Created
October 1, 2023 00:37
-
-
Save penut85420/54765b2ed8a3e2e127ffdc0caf2dcaab to your computer and use it in GitHub Desktop.
An example of Hugging Face Text Generation Inference launch script.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
import os | |
import subprocess as sp | |
def launch(): | |
container_name = "HelloTGI" | |
local_model = True # Model 是否已下載到本機內 | |
download_dir = "Models" # 如果不是 Local Model 要下載到哪裡 | |
model_path = "Models/Llama-2-7b-chat-fp16" | |
image_name = "ghcr.io/huggingface/text-generation-inference:latest" | |
quantize = "bitsandbytes-nf4" # gptq, bitsandbytes | |
quantize = None | |
port = 8080 | |
gpu_device = 0 | |
batch_size = 8 # 同時推論數量 | |
max_input_length = 1500 # 最大輸入長度 | |
max_new_tokens = 500 # 最大輸出長度 | |
max_total_tokens = max_input_length + max_new_tokens | |
max_batch_prefill_tokens = max_input_length * batch_size | |
cwd_dir = os.getcwd() | |
if local_model: | |
host_model_dir = os.path.join(cwd_dir, model_path) | |
guest_model_dir = os.path.join("/", model_path) | |
else: | |
host_model_dir = os.path.join(cwd_dir, download_dir) | |
guest_model_dir = "/data" | |
max_best_of = 1 | |
num_shard = 1 | |
max_concurrent_requests = 128 | |
cuda_memory_fraction = 1.0 | |
# Docker Arguments | |
cmds = list() | |
cmds += ["docker", "run", "--rm", "--shm-size", "1g"] | |
cmds += ["-p", f"{port}:80", "--name", container_name] | |
cmds += ["--gpus", f"device={gpu_device}"] | |
cmds += ["-v", f"{host_model_dir}:{guest_model_dir}"] | |
cmds += [image_name] | |
# TGI Arguments | |
if local_model: | |
cmds += ["--model-id", guest_model_dir] | |
else: | |
cmds += ["--model-id", model_path] | |
cmds += ["--num-shard", num_shard] | |
cmds += ["--quantize", quantize] if quantize else [] | |
cmds += ["--max-input-length", max_input_length] | |
cmds += ["--max-total-tokens", max_total_tokens] | |
cmds += ["--max-batch-prefill-tokens", max_batch_prefill_tokens] | |
cmds += ["--max-concurrent-requests", max_concurrent_requests] | |
cmds += ["--max-best-of", max_best_of] | |
cmds += ["--cuda-memory-fraction", cuda_memory_fraction] | |
cmds += ["--trust-remote-code"] | |
cmds += ["--hostname", "0.0.0.0"] | |
cmds = map(str, cmds) | |
try: | |
proc = sp.Popen(cmds) | |
proc.wait() | |
exit(proc.returncode) | |
except KeyboardInterrupt: | |
proc.kill() | |
if __name__ == "__main__": | |
launch() |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment