LateOn-Code

The LateOn-Code collection is composed of PyLate models optimized for code retrieval. These late interaction models are first pre-trained following the methodology of CoRNStack. These pre-trained models are then further fine-tuned on train sets of CoIR using the nv-retriever methodology to mine hard negatives while preventing false negatives.

We started from the two best ColBERT models on the BEIR benchmark for their respective sizes. The first one, LateOn-Code is based on in-house LateOn model, a new version of GTE-ModernColBERT-v1 built on ModernBERT-base (also developed at LightOn). This version underwent significantly deeper training, crossing the 57 mark on BEIR, almost a 2.5-point improvement and is thus SOTA by a large margin. We'll release this base model along with training data and boilerplates in the near future, so stay tuned! The second, LateOn-Code-edge is a smaller model based on the edge-colbert model family from mixedbread, using the smallest variant (Ettin-17M) for maximum efficiency. For more details on the training setup, please refer to our blogpost.

The original CoRNStack data in a format compatible with PyLate can be found here while the fine-tuning data can be found here. Training boilerplates can be found here in the PyLate repository

MTEB (Code, v1) benchmark results

Pre-trained models achieve very competitive results as the 17M model outperforms the very strong granite-embedding-small-english-r2 by an average of 1.7. This is truly impressive, as the granite model is almost three times bigger (17M vs 48M), but is also a beast on its own in the <100M parameters range. It also outperforms the larger granite variant (149M). The larger version nicely scales by improving over the performance of its little sibling by 6.5 on average.

Although the pre-training results are already very impressive given that they are mostly out-of-domain, running a proper fine-tuning using the training data of CoIR significantly boost the performance of the models. Notably, the 17M model increases from 57.50 to 66.64 (+9.14), getting pretty close to EmbeddingGemma-300M while being 17 times smaller. The larger one increases from 63.77 to 74.12 (+10.35), strongly outperforming EmbeddingGemma-300M and getting closer to strong LLM models such as Qwen3-Embedding-0.6B and C2LLM-0.5B while being much smaller.

Model	Params	Type	Avg	Apps	COIR CSNet	CodeEdit	CodeFB MT	CodeFB ST	CSNet CC	CSNet	CodeTrans Contest	CodeTrans DL	CosQA	StackOF QA	Synth T2SQL
Baseline
BM25	-	Lexical	44.41	4.76	40.86	49.85	59.19	68.15	53.97	60.01	47.78	34.42	18.75	70.26	24.94
Small (≤50M)
granite-embedding-small-english-r2	47M	Single vector	55.84	13.54	60.46	57.16	52.19	76.85	48.42

Best result across all sizes is underlined. Best within each size category is bolded.

Colgrep

The LateOn-Code family model can easily be used within ColGrep, an easy-to-use search tool that give their powerful search capabilities to coding agent. It has been designed to extend grep capabilities to get the best of both world and is very effective to enhance the quality of the answer while diminishing answer time and tokens consumption. Given the performance of the very light-weight 17M model, it can easily run quickly on any computer.

Install ColGrep

# macOS / Linux
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/lightonai/next-plaid/releases/latest/download/colgrep-installer.sh | sh

# Windows (PowerShell)
powershell -c "irm https://github.com/lightonai/next-plaid/releases/latest/download/colgrep-installer.ps1 | iex"

Search

# Semantic search — find code by meaning
colgrep "function that retries HTTP requests"

# Regex search
colgrep -e "async fn\s+\w+"

# Hybrid — regex narrows candidates, semantics ranks them
colgrep -e "Result<" "error handling" --include="*.rs"

Install for Claude Code

colgrep --install-claude-code

Choose a Model

# Set the model
colgrep set-model lightonai/LateOn-Code  # default: lightonai/LateOn-Code-edge

For more information about ColGrep, please refer to the official documentation

Model

This is a multi-vector (ColBERT-style late interaction) embedding model finetuned from lightonai/LateOn-Code-edge-pretrain on the apps, synthetictext2sql, cosqa, codefeedbackst, codefeedbackmt, stackoverflowqa, codetranscontest, codetransdl, CodeSearchNet_go, CodeSearchNet_java, CodeSearchNet_javascript, CodeSearchNet_php, CodeSearchNet_python, CodeSearchNet_ruby, CodeSearchNet_ccr_go, CodeSearchNet_ccr_java, CodeSearchNet_ccr_javascript, CodeSearchNet_ccr_php, CodeSearchNet_ccr_python and datasets. It maps sentences & paragraphs to sequences of 48-dimensional dense vectors and can be used for semantic textual similarity using the MaxSim operator.

