Spaces:
Running
Running
Jae-Won Chung
commited on
Commit
·
764dce6
1
Parent(s):
9f1c84b
Push `benchmark.py` from fix_stop_str
Browse files- scripts/benchmark.py +124 -138
scripts/benchmark.py
CHANGED
|
@@ -7,8 +7,8 @@ import json
|
|
| 7 |
import copy
|
| 8 |
import atexit
|
| 9 |
from typing import Generator, Literal, Iterable, Dict
|
|
|
|
| 10 |
|
| 11 |
-
import gc
|
| 12 |
import numpy as np
|
| 13 |
import tyro
|
| 14 |
import torch
|
|
@@ -16,6 +16,7 @@ import rich
|
|
| 16 |
from rich.table import Table
|
| 17 |
from fastchat.serve.inference import prepare_logits_processor
|
| 18 |
from fastchat.model.model_adapter import load_model, get_conversation_template
|
|
|
|
| 19 |
from zeus.monitor import ZeusMonitor
|
| 20 |
|
| 21 |
SYSTEM_PROMPTS = {
|
|
@@ -39,21 +40,20 @@ SYSTEM_PROMPTS = {
|
|
| 39 |
),
|
| 40 |
}
|
| 41 |
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
return False
|
| 48 |
|
| 49 |
@torch.inference_mode()
|
| 50 |
-
def
|
| 51 |
model,
|
| 52 |
tokenizer,
|
| 53 |
params: Dict,
|
| 54 |
device: str,
|
| 55 |
context_len: int = 2048,
|
| 56 |
-
):
|
| 57 |
# Read parameters
|
| 58 |
prompts = params["prompt"]
|
| 59 |
temperature = float(params.get("temperature", 1.0))
|
|
@@ -62,10 +62,16 @@ def generate_stream(
|
|
| 62 |
top_k = int(params.get("top_k", -1)) # -1 means disable
|
| 63 |
max_new_tokens = int(params.get("max_new_tokens", 256))
|
| 64 |
stop_str = params.get("stop", None)
|
| 65 |
-
stop_token_ids = params.get("stop_token_ids", None) or []
|
| 66 |
stop_token_ids.append(tokenizer.eos_token_id)
|
| 67 |
batch_size = len(prompts)
|
| 68 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
# left append prompts with eos to make all input prompts the same length
|
| 70 |
tokenizer.padding_side = "left"
|
| 71 |
tokenizer.pad_token = tokenizer.eos_token
|
|
@@ -75,15 +81,14 @@ def generate_stream(
|
|
| 75 |
)
|
| 76 |
|
| 77 |
input_ids = tokenizer(prompts, padding=True).input_ids
|
| 78 |
-
output_ids =
|
| 79 |
|
| 80 |
if model.config.is_encoder_decoder:
|
| 81 |
max_src_len = context_len
|
| 82 |
else: # truncate
|
| 83 |
-
max_src_len = context_len - max_new_tokens -
|
| 84 |
|
| 85 |
input_ids = [input_id[-max_src_len:] for input_id in input_ids]
|
| 86 |
-
input_len = len(input_ids[0])
|
| 87 |
|
| 88 |
if model.config.is_encoder_decoder:
|
| 89 |
encoder_output = model.encoder(
|
|
@@ -141,10 +146,10 @@ def generate_stream(
|
|
| 141 |
else:
|
| 142 |
last_token_logits = logits[:, -1, :]
|
| 143 |
|
| 144 |
-
|
| 145 |
-
|
| 146 |
-
|
| 147 |
-
|
| 148 |
if temperature < 1e-5 or top_p < 1e-8: # greedy
|
| 149 |
_, indices = torch.topk(last_token_logits, 2)
|
| 150 |
tokens = [[int(token) for token in query] for query in indices.tolist()]
|
|
@@ -152,81 +157,70 @@ def generate_stream(
|
|
| 152 |
probs = torch.softmax(last_token_logits, dim=-1)
|
| 153 |
indices = torch.multinomial(probs, num_samples=2)
|
| 154 |
tokens = [[int(token) for token in query] for query in indices.tolist()]
|
|
|
|
|
|
|
| 155 |
|
|
|
|
| 156 |
old_stopped = stopped
|
| 157 |
stopped = np.logical_or(old_stopped, np.array([True if token[0] in stop_token_ids else False for token in tokens]))
|
| 158 |
-
|
| 159 |
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
| 178 |
-
|
| 179 |
-
|
| 180 |
-
|
| 181 |
-
|
|
|
|
| 182 |
find_stop = pos_array != -1
|
| 183 |
-
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
| 188 |
-
|
| 189 |
-
|
| 190 |
-
|
| 191 |
-
|
| 192 |
-
|
| 193 |
-
else:
|
| 194 |
-
raise ValueError("Invalid stop field type.")
