Utilizing the GPU of MacBook Pro M3 to accelerate the process of fit_transform

I am running BERTopic on a MacBook Pro M3, and the text data is nearly 190,000 Chinese sentences. It has taken two full days and still is not finished. By now it has run 2928 minutes. I wonder if that's because I did not utilize the GPU, but I do not know how. Any help is much appreciated!

Following is the code I used:

sampled_data2 = data1['2grams_filtered'].tolist()
sampled_data2_as_strings = [' '.join(tokens) for tokens in sampled_data2]

sentence_model = SentenceTransformer("paraphrase-multilingual-MiniLM-L12-v2")
embeddings = sentence_model.encode(sampled_data2_as_strings, show_progress_bar=True)

stopWord_Path='hit_stopwords.txt'
stopWord=[]
with open(stopWord_Path,'r',encoding='utf-8') as fr:
for word in fr.readlines():
stopWord.append(word.strip())

umap_model = UMAP(n_neighbors=100, n_components=2, min_dist=0.1, metric='cosine', random_state=42)

hdbscan_model = HDBSCAN(min_cluster_size=5, metric='euclidean', cluster_selection_method='eom', prediction_data=True,
min_samples=5, gen_min_span_tree=True)

vectorizer_model = CountVectorizer(ngram_range=(1, 2), stop_words=stopWord)

ctfidf_model = ClassTfidfTransformer(seed_words=['台独','蔡英文', '佩洛西'], bm25_weighting=True, reduce_frequent_words=True)

topic_model = BERTopic(embedding_model="distiluse-base-multilingual-cased-v1",
language="multilingual",
verbose=True,
calculate_probabilities=True,
umap_model=umap_model,
hdbscan_model=hdbscan_model,
vectorizer_model=vectorizer_model,
ctfidf_model=ctfidf_model,
nr_topics="auto",
top_n_words=15,
min_topic_size=120)

topics, probs = topic_model.fit_transform(sampled_data2_as_strings, embeddings)

Start with the BERTopic.fit_transform entry point and the supplied SentenceTransformer.encode call, then trace how the provided UMAP and HDBSCAN models are executed. Done means the issue can state whether this configuration supports GPU use on an M3 Mac and document the supported path or limitations.