Model Details

Model Description

• Model Type: Multi-vector embedding model

• Document Length: 2048 tokens
• Query Length: 256 tokens
• Output Dimensionality: 48 tokens
• Similarity Function: MaxSim
• Training Datasets:
  • apps
  • synthetictext2sql
  • cosqa
  • codefeedbackst
  • codefeedbackmt
  • stackoverflowqa
  • codetranscontest
  • codetransdl
  • CodeSearchNet_go
  • CodeSearchNet_java
  • CodeSearchNet_javascript
  • CodeSearchNet_php
  • CodeSearchNet_python
  • CodeSearchNet_ruby
  • CodeSearchNet_ccr_go
  • CodeSearchNet_ccr_java
  • CodeSearchNet_ccr_javascript

Model Sources

• Documentation: PyLate Documentation
• Repository: PyLate on GitHub
• Hugging Face: PyLate models on Hugging Face

Full Model Architecture

ColBERT(
  (0): Transformer({'max_seq_length': 2047, 'do_lower_case': True, 'architecture': 'ModernBertModel'})
  (1): Dense({'in_features': 256, 'out_features': 512, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'use_residual': False})
  (2): Dense({'in_features': 512, 'out_features': 48, 'bias': False, 'activation_function': 'torch.nn.modules.linear.Identity', 'use_residual': False})
)

Usage

Sentence Transformers

This model can be used with Sentence Transformers as a multi-vector (ColBERT-style late interaction) retriever via the MultiVectorEncoder:

pip install "sentence-transformers>=6.0.0"

from sentence_transformers import MultiVectorEncoder

model = MultiVectorEncoder("lightonai/LateOn-Code-edge")

query = "Which planet is known as the Red Planet?"
documents = [
    "Venus is often called Earth's twin because of its similar size and proximity.",
    "Mars, known for its reddish appearance, is often referred to as the Red Planet.",
    "Jupiter, the largest planet in our solar system, has a prominent red spot.",
    "Saturn, famous for its rings, is sometimes mistaken for the Red Planet.",
]

query_embeddings = model.encode_query(query)
document_embeddings = model.encode_document(documents)
print(query_embeddings.shape, document_embeddings[0].shape)
# (12, 48) (18, 48)

# MaxSim late-interaction scoring (higher is more relevant)
scores = model.similarity(query_embeddings, document_embeddings)
print(scores)
# tensor([[4.7185, 7.8521, 7.0537, 7.3523]])

PyLate

First install the PyLate library:

pip install -U pylate

Retrieval

Use this model with PyLate to index and retrieve documents. The index uses FastPLAID for efficient similarity search.

Indexing documents

Load the ColBERT model and initialize the PLAID index, then encode and index your documents:

from pylate import indexes, models, retrieve

# Step 1: Load the ColBERT model
model = models.ColBERT(
    model_name_or_path="pylate_model_id",
)

# Step 2: Initialize the PLAID index
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
    override=True,  # This overwrites the existing index if any
)

# Step 3: Encode the documents
documents_ids = ["1", "2", "3"]
documents = ["document 1 text", "document 2 text", "document 3 text"]

documents_embeddings = model.encode(
    documents,
    batch_size=32,
    is_query=False,  # Ensure that it is set to False to indicate that these are documents, not queries
    show_progress_bar=True,
)

# Step 4: Add document embeddings to the index by providing embeddings and corresponding ids
index.add_documents(
    documents_ids=documents_ids,
    documents_embeddings=documents_embeddings,
)

Note that you do not have to recreate the index and encode the documents every time. Once you have created an index and added the documents, you can re-use the index later by loading it:

# To load an index, simply instantiate it with the correct folder/name and without overriding it
index = indexes.PLAID(
    index_folder="pylate-index",
    index_name="index",
)

Retrieving top-k documents for queries

Once the documents are indexed, you can retrieve the top-k most relevant documents for a given set of queries. To do so, initialize the ColBERT retriever with the index you want to search in, encode the queries and then retrieve the top-k documents to get the top matches ids and relevance scores:

# Step 1: Initialize the ColBERT retriever
retriever = retrieve.ColBERT(index=index)

# Step 2: Encode the queries
queries_embeddings = model.encode(
    ["query for document 3", "query for document 1"],
    batch_size=32,
    is_query=True,  #  # Ensure that it is set to False to indicate that these are queries
    show_progress_bar=True,
)

# Step 3: Retrieve top-k documents
scores = retriever.retrieve(
    queries_embeddings=queries_embeddings,
    k=10,  # Retrieve the top 10 matches for each query
)

Reranking

If you only want to use the ColBERT model to perform reranking on top of your first-stage retrieval pipeline without building an index, you can simply use rank function and pass the queries and documents to rerank:

from pylate import rank, models

queries = [
    "query A",
    "query B",
]

documents = [
    ["document A", "document B"],
    ["document 1", "document C", "document B"],
]

documents_ids = [
    [1, 2],
    [1, 3, 2],
]

model = models.ColBERT(
    model_name_or_path="pylate_model_id",
)

queries_embeddings = model.encode(
    queries,
    is_query=True,
)

documents_embeddings = model.encode(
    documents,
    is_query=False,
)

reranked_documents = rank.rerank(
    documents_ids=documents_ids,
    queries_embeddings=queries_embeddings,
    documents_embeddings=documents_embeddings,
)

Evaluation

Metrics

Py Late Information Retrieval

• Dataset: ['CodeSearchNetPython', 'CodeSearchNetJavascript', 'CodeSearchNetGo', 'CodeSearchNetRuby', 'CodeSearchNetJava', 'CodeSearchNetPhp']
• Evaluated with pylate.evaluation.pylate_information_retrieval_evaluator.PyLateInformationRetrievalEvaluator

Metric	CodeSearchNetPython	CodeSearchNetJavascript	CodeSearchNetGo	CodeSearchNetRuby	CodeSearchNetJava	CodeSearchNetPhp
MaxSim_accuracy@1	0.855	0.707	0.92	0.737	0.755	0.802
MaxSim_accuracy@3	0.958	0.815	0.978	0.87	0.914	0.91
MaxSim_accuracy@5	0.972	0.845	0.987	0.899	0.937	0.932
MaxSim_accuracy@10	0.98	0.877	0.991	0.921	0.951	0.953
MaxSim_precision@1	0.855	0.707	0.92	0.737	0.755	0.802
MaxSim_precision@3	0.3193	0.2717	0.326	0.29	0.3047	0.3033
MaxSim_precision@5	0.1944	0.169	0.1974	0.1798	0.1874	0.1864
MaxSim_precision@10	0.098	0.0877	0.0991	0.0921	0.0951	0.0953
MaxSim_recall@1	0.855	0.707

Code Search Network

• Dataset: CodeSearchNet_mean
• Evaluated with pylate.evaluation.code_search_network_evaluator.CodeSearchNetworkEvaluator

Metric	Value
MaxSim_accuracy@1	0.796
MaxSim_accuracy@3	0.9075
MaxSim_accuracy@5	0.9287
MaxSim_accuracy@10	0.9455
MaxSim_precision@1	0.796
MaxSim_precision@3	0.3025
MaxSim_precision@5	0.1857
MaxSim_precision@10	0.0946
MaxSim_recall@1	0.796
MaxSim_recall@3	0.9075
MaxSim_recall@5	0.9287
MaxSim_recall@10	0.9455
MaxSim_ndcg@10	0.8771
MaxSim_mrr@10	0.8544
MaxSim_map@100	0.8557

Training Details

Training Datasets

apps

• Dataset: apps at 68d15dc
• Size: 4,985 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

synthetictext2sql

• Dataset: synthetictext2sql at 68d15dc
• Size: 99,996 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

cosqa

• Dataset: cosqa at 68d15dc
• Size: 9,018 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

codefeedbackst

• Dataset: codefeedbackst at 68d15dc
• Size: 125,124 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

codefeedbackmt

• Dataset: codefeedbackmt at 68d15dc
• Size: 52,941 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

stackoverflowqa

• Dataset: stackoverflowqa at 68d15dc
• Size: 13,934 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

codetranscontest

• Dataset: codetranscontest at 68d15dc
• Size: 561 training samples
• Approximate statistics based on the first 561 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

codetransdl

• Dataset: codetransdl at 68d15dc
• Size: 564 training samples
• Approximate statistics based on the first 564 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_go