|
| 195 |
-
|
| 196 |
-
# Prevent yielding partial stop sequence
|
| 197 |
-
if not any(partially_stopped):
|
| 198 |
-
# indicates which request in batch stopped
|
| 199 |
-
different_indices = np.where(stopped != old_stopped)[0]
|
| 200 |
-
stop_length = np.array([(j, i+1) for j in different_indices])
|
| 201 |
-
yield {
|
| 202 |
-
"text": output,
|
| 203 |
-
"stop_length": stop_length,
|
| 204 |
-
}
|
| 205 |
|
| 206 |
if all(stopped):
|
| 207 |
break
|
| 208 |
|
| 209 |
-
|
| 210 |
if any(stopped) == False:
|
| 211 |
-
tmp_output_ids = [ids[input_len:] for ids in output_ids]
|
| 212 |
output = tokenizer.batch_decode(
|
| 213 |
-
|
| 214 |
skip_special_tokens=True,
|
| 215 |
spaces_between_special_tokens=False,
|
| 216 |
clean_up_tokenization_spaces=True,
|
| 217 |
)
|
| 218 |
-
stop_length = np.array([(i, max_new_tokens) for i in false_indices])
|
| 219 |
|
| 220 |
-
|
| 221 |
-
|
| 222 |
-
|
| 223 |
-
}
|
| 224 |
|
| 225 |
-
|
| 226 |
-
del past_key_values, out
|
| 227 |
-
gc.collect()
|
| 228 |
-
torch.cuda.empty_cache()
|
| 229 |
|
|
|
|
|
|
|
|
|
|
| 230 |
|
| 231 |
def main(
|
| 232 |
model_path: str,
|
|
@@ -347,108 +341,100 @@ def main(
|
|
| 347 |
"temperature": temperature,
|
| 348 |
"repitition_penalty": repitition_penalty,
|
| 349 |
"max_new_tokens": max_new_tokens,
|
|
|
|
| 350 |
},
|
| 351 |
config_json,
|
| 352 |
indent=4,
|
| 353 |
)
|
| 354 |
config_json.write("\n")
|
| 355 |
|
| 356 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 357 |
"""Yields a tuple of whether this is a warmup run and the input prompt."""
|
| 358 |
-
for _ in range(3
|
| 359 |
-
yield True, "Say something long and random. I don't care about the content."
|
| 360 |
-
|
| 361 |
-
|
| 362 |
-
|
|
|
|
|
|
|
| 363 |
|
| 364 |
# Warm up the GPU with some random prompts.
|
| 365 |
# Forward through all the prompts.
|
| 366 |
is_first = True
|
| 367 |
convs = []
|
| 368 |
prompts = []
|
| 369 |
-
data_iter = iter(dataloader(input_file))
|
| 370 |
-
|
| 371 |
-
|
| 372 |
-
while True:
|
| 373 |
-
try:
|
| 374 |
-
is_warmup, input_prompt = next(data_iter)
|
| 375 |
-
except StopIteration:
