The story so far
Weights, every dated article, and repo commits on one rail.
18 repo commits since the weights landed
Teal is the weights (2023-09-12), grey ticks are repo commits.
bge-reranker-large
560M parameters, text in, embeddings out, 514 tokens of context. 1 provider serves it.
Verified 2026-10-25 Rank #24 in Embeddings · score 96.3 method v1 Sources 3 Downloads 30d 1.9M · #64 of 619 tracked
Do people use it?1.9Mdownloads, 30 days#64 of 619 we track · top 10%+3 likes, downloads −9% this week484 likesHugging Face2026-10-24
Where to run it
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
| # | Cheapest live row | In /1M | Out /1M | Context | Quant | Seen |
|---|---|---|---|---|---|---|
| 1 | HF Inference | — | — | — | — | Oct 25 |
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.
Served by 1 API provider; 22 quantized builds for local use.
- As a service1HF Inference
- On own hardware22quantized builds
- Built on1010 fine-tunes · 0 adapters · 0 merges
- In apps100+Hugging Face Spaces
Top Spaces running it: embedl/hfviewer ♥ 181 · lightmate/llm-chatbot ♥ 9 · emanforhelp/model-pulse ♥ 5 · sambanovasystems/enterprise_knowledge_retriever ♥ 4
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 0.3 GB at INT4, 0.7 GB at FP8, 1.3 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
What people built on it
32 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
32 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.
Use-case evidence
4 use cases
What it is good at 4 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
- bge-reranker-large has 560M parameters. (safetensors metadata, observed 2026-10-24)
- Context length is 514 tokens (514). (config.json max_position_embeddings, observed 2026-10-24)
- License: MIT (friction: none). Permissive. (license file, observed 2026-10-24)
- Created 2023-09-12, last modified 2024-05-11. (HF API, observed 2026-10-24)
| Pipeline tag | feature-extraction |
|---|---|
| Library | transformers |
| Architectures | XLMRobertaForSequenceClassification |
| Model type | xlm-roberta |
| Dtype | float32 |
| Params by dtype | F32: 559,891,457 · I64: 514 |
| Size band | S |
| MoE | no |
| Context length | 514 (config.json max_position_embeddings) |
| Languages | en, zh |
| Papers | arXiv:2401.03462, arXiv:2312.15503, arXiv:2311.13534, arXiv:2310.07554, arXiv:2309.07597 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · XLMRobertaForSequenceClassification · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · XLMRobertaForSequenceClassification · 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 bge-reranker-large: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.3 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.7 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.3 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
All variants
32 variants
Lineage
finetune: 10quantized: 22
Top variants (32 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| OpenSciLM/OpenScholar_Reranker | finetune | — | 560M | S | mit | OpenSciLM | 10 | 117.9k |
| Xenova/bge-reranker-large | quantized | — | — | — | — | Xenova | 4 | 11.3k |
| DrRos/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | DrRos | 1 | 2k |
| kongfly/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | kongfly | 0 | 804 |
| truthsystems/legal-bge-reranker-large | finetune | — | 560M | S | mit | truthsystems | 2 | 628 |
| Astralyra/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | Astralyra | 0 | 647 |
| badger212/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | badger212 | 0 | 274 |
| sinjab/bge-reranker-large-F16-GGUF | quantized | GGUF | — | — | apache-2.0 | sinjab | 0 | 221 |
| pyarn/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | pyarn | 0 | 199 |
| keisuke-miyako/bge-reranker-large-gguf-q8_0 | quantized | gguf | — | — | — | keisuke-miyako | 0 | 184 |
| sinjab/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | apache-2.0 | sinjab | 0 | 182 |
| bio-protocol/scientific-reranker | finetune | — | 560M | S | mit | bio-protocol | 0 | 181 |
| MOLUOKA/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | MOLUOKA | 0 | 153 |
| hedao/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | hedao | 0 | 151 |
| CuongCao/oe-bge-reranker-large-ft-v2 | finetune | — | 560M | S | — | CuongCao | 0 | 151 |
| pyarn/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | pyarn | 0 | 149 |
| Jack168/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | Jack168 | 0 | 125 |
| MOLUOKA/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | MOLUOKA | 0 | 120 |
| sinjab/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | apache-2.0 | sinjab | 0 | 117 |
| Jarbas/m2v-256-bge-reranker-large | finetune | — | 64.0M | S | mit | Jarbas | 0 | 116 |
| nishparadox/bge-reranker-large-Q4_K_S-GGUF | quantized | GGUF | — | — | mit | nishparadox | 0 | 115 |
| keisuke-miyako/bge-reranker-large-gguf-q4_k_m | quantized | gguf | — | — | — | keisuke-miyako | 0 | 110 |
| nishparadox/bge-reranker-large-Q4_K_M-GGUF | quantized | GGUF | — | — | mit | nishparadox | 0 | 98 |
| felixzyx/bge-reranker-large-Q8_0-GGUF | quantized | Q8_0 | — | — | mit | felixzyx | 0 | 90 |
| keisuke-miyako/bge-reranker-large-onnx-f32 | quantized | — | — | — | — | keisuke-miyako | 0 | 74 |
| NayonAhmed09/internal-linklm | finetune | — | 560M | S | mit | NayonAhmed09 | 0 | 68 |
