The story so far
Weights, every dated article, and repo commits on one rail.
2 dated articles and 7 repo commits so far
Teal is the weights (2025-07-17), amber triangles are 2 dated articles (each links to it), grey ticks are repo commits.
Models › granite-embedding family
granite-embedding-small-english-r2
ibm-granite Apache 2.0 out 15 months ago
47.7M parameters, text in, embeddings out, 8K tokens of context.
Verified 2026-10-25 Rank #50 in Embeddings · score 89.3 method v1 Sources 3 Downloads 30d 6.1M · #23 of 619 tracked
Do people use it?6.1Mdownloads, 30 days#23 of 619 we track · top 4%+6 likes, downloads +9% this week111 likesHugging Face2026-10-24
Where it's used
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 14 quantized builds and 6 Spaces.
- As a service0no API provider listed
- On own hardware14quantized builds
- Built on22 fine-tunes · 0 adapters · 0 merges
- In apps6Hugging Face Spaces
Top Spaces running it: embedl/hfviewer ♥ 181 · emanforhelp/model-pulse ♥ 5 · ibm-granite/granite-embedding ♥ 1
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at FP8
About 0.1 GB at FP8, 0.1 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
How it scores
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
#10 of 33 on MTEB Reranking so far
47.63. Each tick is a model on this site with a measured MTEB Reranking row; author-reported numbers stay in Reference. MAP x100 (mean of 2 tasks). 2 tasks, split=test
Source · leaderboard · 2026-10-24
What people built on it
16 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
16 derivatives on the Hub, most of them quantized
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Benchmarks
17 rows
Benchmarks 17 rows
measured third-party leaderboard row (3)author-reported number from the model card or eval files (14)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| measured MTEB · Embed text for search | 53.93 nDCG@10 x100 (mean of 10 tasks) | 10 tasks, split=test | leaderboard | 2026-10-24 |
| measured MTEB · Embed text for search | 61.07 mean(task) x100 over 41 tasks | 41 tasks, split=test | leaderboard | 2026-10-24 |
| measured MTEB Reranking · Rerank search results | 47.63 MAP x100 (mean of 2 tasks) | 2 tasks, split=test | leaderboard | 2026-10-24 |
| author-reported AVERAGE | 55.6 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported BEIR · Embed text for search | 50.9 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported CoIR · Embed text for search | 53.8 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported CoIR · Embed text for search | 53.4 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported Encoding Speed (docs/sec) | 199 other | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported Encoding Speed (dosc/sec) | 199 other | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported LongEmbed (6) | 61.9 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MLDR · Embed text for search | 39.8 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MLDR · Embed text for search | 40.1 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MTEB · Embed text for search | 61.1 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MTEB · Embed text for search | 53.9 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MTRAG (4) | 48.1 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported MTRAG (4) | 48.9 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
| author-reported Table IR (5) | 75.51 | granite-embedding-small-english-r2 | model author | 2026-10-07 |
What the card says
2 cited sentences
Summary generated, cited
The model generates 384-dimensional embedding vectors and supports a context length of up to 8192 tokens.[1][2] It is released under the Apache 2.0 license for use in text similarity and retrieval applications.[3][4]
- “output embedding size (_384_)”
- “context length of upto 8192 tokens”
- license
- “designed to produce fixed length vector representations f…”
Generated 2026-10-07 15:04 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Use-case evidence
2 use cases
What it is good at 2 use cases
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Facts
22 fields
Listed in Embeddings
Facts
- granite-embedding-small-english-r2 has 47.7M parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 8,192 tokens (8K). (config.json max_position_embeddings, observed 2026-10-24)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
- Created 2025-07-17, last modified 2026-01-21. (HF API, observed 2026-10-24)
| Pipeline tag | feature-extraction |
|---|---|
| Library | sentence-transformers |
| Architectures | ModernBertModel |
| Model type | modernbert |
| Dtype | bfloat16 |
| Params by dtype | BF16: 47,662,464 |
| Size band | S |
| MoE | no |
| Context length | 8,192 (config.json max_position_embeddings) |
| Languages | en |
| Papers | arXiv:2508.21085 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · ModernBertModel · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for granite-embedding-small-english-r2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 0.1 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.1 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.0 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
16 variants
Lineage
quantized: 14finetune: 2
Top variants (16 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| onnx-community/granite-embedding-small-english-r2-ONNX | quantized | — | — | — | apache-2.0 | onnx-community | 2 | 32k |
| mradermacher/granite-embedding-small-english-r2-GGUF | quantized | GGUF | — | — | apache-2.0 | mradermacher | 2 | 2.6k |
| keisuke-miyako/granite-embedding-small-english-r2-gguf-q4_k_m | quantized | gguf | — | — | — | keisuke-miyako | 0 | 629 |
| aisquared/bolt-embedding-small | finetune | — | 47.7M | S | other | aisquared | 0 | 454 |