|
| 376 |
-
end_of_file = True # no more data
|
| 377 |
-
|
| 378 |
# Construct the input prompt.
|
| 379 |
-
|
| 380 |
conv = copy.deepcopy(conv_base)
|
| 381 |
-
conv.append_message(conv.roles[0],
|
| 382 |
conv.append_message(conv.roles[1], "")
|
| 383 |
prompt = conv.get_prompt()
|
| 384 |
prompts.append(prompt)
|
| 385 |
convs.append(conv)
|
| 386 |
-
|
| 387 |
gen_params["prompt"] = prompts
|
| 388 |
-
if end_of_file and len(prompts) == 0:
|
| 389 |
-
break
|
| 390 |
|
| 391 |
# Print input prompt.
|
| 392 |
for i in range(len(convs)):
|
| 393 |
console.print(f"\n[u cyan]{'Warmup ' if is_warmup else ''}Prompt[/u cyan](batch_{i}):")
|
| 394 |
console.print(prompts[i].strip() + "\n", markup=False)
|
| 395 |
|
| 396 |
-
# Generate the ouptut from the model.
|
| 397 |
-
output_stream = generate_stream(model, tokenizer, gen_params, device="cuda", context_len=2048)
|
| 398 |
-
output = {}
|
| 399 |
-
batch_token_len = {}
|
| 400 |
-
|
| 401 |
#################################################
|
| 402 |
# Inference and measurement zone!
|
| 403 |
#################################################
|
| 404 |
monitor.begin_window("inference")
|
| 405 |
-
|
| 406 |
-
stop_length = output["stop_length"]
|
| 407 |
-
for it in stop_length:
|
| 408 |
-
batch_token_len[it[0]] = it[1]
|
| 409 |
measurements = monitor.end_window("inference")
|
| 410 |
#################################################
|
| 411 |
-
|
| 412 |
-
|
| 413 |
-
output_text = output["text"]
|
| 414 |
-
if not is_warmup:
|
| 415 |
-
total_length = int(sum(batch_token_len.values())) # number of valid tokens
|
| 416 |
-
response_length = float(total_length) / len(convs)
|
| 417 |
-
latency = measurements.time
|
| 418 |
-
throughput = response_length / latency
|
| 419 |
-
energy = measurements.total_energy
|
| 420 |
-
output = {
|
| 421 |
-
"model": model_name_cleaned,
|
| 422 |
-
"batch": len(convs),
|
| 423 |
-
"throughput": throughput,
|
| 424 |
-
"response_length": response_length,
|
| 425 |
-
"latency": latency,
|
| 426 |
-
"energy": energy,
|
| 427 |
-
"input": [prompt.strip() for prompt in prompts],
|
| 428 |
-
"output": [output_text[i][:batch_token_len[i]].strip() for i in range(len(convs))],
|
| 429 |
-
}
|
| 430 |
-
output_str = json.dumps(output, indent=4)
|
| 431 |
if not is_warmup:
