Skip to content

Adding nvidia/nemotron-colembed models #3941

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
Samoed merged 7 commits into from
Jan 27, 2026
Merged

Adding nvidia/nemotron-colembed models #3941

Changes from all commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
b7a0567
Adding nvidia/nemotron-colembed models
rnyak Jan 15, 2026
af3fe4d
add colembed 4b, 8b model meta
rnyak Jan 17, 2026
db6adb4
fix colembed-3b-v2 model name
rnyak Jan 17, 2026
8f154d6
update revision for colembed 3b
rnyak Jan 17, 2026
1d7c332
update revisions
rnyak Jan 22, 2026
c538850
Update mteb/models/model_implementations/nvidia_llama_nemoretriever_c…
rnyak Jan 26, 2026
2a623c7
lint
Samoed Jan 27, 2026
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
120 changes: 110 additions & 10 deletions mteb/models/model_implementations/nvidia_llama_nemoretriever_colemb.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
from typing import TYPE_CHECKING, Any

import torch
from packaging.version import Version
from packaging.specifiers import SpecifierSet
from torch.utils.data import DataLoader
from transformers import __version__ as transformers_version

Expand Down Expand Up @@ -31,18 +31,20 @@ def __init__(
model_name_or_path: str,
revision: str,
trust_remote_code: bool,
transformers_version_constraint: str | None = None,
device_map="cuda",
torch_dtype=torch.bfloat16,
attn_implementation="flash_attention_2",
**kwargs,
):
required_transformers_version = "4.49.0"

if Version(transformers_version) != Version(required_transformers_version):
raise RuntimeError(
f"transformers version {transformers_version} is not match with required "
f"install version {required_transformers_version} to run `nvidia/llama-nemoretriever-colembed`"
)
if transformers_version_constraint is not None:
spec = SpecifierSet(transformers_version_constraint)
if transformers_version not in spec:
raise RuntimeError(
f"Model `{model_name_or_path}` requires transformers{transformers_version_constraint}, "
f"but {transformers_version} is installed. "
f"Run: pip install 'transformers{transformers_version_constraint}'"
)

from transformers import AutoModel

Expand Down Expand Up @@ -150,10 +152,24 @@ def encode(
"wiki-ss-nq",
}


TRAINING_DATA_v2 = {
"VidoreDocVQARetrieval",
"VidoreInfoVQARetrieval",
"VidoreTatdqaRetrieval",
"VidoreArxivQARetrieval",
"docmatix-ir",
"VDRMultilingualRetrieval",
"VisRAG-Ret-Train-Synthetic-data",
"VisRAG-Ret-Train-In-domain-data",
"wiki-ss-nq",
}

llama_nemoretriever_colembed_1b_v1 = ModelMeta(
loader=LlamaNemoretrieverColembed,
loader_kwargs=dict(
trust_remote_code=True,
transformers_version_constraint="==4.49.0",
),
name="nvidia/llama-nemoretriever-colembed-1b-v1",
model_type=["late-interaction"],
Expand All @@ -168,7 +184,7 @@ def encode(
embed_dim=2048,
license="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE",
open_weights=True,
public_training_code="Proprietary Code",
public_training_code=None,
public_training_data="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1#training-dataset",
framework=["PyTorch", "Transformers", "safetensors"],
reference="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1",
Expand All @@ -182,6 +198,7 @@ def encode(
loader=LlamaNemoretrieverColembed,
loader_kwargs=dict(
trust_remote_code=True,
transformers_version_constraint="==4.49.0",
),
name="nvidia/llama-nemoretriever-colembed-3b-v1",
model_type=["late-interaction"],
Expand All @@ -196,7 +213,7 @@ def encode(
embed_dim=3072,
license="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE",
open_weights=True,
public_training_code="Proprietary Code",
public_training_code=None,
public_training_data="https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1#training-dataset",
framework=["PyTorch", "Transformers", "safetensors"],
reference="https://huggingface.co/nvidia/llama-nemoretriever-colembed-3b-v1",
Expand All @@ -205,3 +222,86 @@ def encode(
training_datasets=TRAINING_DATA,
citation=LLAMA_NEMORETRIEVER_CITATION,
)

