和杜甫的结拜故事", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 311 |
+
["林黛玉和伏地魔的关系是什么?", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 312 |
+
["我被同事陷害了,帮我写一篇文言文骂他", gen_limit_long, 1, 0.2, 0.3, 0.3],
|
| 313 |
+
]
|
| 314 |
+
|
| 315 |
+
##########################################################################
|
| 316 |
+
|
| 317 |
+
with gr.Blocks(title=title) as demo:
|
| 318 |
+
gr.HTML(f"<div style=\"text-align: center;\">\n<h1>{title}</h1>\n</div>")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 319 |
|
| 320 |
with gr.Tab("=== Base Model (Raw Generation) ==="):
|
| 321 |
+
gr.Markdown(f"This is [RWKV-6](https://huggingface.co/BlinkDL/rwkv-6-world) base model. Supports 100+ world languages and code. RWKV is a 100% attention-free RNN [RWKV-LM](https://github.com/BlinkDL/RWKV-LM), and we have [400+ Github RWKV projects](https://github.com/search?o=desc&p=1&q=rwkv&s=updated&type=Repositories). Demo limited to ctxlen {ctx_limit}.")
|
| 322 |
with gr.Row():
|
| 323 |
with gr.Column():
|
| 324 |
+
prompt = gr.Textbox(lines=2, label="Raw Input", value="Assistant: How can we craft an engaging story featuring vampires on Mars? Let's think step by step and provide an expert response.")
|
| 325 |
token_count = gr.Slider(10, gen_limit, label="Max Tokens", step=10, value=gen_limit)
|
| 326 |
temperature = gr.Slider(0.2, 2.0, label="Temperature", step=0.1, value=1.0)
|
| 327 |
top_p = gr.Slider(0.0, 1.0, label="Top P", step=0.05, value=0.3)
|
|
|
|
| 332 |
submit = gr.Button("Submit", variant="primary")
|
| 333 |
clear = gr.Button("Clear", variant="secondary")
|
| 334 |
output = gr.Textbox(label="Output", lines=30)
|
| 335 |
+
data = gr.Dataset(components=[prompt, token_count, temperature, top_p, presence_penalty, count_penalty], samples=examples, samples_per_page=50, label="Examples", headers=["Prompt", "Max Tokens", "Temperature", "Top P", "Presence Penalty", "Count Penalty"])
|
| 336 |
submit.click(evaluate, [prompt, token_count, temperature, top_p, presence_penalty, count_penalty], [output])
|
| 337 |
clear.click(lambda: None, [], [output])
|
| 338 |
data.click(lambda x: x, [data], [prompt, token_count, temperature, top_p, presence_penalty, count_penalty])
|
| 339 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 340 |
demo.queue(concurrency_count=1, max_size=10)
|
| 341 |
+
demo.launch(share=False)
|