|
| 432 |
-
|
| 433 |
-
|
| 434 |
-
|
| 435 |
-
|
| 436 |
-
|
| 437 |
-
|
| 438 |
-
|
| 439 |
-
|
| 440 |
-
|
| 441 |
-
|
| 442 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 443 |
|
| 444 |
# Print measurement.
|
| 445 |
console.print(measurements)
|
| 446 |
convs = []
|
| 447 |
prompts = []
|
| 448 |
|
| 449 |
-
if end_of_file:
|
| 450 |
-
break
|
| 451 |
-
|
| 452 |
-
|
| 453 |
if __name__ == "__main__":
|
| 454 |
tyro.cli(main)
|
|
|
|
| 7 |
import copy
|
| 8 |
import atexit
|
| 9 |
from typing import Generator, Literal, Iterable, Dict
|
| 10 |
+
from dataclasses import dataclass
|
| 11 |
|
|
|
|
| 12 |
import numpy as np
|
| 13 |
import tyro
|
| 14 |
import torch
|
|
|
|
| 16 |
from rich.table import Table
|
| 17 |
from fastchat.serve.inference import prepare_logits_processor
|
| 18 |
from fastchat.model.model_adapter import load_model, get_conversation_template
|
| 19 |
+
from torch.utils.data import Dataset, DataLoader
|
| 20 |
from zeus.monitor import ZeusMonitor
|
| 21 |
|
| 22 |
SYSTEM_PROMPTS = {
|
|
|
|
| 40 |
),
|
| 41 |
}
|
| 42 |
|
| 43 |
+
@dataclass
|
| 44 |
+
class Output:
|
| 45 |
+
response_length: int
|
| 46 |
+
input: str
|
| 47 |
+
output: str
|
|
|
|
| 48 |
|
| 49 |
@torch.inference_mode()
|
| 50 |
+
def run_inference(
|
| 51 |
model,
|
| 52 |
tokenizer,
|
| 53 |
params: Dict,
|
| 54 |
device: str,
|
| 55 |
context_len: int = 2048,
|
| 56 |
+
) ->list[Output]:
|
| 57 |
# Read parameters
|
| 58 |
prompts = params["prompt"]
|
| 59 |
temperature = float(params.get("temperature", 1.0))
|
|
|
|
| 62 |
top_k = int(params.get("top_k", -1)) # -1 means disable
|
| 63 |
max_new_tokens = int(params.get("max_new_tokens", 256))
|
| 64 |
stop_str = params.get("stop", None)
|
| 65 |
+
stop_token_ids = list(params.get("stop_token_ids", None) or [])
|
| 66 |
stop_token_ids.append(tokenizer.eos_token_id)
|
| 67 |
batch_size = len(prompts)
|
| 68 |
|
| 69 |
+
empty_result = Output(response_length=-1, input="", output="")
|
| 70 |
+
result = []
|
| 71 |
+
for i, prompt in enumerate(prompts):
|
| 72 |
+
result.append(copy.deepcopy(empty_result))
|
| 73 |
+
result[i].input = prompt
|
| 74 |
+
|
| 75 |
# left append prompts with eos to make all input prompts the same length
|
| 76 |
tokenizer.padding_side = "left"
|
| 77 |
tokenizer.pad_token = tokenizer.eos_token
|
|
|
|
| 81 |
)
|
| 82 |
|
| 83 |
input_ids = tokenizer(prompts, padding=True).input_ids
|
| 84 |
+
output_ids = [[] for _ in range(batch_size)]
|
| 85 |
|
| 86 |
if model.config.is_encoder_decoder:
|
| 87 |
max_src_len = context_len
|
| 88 |
else: # truncate
|
| 89 |
+
max_src_len = context_len - max_new_tokens - 1
|
| 90 |
|
| 91 |
input_ids = [input_id[-max_src_len:] for input_id in input_ids]
|
|
|
|
| 92 |
|
| 93 |
if model.config.is_encoder_decoder:
|
| 94 |
encoder_output = model.encoder(
|
|
|
|
| 146 |
else:
|
| 147 |
last_token_logits = logits[:, -1, :]
|
| 148 |
|
| 149 |
+
# handle unexpected Nan issue for llama 2 7b chat
|
| 150 |
+
if torch.any(torch.isnan(last_token_logits)) == True:
|
| 151 |
+
return []
|
| 152 |
+
|
| 153 |
if temperature < 1e-5 or top_p < 1e-8: # greedy
|
| 154 |
_, indices = torch.topk(last_token_logits, 2)
|
| 155 |
tokens = [[int(token) for token in query] for query in indices.tolist()]
|
|
|
|
| 157 |
probs = torch.softmax(last_token_logits, dim=-1)
|
| 158 |
indices = torch.multinomial(probs, num_samples=2)
|
| 159 |
tokens = [[int(token) for token in query] for query in indices.tolist()]
|
| 160 |
+
|
| 161 |
+
output_ids = [ids + [token[0]] for ids, token in zip(output_ids, tokens)]
|
| 162 |
|
| 163 |
+
# deal with stop_token_ids
|
| 164 |
old_stopped = stopped
|
| 165 |
stopped = np.logical_or(old_stopped, np.array([True if token[0] in stop_token_ids else False for token in tokens]))
|
| 166 |
+
different_indices = np.where(stopped != old_stopped)[0]
|
| 167 |
|
| 168 |
+
rfind_start = 0
|
| 169 |
+
output = tokenizer.batch_decode(
|
| 170 |
+
output_ids,
|
| 171 |
+
skip_special_tokens=True,
|
| 172 |
+
spaces_between_special_tokens=False,
|
| 173 |
+
clean_up_tokenization_spaces=True,
|
| 174 |
+
)
|
| 175 |
+
output_np = np.array(output)
|
| 176 |
+
|
| 177 |
+
if different_indices.size > 0:
|
| 178 |
+
# here i but not i+1 is because the i+1 token generated is in stop_token_ids
|
| 179 |
+
for j in different_indices:
|
| 180 |
+
result[j].response_length = i
|
| 181 |
+
result[j].output = output[j]
|
| 182 |
+
|
| 183 |
+
# deal with stop_str
|
| 184 |
+
if stop_str:
|
| 185 |
+
if isinstance(stop_str, str):
|
| 186 |
+
pos_array = np.char.rfind(output_np, stop_str, rfind_start)
|
| 187 |
+
find_stop = pos_array != -1
|
| 188 |
+
elif isinstance(stop_str, Iterable):
|
| 189 |
+
for each_stop in stop_str:
|
| 190 |
+
pos_array = np.char.rfind(output_np, each_stop, rfind_start)
|
| 191 |
find_stop = pos_array != -1
|
| 192 |
+
else:
|
| 193 |
+
raise ValueError("Invalid stop field type.")