• Dataset: CodeSearchNet_go at 9f89bdc
• Size: 166,972 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_java

• Dataset: CodeSearchNet_java at 9f89bdc
• Size: 162,773 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_javascript

• Dataset: CodeSearchNet_javascript at 9f89bdc
• Size: 56,734 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_php

• Dataset: CodeSearchNet_php at 9f89bdc
• Size: 240,327 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_python

• Dataset: CodeSearchNet_python at 9f89bdc
• Size: 251,063 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ruby

• Dataset: CodeSearchNet_ruby at 9f89bdc
• Size: 24,731 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_go

• Dataset: CodeSearchNet_ccr_go at 9f89bdc
• Size: 167,278 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_java

• Dataset: CodeSearchNet_ccr_java at 9f89bdc
• Size: 164,900 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_javascript

• Dataset: CodeSearchNet_ccr_javascript at 9f89bdc
• Size: 58,017 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_php

• Dataset: CodeSearchNet_ccr_php at 9f89bdc
• Size: 241,177 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_python

• Dataset: CodeSearchNet_ccr_python at 9f89bdc
• Size: 251,758 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

CodeSearchNet_ccr_ruby

• Dataset: CodeSearchNet_ccr_ruby at 9f89bdc
• Size: 24,918 training samples
• Approximate statistics based on the first 1000 samples:

  	query	positive	negative_0	negative_1	negative_2	negative_3	negative_4	negative_5	negative_6	negative_7	negative_8	negative_9	negative_10	negative_11	negative_12	negative_13	negative_14	negative_15	negative_16	negative_17	negative_18	negative_19	negative_20	negative_21	negative_22	negative_23	negative_24	negative_25	negative_26	negative_27	negative_28	negative_29	negative_30	negative_31	negative_32	negative_33	negative_34	negative_35	negative_36	negative_37	negative_38	negative_39	negative_40	negative_41	negative_42	negative_43	negative_44	negative_45	negative_46	negative_47	negative_48	negative_49

Training Hyperparameters

Non-Default Hyperparameters

• eval_strategy: steps
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• learning_rate: 3e-05
• num_train_epochs: 1
• bf16: True
• dataloader_num_workers: 8
• accelerator_config: {'split_batches': True, 'dispatch_batches': None, 'even_batches': True, 'use_seedable_sampler': True, 'non_blocking': False, 'gradient_accumulation_kwargs': None}

All Hyperparameters

• overwrite_output_dir: False
• do_predict: False
• eval_strategy: steps
• prediction_loss_only: True
• per_device_train_batch_size: 128
• per_device_eval_batch_size: 128
• per_gpu_train_batch_size: None
• per_gpu_eval_batch_size: None
• gradient_accumulation_steps: 1
• eval_accumulation_steps: None
• torch_empty_cache_steps: None
• learning_rate: 3e-05
• weight_decay: 0.0
• adam_beta1: 0.9
• adam_beta2: 0.999
• adam_epsilon: 1e-08
• max_grad_norm: 1.0
• num_train_epochs: 1
• max_steps: -1
• lr_scheduler_type: linear
• lr_scheduler_kwargs: {}
• warmup_ratio: 0.0
• warmup_steps: 0
• log_level: passive
• log_level_replica: warning
• log_on_each_node: True
• logging_nan_inf_filter: True
• save_safetensors: True
• save_on_each_node: False
• save_only_model: False
• restore_callback_states_from_checkpoint: False
• no_cuda: False
• use_cpu: False
• use_mps_device: False
• seed: 42
• data_seed: None
• jit_mode_eval: False
• bf16: True
• fp16: False
• fp16_opt_level: O1
• half_precision_backend: auto
• bf16_full_eval: False
• fp16_full_eval: False
• tf32: None
• local_rank: 0
• ddp_backend: None
• tpu_num_cores: None