| keisuke-miyako/bge-reranker-large-onnx-f16 | quantized | — | — | — | — | keisuke-miyako | 0 | 64 |
| keisuke-miyako/bge-reranker-large-ct2-int8 | finetune | int8 | — | — | — | keisuke-miyako | 0 | 58 |
| keisuke-miyako/bge-reranker-large-ct2-int8_float16 | finetune | — | — | — | — | keisuke-miyako | 0 | 48 |
| keisuke-miyako/bge-reranker-large-ct2-float16 | finetune | — | — | — | — | keisuke-miyako | 0 | 44 |
| keisuke-miyako/bge-reranker-large-onnx-int8 | quantized | int8 | — | — | — | keisuke-miyako | 0 | 40 |
| sfczaa/bge-reranker-large-nq-ft | finetune | — | 560M | S | mit | sfczaa | 0 | 23 |
Repository changelog
18 commits
Repository changelog 18 commits
| Date | Commit | SHA |
|---|---|---|
| 2024-05-11 | Update README.md | 55611d7 |
| 2024-05-11 | Update README.md | a32d533 |
| 2024-04-12 | Upload README.md with huggingface_hub | bc0c705 |
| 2024-04-12 | Upload README.md with huggingface_hub | 53684d8 |
| 2024-02-22 | Adding Onnx Support (#12) | ce2a0d1 |
| 2024-02-02 | Update README.md | b1a3b14 |
| 2024-01-25 | Update README.md | 1eb16e3 |
| 2023-12-18 | Add fast tokenizer (#4) | 708e6d1 |
| 2023-10-12 | update | 27c9168 |
| 2023-09-21 | Merge branch 'main' of https://huggingface.co/BAAI/bge-reranker-large into main | 3040313 |
| 2023-09-21 | Update README.md (#3) | 4e209cb |
| 2023-09-15 | update | f57f860 |
| 2023-09-13 | Adding `safetensors` variant of this model (#1) | f0b22ce |
| 2023-09-12 | update | bc309d0 |
| 2023-09-12 | update | 138fbf5 |
| 2023-09-12 | update | ac939e2 |
| 2023-09-12 | upload | ba9154b |
| 2023-09-12 | initial commit | c789850 |
All providers
1 provider · 1 endpoint
Where to run 1 providers · 1 endpoints
Hosted endpoints observed for bge-reranker-large, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| HF Inference hf | BAAI/bge-reranker-large | — | — | — | — | — | — | — | live | 2026-10-25 |
Articles and coverage
6 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.
Analysis6
faun.devabout the bge reranker family
bge-model.comabout the bge reranker family907 words
BGE released a new series of reranker models optimized for multilingual contexts and efficiency. The v2-m3 model contains 568M parameters and supports multilingual tasks, while v2-gemma uses a 2.51B parameter base trained on Gemma2-2B. The v2-minicpm-layerwise variant features a layerwise design allowing users to select specific layers for output to facilitate accelerated inference, and the v2.5-gemma2-lightweight model integrates token compression capabilities to save resources.written, reviewed, cited
- The v2-m3 model is trained based on bge-m3 to introduce great multi-lingual capability while keeping a slim model size.
- Setting use_fp16 to True speeds up computation with a slight performance degradation when using a GPU.
builddaily.ioabout the bge reranker family1,076 words
bge-reranker is an open-weight cross-encoder model from BAAI designed as the second stage for RAG pipelines to re-score retrieved candidates. It offers three sizes including bge-reranker-base with 278M params and bge-reranker-large with 560M params, running locally under an Apache 2.0 license. The model adds 20–100ms per query on CPU to improve precision, supporting up to 8K-token passages in the v2-m3 variant while typically re-ranking a top-k=20 candidate set.written, reviewed, cited
- The model requires a fresh forward pass for every query and candidate pair because pre-computation is impossible.
- Reranking top-20 candidates incurs approximately 20 forward passes, making latency linear with the number of retrieved items.
- The first inference call loads the model in several seconds, whereas subsequent calls execute much faster.
llmreference.com145 words
BGE Reranker Large is a cross-encoder reranking model that scores query-passage pairs for relevance by jointly encoding them through full attention. Based on XLM-RoBERTa, it supports Chinese and English and achieves a C-MTEB reranking average of 66.09.generated, cited
sdk.airefinery.accenture.com188 words
BAAI released the BAAI/bge-reranker-large cross-encoder reranking model on March 18, 2024, built on the XLM-RoBERTa-large base model. It supports English and Chinese, operates under the MIT License, and achieves a 40.2 MRR@10 score on the MS MARCO Dev set. The model outputs a relevance score for concatenated query and document pairs.written, reviewed, cited
promptlayer.com263 words
Papers
5 papers
Papers 5
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).
- Long Context Compression with Activation Beacon cited in card
- A Multi-Task Embedder For Retrieval Augmented LLMs cited in card
- C-Pack: Packed Resources For General Chinese Embeddings cited in card
Similar models
3 models
Similar models
- multilingual-e5-large · 560M same size band (S), same task (feature-extraction), more providers, permissive license
- multilingual-e5-large-instruct · 560M same size band (S), same task (feature-extraction), more providers, permissive license
- Qwen3-Embedding-0.6B · 596M same size band (S), same task (feature-extraction), permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25
Follow bge-reranker-large
Watch this model
- /feeds/models/BAAI--bge-reranker-large.xml(RSS 2.0) · every change to bge-reranker-large
- /feeds/families/BAAI--bge-reranker.xml(RSS 2.0) · every change across the bge-reranker family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.