As Figure 1 illustrates, ColBERT relies on fine-grained contextual late interaction: it encodes each passage into a matrix of token-level embeddings (shown above in blue).
Runs On
What it takes to serve colbertv2.0 (110M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 0.2 GB | 0.3 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.1 GB | 0.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.1 GB | 0.1 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Sep 18, 2026.
SAVRN's Notes on colbertv2.0
Do not size a ColBERT deployment off the model line alone. The weights are small: 110M parameters need 0.3 GB of memory at 16-bit, and a 192 GB MI300X at $1.85 per hour on-demand is the lowest-priced setup on our list. What grows is the index: late interaction encodes every passage into a matrix of token-level embeddings, and at query time a MaxSim operator scores the query matrix against them. The card runs the 12-layer encoder with its 512 token context; your storage plan carries the passages.
On licensing there is nothing to negotiate: MIT permits commercial use, modification and redistribution with the copyright and permission notices kept. What to check: the page carries no task label and no host prices, the checkpoint is float32, 1.31 GB across 11 files, and weights were last updated April 5, 2024. Five papers describe the method, from arXiv:2004.12832 through arXiv:2205.09707.
Model Card
By ColBERT, published under mit, revision c1e84128e85e.
# ColBERT (v2) ### ColBERT is a _fast_ and _accurate_ retrieval model, enabling scalable BERT-based search over large text collections in tens of milliseconds. [](https://colab.research.google.com/github/stanford-futuredata/ColBERT/blob/main/docs/intro2new.ipynb) Figure 1: ColBERT's late interaction, efficiently scoring the fine-grained similarity between a queries and a passage. As Figure 1 illustrates, ColBERT relies on fine-grained **contextual late interaction**: it encodes each passage into a **matrix** of token-level embeddings (shown above in blue). Then at search time, it embeds every query into another matrix (shown in green) and efficiently finds passages that contextually match the query using scalable vector-similarity (`MaxSim`) operators. These rich interactions allow ColBERT to surpass the quality of _single-vector_ representation models, while scaling efficiently to large corpora. You can read more in our papers: * [**ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERT**](https://arxiv.org/abs/2004.12832) (SIGIR'20). * [**Relevance-guided Supervision for OpenQA with ColBERT**](https://arxiv.org/abs/2007.00814) (TACL'21). * [**Baleen: Robust Multi-Hop Reasoning at Scale via Condensed Retrieval**](https://arxiv.org/abs/2101.00436) (NeurIPS'21). * [**ColBERTv2: Effective and Efficient Retrieval via…
Configuration
- Architecture
- HF_ColBERT
- Context length (tokens)
- 512
- Layers
- 12
- Hidden size
- 768
- Feed-forward size
- 3,072
- Attention heads
- 12
- Vocabulary size
- 30,522
- Stored precision
- float32
- Model type
- bert
Identity and Version
- Repository
- colbert-ir/colbertv2.0
- Publisher
- ColBERT
- Task
- Not stated by the source
- Modality
- Other
- Library
- transformers
- Parameters
- 110M parameters
- Languages
- en
- Revision
- c1e84128e85ef755c096a95bdb06b47793b13acf
- First published
- 2023-06-27
- Last updated
- 2024-04-05
Files and Weights
11 files, 1.3 GB in total. The weights are 3 files totalling 1.3 GB in bin, onnx, safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.onnx | Weights | 436.2 MB | bfe81fa313c4 |
| model.safetensors | Weights | 438.3 MB | 3f58890b1dfd |
| pytorch_model.bin | Weights | 438.4 MB | 26e4c2f9f95a |
| config.json | Configuration | 743 B | — |
| special_tokens_map.json | Configuration | 112 B | — |
| README.md | Documentation | 10.4 KB | — |
| artifact.metadata | Other | 1.6 KB | — |
| .gitattributes | Repository | 1.5 KB | — |
| tokenizer.json | Tokenizer | 466.1 KB | — |
| tokenizer_config.json | Tokenizer | 405 B | — |
| vocab.txt | Tokenizer | 231.5 KB | — |
License and Download
- License
- mit
- Access
- Open weights, no gate
- Download size
- 1.3 GB
Released by ColBERT through its official repository on Hugging Face. Read the license.
Built From
- Described by arXiv:2004.12832
- Described by arXiv:2007.00814
- Described by arXiv:2101.00436
- Described by arXiv:2112.01488
- Described by arXiv:2205.09707
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 1.3 GB |
| 16-bit | 0.2 GB |
| 8-bit | 0.1 GB |
| 4-bit | 0.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About colbertv2.0
How much GPU memory does colbertv2.0 need?
About 0.3 GB at 16-bit and 0.1 GB at 4-bit: the weights (110M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run colbertv2.0 on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use colbertv2.0 commercially?
Yes. colbertv2.0 is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.
What is colbertv2.0's context length?
512 tokens, from the maximum position embeddings in its published configuration.