| brokkai/Muninn-small | finetune | — | 47.7M | S | apache-2.0 | brokkai | 2 | 304 |
| Teradata/granite-embedding-small-english-r2 | quantized | — | — | — | apache-2.0 | Teradata | 0 | 201 |
| Arraasz/granite-embedding-small-english-r2-ane | quantized | — | — | — | apache-2.0 | Arraasz | 1 | 97 |
| sirasagi62/granite-embedding-small-english-r2-ONNX | quantized | — | — | — | apache-2.0 | sirasagi62 | 0 | 145 |
| jrc2139/granite-embedding-small-english-r2-ONNX | quantized | — | — | — | — | jrc2139 | 0 | 109 |
| keisuke-miyako/granite-embedding-small-english-r2-gguf-q8_0 | quantized | gguf | — | — | — | keisuke-miyako | 0 | 107 |
| daniel-gomm/scientific-granite | quantized | — | 47.7M | S | apache-2.0 | daniel-gomm | 1 | 27 |
| aisquared/bolt-embedding-small-gguf | quantized | gguf | — | — | other | aisquared | 0 | 73 |
| keisuke-miyako/granite-embedding-small-english-r2-onnx-int8 | quantized | int8 | — | — | — | keisuke-miyako | 0 | 53 |
| keisuke-miyako/granite-embedding-small-english-r2-onnx-fp16 | quantized | -fp16 | — | — | — | keisuke-miyako | 0 | 45 |
| keisuke-miyako/granite-embedding-small-english-r2-onnx-fp32 | quantized | — | — | — | — | keisuke-miyako | 0 | 39 |
| jevhome/jevhome-E | quantized | — | — | — | cc-by-nc-4.0 | jevhome | 0 | 29 |
Repository changelog
7 commits
Repository changelog 7 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-01-21 | Update pipeline tag and add project page link (#1) | 2ab6fa8 |
| 2026-01-06 | Upload model.safetensors | 0d44fac |
| 2025-09-22 | Update README.md | c949f23 |
| 2025-09-02 | Update README.md | d76b137 |
| 2025-08-14 | Upload 10 files | 54a8d26 |
| 2025-08-12 | Update README.md | 221d972 |
| 2025-07-17 | initial commit | 1ef0b36 |
Articles and coverage
8 pieces
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Official1
ibm.comofficialauthorityabout the granite embedding family920 words
IBM Granite Embedding models generate vector representations for semantic search and retrieval-augmented generation under the Apache 2.0 license. The granite-embedding-97m-multilingual-r2 model ranks #1 on the MTEB Leaderboard for multilingual embedding models under 100M parameters, while granite-embedding-311m-multilingual-r2 ranks #2 for models under 500M parameters. These models support 200+ languages and include Flash Attention 2 for efficient inference, plus Matryoshka Representation Learning to truncate embeddings to dimensions like 512, 256, or 128.written, reviewed, cited
- The ONNX model supports any ONNX Runtime backend including CPU, CUDA, TensorRT, and DirectML.
- The OpenVINO model is specifically optimized for Intel hardware including CPUs and integrated GPUs.
- The granite-embedding-311m-multilingual-r2 model uses CLS Pooling to generate its final embeddings.
Analysis7
arxiv.orgauthorityabout the granite embedding family10,314 words
The models deliver 16x expanded context length of 8,192 tokens and measurable speed advantages of 19-44% over leading competitors. The architecture includes a 22-layer retriever model with 149M parameters and an efficient 12-layer counterpart with 47M parameters.generated, cited
arxiv.orgauthorityabout the granite embedding family8,317 words
IBM releases Granite Embedding models under the Apache 2.0 license for retrieval tasks. The family includes 12 layer base models and efficient 6 layer distilled counterparts supporting a maximum context length of 512 tokens. Specific variants include 30M, 125M, 107M, and 278M parameter options for English and multilingual use cases.written, reviewed, cited
- Dense models use CLS pooling while the sparse model uses max-pooling across all sequence tokens.
- Multilingual models support 12 languages but can be finetuned on other languages covered by the XLM-RoBERTa vocabulary.
github.comauthorityabout the granite embedding family2,061 words
IBM released the Granite Embedding R2 collection, featuring four ModernBERT-based models optimized for retrieval tasks. The English models (47M and 149M parameters) launched on Aug 15, 2025, while multilingual variants (97M and 311M parameters) arrive on April 29, 2026. The 311M model supports Matryoshka dimension support (768/512/384/256/128), where cutting from 768 to 256 dimensions (a 3x reduction in storage and computation cost) drops MTEB Multilingual Retrieval by just 0.4 points (63.9 to 63.5).written, reviewed, cited
github.comauthorityabout the granite embedding family370 words
The IBM Granite repository provides a Jupyter notebook with example code for R2 embedding models. The notebook demonstrates how to use these models for embedding tasks within the open-source project. It is located in the main branch under the code directory.written, reviewed, cited
huggingface.co2025-10-14authorityabout the granite embedding family1,170 words
savrn.com1,621 words
arxiv.science2025-08-26about the granite embedding family263 words
Papers
1 paper
Papers 1
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
- Granite Embedding R2 Models this model
Similar models
3 models
Similar models
- bge-small-en-v1.5 · 33.4M same size band (S), same task (feature-extraction), more providers, permissive license
- jina-embeddings-v2-small-en · 32.7M same size band (S), same task (feature-extraction), permissive license
- bge-small-zh-v1.5 · 24.0M same size band (S), same task (feature-extraction), more providers, permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:37 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:04 UTC · Last verified 2026-10-25
Follow granite-embedding-small-english-r2
Watch this model
- /feeds/models/ibm-granite--granite-embedding-small-english-r2.xml(RSS 2.0) · every change to granite-embedding-small-english-r2
- /feeds/families/ibm-granite--granite-embedding.xml(RSS 2.0) · every change across the granite-embedding family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.