Spaces:

HuggingFaceTB

/

SmolLM-360M-Instruct-WebGPU

like 125

Running

To use the onnx version you can do it like this

from transformers import AutoTokenizer
from optimum.onnxruntime import ORTModelForCausalLM
import torch
import os

# Specify the local path to your cloned repository
# git clone https://huggingface.co/HuggingFaceTB/SmolLM-360M-Instruct
repo_path = "path/to/local/repo"

# cp repo_path/onnx/model.onnx repo_path/model.onnx -> moove the onnx model you want to the main folder

# Load the tokenizer and model from local paths
tokenizer = AutoTokenizer.from_pretrained(repo_path)
model = ORTModelForCausalLM.from_pretrained(repo_path)

# Prepare the input using the chat template
messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "What is the capital of France ?"}
]

# Apply the chat template
input_text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

# Generate the response
inputs = tokenizer(input_text, return_tensors="pt")
gen_tokens = model.generate(**inputs, do_sample=True, temperature=0.2, top_p=0.9, min_length=20, max_length=100)

# Decode and print the output
output = tokenizer.batch_decode(gen_tokens, skip_special_tokens=True)
print(output[0])
``