llama_nemotron_colembed_vl_3b_v2 = ModelMeta(
loader=LlamaNemoretrieverColembed,
loader_kwargs=dict(
trust_remote_code=True,
transformers_version_constraint="==4.49.0",
),
name="nvidia/llama-nemotron-colembed-vl-3b-v2",
model_type=["late-interaction"],
languages=["eng-Latn"],
revision="75f03c712cb3a252e062295f9a0966e5d95d6156",
release_date="2026-01-21",
modalities=["image", "text"],
n_parameters=4_407_000_000,
memory_usage_mb=8403,
max_tokens=8192,
embed_dim=3072,
license="https://huggingface.co/nvidia/llama-nemotron-colembed-vl-3b-v2/blob/main/LICENSE",
open_weights=True,
public_training_code=None,
public_training_data="https://huggingface.co/nvidia/llama-nemotron-colembed-vl-3b-v2#training-dataset",
framework=["PyTorch", "Transformers", "safetensors"],
reference="https://huggingface.co/nvidia/llama-nemotron-colembed-vl-3b-v2",
similarity_fn_name="MaxSim",
use_instructions=True,
training_datasets=TRAINING_DATA,
citation=LLAMA_NEMORETRIEVER_CITATION,
)

nemotron_colembed_vl_4b_v2 = ModelMeta(
loader=LlamaNemoretrieverColembed,
loader_kwargs=dict(
trust_remote_code=True,
transformers_version_constraint="==5.0.0rc0",
),
name="nvidia/nemotron-colembed-vl-4b-v2",
revision="823b1625c15fe3da73fa094205e538a7a2301a2a",
languages=["eng-Latn"],
release_date="2026-01-07",
modalities=["image", "text"],
n_parameters=4_800_000_000,
memory_usage_mb=9206,
max_tokens=262144,
embed_dim=2560,
license="https://huggingface.co/nvidia/nemotron-colembed-vl-4b-v2/blob/main/LICENSE",
open_weights=True,
public_training_code=None,
public_training_data="https://huggingface.co/nvidia/nemotron-colembed-vl-4b-v2#training-dataset",
framework=["PyTorch", "Transformers"],
reference="https://huggingface.co/nvidia/nemotron-colembed-vl-4b-v2",
similarity_fn_name="MaxSim",
use_instructions=True,
training_datasets=TRAINING_DATA_v2,
citation=LLAMA_NEMORETRIEVER_CITATION,
)


nemotron_colembed_vl_8b_v2 = ModelMeta(
loader=LlamaNemoretrieverColembed,
loader_kwargs=dict(
trust_remote_code=True,
transformers_version_constraint="==5.0.0rc0",
),
name="nvidia/nemotron-colembed-vl-8b-v2",
revision="6cbe43579dda6237768fc373768ad372cc5cdfec",
languages=["eng-Latn"],
release_date="2026-01-07",
modalities=["image", "text"],
n_parameters=8_700_000_000,
memory_usage_mb=16722,
max_tokens=262144,
embed_dim=4096,
license="https://huggingface.co/nvidia/nemotron-colembed-vl-8b-v2/blob/main/LICENSE",
open_weights=True,
public_training_code=None,
public_training_data="https://huggingface.co/nvidia/nemotron-colembed-vl-8b-v2#training-dataset",
framework=["PyTorch", "Transformers"],
reference="https://huggingface.co/nvidia/nemotron-colembed-vl-8b-v2",
similarity_fn_name="MaxSim",
use_instructions=True,
training_datasets=TRAINING_DATA_v2,
citation=LLAMA_NEMORETRIEVER_CITATION,
)