[Python] Hey can anyone please verify if i am doing something wrong or the vertex api is...

please check out the stuff below

• ok so i locally tested both the api i am using with sample small data
• here is the embedding api use and output

client = genai.Client()

start_time = time.time()

response = client.models.embed_content(
model="text-embedding-005",
contents=["text to be embedded", "idk how is this working", "hello i am you"],
config=EmbedContentConfig(
task_type="CLASSIFICATION",
output_dimensionality=768,
),
)

end_time = time.time()
print(f"Time taken: {end_time - start_time:.4f} seconds")
print(response.embeddings[1].values)


# output
Time taken: 12.6191 seconds
...embedding vector

• and here is the token count api

client = genai.Client(http_options=HttpOptions(api_version="v1"))

prompt = "Why is the sky blue?"

strt_time = time.time()
response = client.models.generate_content(
model="gemini-2.0-flash", contents=prompt
)
end_time = time.time()

print(f"Time taken: {end_time - strt_time:.4f} seconds")
print(response.usage_metadata.prompt_token_count)


# output
Time taken: 3.5657 seconds
6

• but when i use these combined to get some embeddings on this data:

"http://www.saman-sna.cf"
"https://libreriabaruqeros45.libreria15.repl.co"
"https://www.endometriosis-uk.org"
"https://www.mca.org.mt"
"http://www.xgdsgj.com" # ...only 100 rows of this kinda strings for testing

• using this code on google colab

import asyncio
from google import genai
from google.genai.types import EmbedContentConfig
from google.genai.types import HttpOptions

# Init client with HTTP options to get token usage
client = genai.Client(http_options=HttpOptions(api_version="v1"))

# Load the dataframe (modify this path or use your own df)
df_100_rows = pd.read_csv("data.csv").head(100) # only first 100 rows
urls = df_100_rows['url'].tolist()

# Constants
BATCH_SIZE = 250
MAX_TOKENS_PER_BATCH = 20000

# Helper: get token count for a single input
def get_token_count(text: str) -> int:
response = client.models.generate_content(
model="gemini-2.0-flash", contents=text
)
return response.usage_metadata.prompt_token_count

# Helper: yield batches with token safety
def get_safe_batches(urls):
batch = []
total_tokens = 0
for url in urls:
token_count = get_token_count(url)
if (len(batch) >= BATCH_SIZE) or (total_tokens + token_count > MAX_TOKENS_PER_BATCH):
yield batch
batch = []
total_tokens = 0
batch.append(url)
total_tokens += token_count
if batch:
yield batch

# Async embed one batch
async def embed_batch(batch):
return client.models.embed_content(
model="text-embedding-005",
contents=batch,
config=EmbedContentConfig(
task_type="CLASSIFICATION",
output_dimensionality=768,
),
).embeddings

# Main embedding loop
async def embed_all(urls):
all_embeddings = []
for i, batch in enumerate(get_safe_batches(urls), 1):
print(f"Embedding batch {i} with {len(batch)} URLs...")
embeddings = await embed_batch(batch)
all_embeddings.extend([e.values for e in embeddings])
return all_embeddings


embeddings = asyncio.run(embed_all(urls))
print(f"Total embedded: {len(embeddings)}")

• more than 15 mins passed and no output yet ... so i terminated it but i wonder how is the vertex ai of google which is pretty much their enterprise ai solution be so slow even for only 100 small strings ... so does that mean i am doing something wrong here ?

Continue reading...