|
| 194 |
+
|
| 195 |
+
stop_str_indices = np.where(find_stop & ~stopped)[0]
|
| 196 |
+
if stop_str_indices.size > 0:
|
| 197 |
+
for j in stop_str_indices:
|
| 198 |
+
# TODO: find a elegant way to figure out the size of stop_str, here just suppose stop_str has one token
|
| 199 |
+
result[j].response_length = i
|
| 200 |
+
result[j].output = output[j][:pos_array[j]]
|
| 201 |
+
stopped[find_stop] = True
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 202 |
|
| 203 |
if all(stopped):
|
| 204 |
break
|
| 205 |
|
| 206 |
+
not_finish_indices = np.where(stopped == False)[0]
|
| 207 |
if any(stopped) == False:
|
|
|
|
| 208 |
output = tokenizer.batch_decode(
|
| 209 |
+
output_ids,
|
| 210 |
skip_special_tokens=True,
|
| 211 |
spaces_between_special_tokens=False,
|
| 212 |
clean_up_tokenization_spaces=True,
|
| 213 |
)
|
|
|
|
| 214 |
|
| 215 |
+
for j in not_finish_indices:
|
| 216 |
+
result[j].response_length = max_new_tokens
|
| 217 |
+
result[j].output = output[j]
|
|
|
|
| 218 |
|
| 219 |
+
return result
|
|
|
|
|
|
|
|
|
|
| 220 |
|
| 221 |
+
def write_error_to_file(filename, error_message):
|
| 222 |
+
with open(filename, 'a') as file:
|
| 223 |
+
file.write(error_message + '\n')
|
| 224 |
|
| 225 |
def main(
|
| 226 |
model_path: str,
|
|
|
|
| 341 |
"temperature": temperature,
|
| 342 |
"repitition_penalty": repitition_penalty,
|
| 343 |
"max_new_tokens": max_new_tokens,
|
| 344 |
+
"batch_size": batch,
|
| 345 |
},
|
| 346 |
config_json,
|
| 347 |
indent=4,
|
| 348 |
)
|
| 349 |
config_json.write("\n")
|
| 350 |
|
| 351 |
+
class CustomDataset(Dataset):
|
| 352 |
+
def __init__(self, data):
|
| 353 |
+
self.data = data
|
| 354 |
+
|
| 355 |
+
def __len__(self):
|
| 356 |
+
return len(self.data)
|
| 357 |
+
|
| 358 |
+
def __getitem__(self, index):
|
| 359 |
+
sample = self.data[index]
|
| 360 |
+
return sample["conversations"][0]["value"]
|
| 361 |
+
|
| 362 |
+
|
| 363 |
+
def dataloader(input_file: str, batch_size: batch) -> Generator[tuple[bool, str], None, None]:
|
| 364 |
"""Yields a tuple of whether this is a warmup run and the input prompt."""
