Skip to content

Correcting the default batch size calculation in the voyage models. #3091

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
KennethEnevoldsen merged 1 commit into from
Sep 1, 2025
Merged

Correcting the default batch size calculation in the voyage models. #3091

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 47 additions & 1 deletion mteb/models/voyage_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,6 +16,24 @@
# synthetic data
}

# Total token limits per model based on VoyageAI documentation
VOYAGE_TOTAL_TOKEN_LIMITS = {
"voyage-3.5-lite": 1_000_000,
"voyage-3.5": 320_000,
"voyage-2": 320_000,
"voyage-3-large": 120_000,
"voyage-code-3": 120_000,
"voyage-large-2-instruct": 120_000,
"voyage-finance-2": 120_000,
"voyage-multilingual-2": 120_000,
"voyage-law-2": 120_000,
"voyage-large-2": 120_000,
"voyage-3": 120_000,
"voyage-3-lite": 120_000,
"voyage-code-2": 120_000,
"voyage-3-m-exp": 120_000,
}


def token_limit(max_tpm: int, interval: int = 60):
limit_interval_start_ts = time.time()
Expand Down Expand Up @@ -75,6 +93,7 @@ def __init__(
max_retries: int = 5,
max_rpm: int = 300,
max_tpm: int = 1_000_000,
max_tokens: int | None = None,
model_prompts: dict[str, str] | None = None,
**kwargs,
) -> None:
Expand All @@ -85,17 +104,32 @@ def __init__(
self._embed_func = rate_limit(max_rpm)(token_limit(max_tpm)(self._client.embed))
self._model_name = model_name
self._max_tpm = max_tpm
self._max_tokens = max_tokens
self.model_prompts = self.validate_task_to_prompt_name(model_prompts)

def _calculate_default_batch_size(self) -> int:
"""Calculate the default batch size based on total token limit and context length.

Formula: floor(total_token_limit / context_length)
"""
if self._max_tokens is None:
return 32 # fallback to original default

total_token_limit = VOYAGE_TOTAL_TOKEN_LIMITS.get(self._model_name, 120_000)
return max(1, total_token_limit // self._max_tokens)

def encode(
self,
sentences: list[str],
*,
batch_size: int = 32,
batch_size: int | None = None,
task_name: str,
prompt_type: PromptType | None = None,
**kwargs: Any,
) -> np.ndarray:
if batch_size is None:
batch_size = self._calculate_default_batch_size()

prompt_name = self.get_prompt_name(self.model_prompts, task_name, prompt_type)
input_type = self.model_prompts.get(prompt_name, "document")

Expand Down Expand Up @@ -149,6 +183,7 @@ def _batched_encode(
loader=partial(
VoyageWrapper,
model_name="voyage-3.5",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -174,6 +209,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-large-2-instruct",
max_tokens=16000,
model_prompts=model_prompts,
),
max_tokens=16000,
Expand All @@ -199,6 +235,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-finance-2",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -224,6 +261,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-law-2",
max_tokens=16000,
model_prompts=model_prompts,
),
max_tokens=16000,
Expand All @@ -249,6 +287,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-code-2",
max_tokens=16000,
model_prompts=model_prompts,
),
max_tokens=16000,
Expand All @@ -274,6 +313,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-code-3",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -300,6 +340,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-large-2",
max_tokens=16000,
model_prompts=model_prompts,
),
max_tokens=16000,
Expand All @@ -325,6 +366,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-2",
max_tokens=4000,
model_prompts=model_prompts,
),
max_tokens=4000,
Expand All @@ -349,6 +391,7 @@ def _batched_encode(
loader=partial( # type: ignore
VoyageWrapper,
model_name="voyage-multilingual-2",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -374,6 +417,7 @@ def _batched_encode(
loader=partial(
VoyageWrapper,
model_name="voyage-3",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -399,6 +443,7 @@ def _batched_encode(
loader=partial(
VoyageWrapper,
model_name="voyage-3-lite",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand All @@ -424,6 +469,7 @@ def _batched_encode(
loader=partial(
VoyageWrapper,
model_name="voyage-3-m-exp",
max_tokens=32000,
model_prompts=model_prompts,
),
max_tokens=32000,
Expand Down
Loading