Training Logs

Epoch	Step	Training Loss	CodeSearchNetPython_MaxSim_ndcg@10	CodeSearchNetJavascript_MaxSim_ndcg@10	CodeSearchNetGo_MaxSim_ndcg@10	CodeSearchNetRuby_MaxSim_ndcg@10	CodeSearchNetJava_MaxSim_ndcg@10	CodeSearchNetPhp_MaxSim_ndcg@10	CodeSearchNet_mean_MaxSim_ndcg@10
0.0000	1	6.4113	-	-	-	-	-	-	-
0.0391	1250	3.2574	-	-	-	-	-	-	-
0.0781	2500	19.7862	0.9377	0.7986	0.9622	0.8487	0.8837	0.8834	0.8857
0.1172	3750	4.6875	-	-	-	-	-	-	-
0.1562	5000	2.3691	0.9335	0.8001	0.9614	0.8435	0.8755	0.8818	0.8826
0.1953	6250	1.4007	-	-	-

Framework Versions

• Python: 3.12.12
• Sentence Transformers: 5.1.1
• PyLate: 1.3.4
• Transformers: 4.57.3
• PyTorch: 2.9.0+cu128
• Accelerate: 1.12.0
• Datasets: 4.4.2
• Tokenizers: 0.22.2

Citation

BibTeX

LateOn-Code

@misc{LateOn-Code,
  title  = {LateOn-Code: a Family of State-Of-The-Art Late Interaction Code Retrieval Models},
  author = {Chaffin, Antoine},
  url    = {https://huggingface.co/collections/lightonai/lateon-code},
  year   = {2026}
}

ColGrep

@software{next-plaid,
  title  = {NextPlaid, ColGREP: Multi-vector search, from database to coding agents.},
  url    = {https://github.com/lightonai/next-plaid},
  author = {Raphaël Sourty},
  year   = {2026},
}

CoRNStack

@inproceedings{DBLP:conf/iclr/SureshRXNMDJ25,
  author       = {Tarun Suresh and
                  Revanth Gangi Reddy and
                  Yifei Xu and
                  Zach Nussbaum and
                  Andriy Mulyar and
                  Brandon Duderstadt and
                  Heng Ji},
  title        = {CoRNStack: High-Quality Contrastive Data for Better Code Retrieval
                  and Reranking},
  booktitle    = {The Thirteenth International Conference on Learning Representations,
                  {ICLR} 2025, Singapore, April 24-28, 2025},
  publisher    = {OpenReview.net},
  year         = {2025},
  url          = {https://openreview.net/forum?id=iyJOUELYir},
  timestamp    = {Sun, 25 May 2025 21:25:19 +0200},
  biburl       = {https://dblp.org/rec/conf/iclr/SureshRXNMDJ25.bib},
  bibsource    = {dblp computer science bibliography, https://dblp.org}
}

CoIR

@inproceedings{li2025coir,
  title     = {Coir: A comprehensive benchmark for code information retrieval models},
  author    = {Li, Xiangyang and Dong, Kuicai and Lee, Yi Quan and Xia, Wei and Zhang, Hao and Dai, Xinyi and Wang, Yasheng and Tang, Ruiming},
  booktitle = {Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  pages     = {22074--22091},
  year      = {2025}
}

Sentence Transformers

@inproceedings{reimers-2019-sentence-bert,
  title     = "Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks",
  author    = "Reimers, Nils and Gurevych, Iryna",
  booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing",
  month     = "11",
  year      = "2019",
  publisher = "Association for Computational Linguistics",
  url       = "https://arxiv.org/abs/1908.10084"
}

PyLate

@inproceedings{DBLP:conf/cikm/ChaffinS25,
  author       = {Antoine Chaffin and
                  Rapha{"{e}}l Sourty},
  editor       = {Meeyoung Cha and
                  Chanyoung Park and
                  Noseong Park and
                  Carl Yang and
                  Senjuti Basu Roy and
                  Jessie Li and
                  Jaap Kamps and
                  Kijung Shin and
                  Bryan Hooi and
                  Lifang He},
  title        = {PyLate: Flexible Training and Retrieval for Late Interaction Models},
  booktitle    = {Proceedings of the 34th {ACM} International Conference on Information
                  and Knowledge Management, {CIKM} 2025, Seoul, Republic of Korea, November
                  10-14, 2025},
  pages        = {6334--6339},
  publisher    = {{ACM}},
  year         = {2025},
  url          = {https://github.com/lightonai/pylate},
  doi          = {10.1145/3746252.3761608},
}