|
| 365 |
+
for _ in range(3):
|
| 366 |
+
yield True, ["Say something long and random. I don't care about the content." for _ in range (batch)]
|
| 367 |
+
data = json.load(open(input_file, "r"))
|
| 368 |
+
custom_dataset = CustomDataset(data)
|
| 369 |
+
data_loader = DataLoader(custom_dataset, batch_size=batch_size, shuffle=False)
|
| 370 |
+
for prompt in data_loader:
|
| 371 |
+
yield False, prompt
|
| 372 |
|
| 373 |
# Warm up the GPU with some random prompts.
|
| 374 |
# Forward through all the prompts.
|
| 375 |
is_first = True
|
| 376 |
convs = []
|
| 377 |
prompts = []
|
| 378 |
+
data_iter = iter(dataloader(input_file, batch))
|
| 379 |
+
|
| 380 |
+
for is_warmup, input_prompts in data_iter:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 381 |
# Construct the input prompt.
|
| 382 |
+
for i in range(batch):
|
| 383 |
conv = copy.deepcopy(conv_base)
|
| 384 |
+
conv.append_message(conv.roles[0], input_prompts[i])
|
| 385 |
conv.append_message(conv.roles[1], "")
|
| 386 |
prompt = conv.get_prompt()
|
| 387 |
prompts.append(prompt)
|
| 388 |
convs.append(conv)
|
| 389 |
+
|
| 390 |
gen_params["prompt"] = prompts
|
|
|
|
|
|
|
| 391 |
|
| 392 |
# Print input prompt.
|
| 393 |
for i in range(len(convs)):
|
| 394 |
console.print(f"\n[u cyan]{'Warmup ' if is_warmup else ''}Prompt[/u cyan](batch_{i}):")
|
| 395 |
console.print(prompts[i].strip() + "\n", markup=False)
|
| 396 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 397 |
#################################################
|
| 398 |
# Inference and measurement zone!
|
| 399 |
#################################################
|
| 400 |
monitor.begin_window("inference")
|
| 401 |
+
results = run_inference(model, tokenizer, gen_params, device="cuda", context_len=2048)
|
|
|
|
|
|
|
|
|
|
| 402 |
measurements = monitor.end_window("inference")
|
| 403 |
#################################################
|
| 404 |
+
if results:
|
| 405 |
+
# Record numbers.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 406 |
if not is_warmup:
|
| 407 |
+
response_length = sum([result.response_length for result in results]) # number of valid tokens
|
| 408 |
+
latency = measurements.time
|
| 409 |
+
throughput = response_length / latency
|
| 410 |
+
energy = measurements.total_energy
|
| 411 |
+
output = {
|
| 412 |
+
"model": model_name_cleaned,
|
| 413 |
+
"throughput": throughput,
|
| 414 |
+
"response_length": response_length,
|
| 415 |
+
"latency": latency,
|
| 416 |
+
"energy": energy,
|
| 417 |
+
"input": [prompt.strip() for prompt in prompts],
|
| 418 |
+
"output": [(result.output).strip() for result in results],
|
| 419 |
+
}
|
| 420 |
+
output_str = json.dumps(output, indent=4)
|
| 421 |
+
if not is_warmup:
|
| 422 |
+
if not is_first:
|
| 423 |
+
output_json.write(",\n" + output_str)
|
| 424 |
+
else:
|
| 425 |
+
is_first = False
|
| 426 |
+
output_json.write(output_str)
|
| 427 |
+
output_json.flush()
|
| 428 |
+
|
| 429 |
+
# Print the response.
|
| 430 |
+
for i in range(len(convs)):
|
| 431 |
+
console.print(f"\n[u cyan]{'Warmup ' if is_warmup else ''}Response[/u cyan](batch_{i}):")
|
| 432 |
+
console.print(results[i].output.strip() + "\n", markup=False)
|
| 433 |
|
| 434 |
# Print measurement.
|
| 435 |
console.print(measurements)
|
| 436 |
convs = []
|
| 437 |
prompts = []
|
| 438 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 439 |
if __name__ == "__main__":
|
| 440 |
tyro.cli(main)
|