File size: 1,668 Bytes
9aa8ed3 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 |
import torch
from transformers import AutoTokenizer
from llama_modeling.front_end import LlamaForCausalLM
from llama_modeling.config import LlamaConfig
import json
import sys
from utils.trainutils import load_checkpoint
def generate_text(model, tokenizer, prompt, max_new_tokens=30):
input_ids = tokenizer.encode(prompt, return_tensors='pt').to("cuda")
with torch.inference_mode():
outputs = model.generate(
input_ids,
max_new_tokens=max_new_tokens,
temperature=0.7
)
return tokenizer.decode(outputs[0], skip_special_tokens=True)
def main():
if len(sys.argv) != 2:
print("Usage: python inference.py <path_to_model>")
sys.exit(1)
model_path = sys.argv[1]
device = "cuda" if torch.cuda.is_available() else "cpu"
with open("config.json") as f:
config_dict = json.load(f)
config = LlamaConfig(**{k: v for k, v in config_dict.items() if k in LlamaConfig.__dataclass_fields__})
model = LlamaForCausalLM(config).to(device)
load_checkpoint(model, model_path)
model.eval()
tokenizer = AutoTokenizer.from_pretrained("./SmolLM2-135M-Instruct")
prompts = [
"Once upon a time,",
"The best way to learn programming is",
"Here's a recipe for chocolate cake:"
]
with torch.no_grad(), torch.autocast(device_type='cuda', dtype=None):
for prompt in prompts:
print(f"\nPrompt: {prompt}")
output = generate_text(model, tokenizer, prompt)
print(f"Generated: {output}")
print("-" * 50)
if __name__ == "__main__":
main() |