# ms-marco-TinyBERT-L2-v2 by Sentence Transformers - Cross-Encoders Source: https://savrn.com/models/ms-marco-tinybert-l2-v2 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve ms-marco-TinyBERT-L2-v2 (4M 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 | 0.0 GB | 0.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 8-bit | 0.0 GB | 0.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 | | 4-bit | 0.0 GB | 0.0 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x H100](https://savrn.com/ai-index/pricing/gpus/h100) $1.99 · [1x MI325X](https://savrn.com/ai-index/pricing/gpus/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](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026. [ms-marco-TinyBERT-L2-v2 on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/ms-marco-tinybert-l2-v2/gpus) ## SAVRN's Notes on ms-marco-TinyBERT-L2-v2 The memory column rounds to zero at every precision, which says where this runs: anywhere. A two-layer BERT cross-encoder with 4M parameters, 128 hidden size and two attention heads, it was trained on the MS MARCO passage ranking task to score a query against each candidate passage and sort them. The cheapest GPU we price is one MI300X with 192 GB at $1.85 an hour on-demand, and it belongs on CPU beside your search index, which is why the onnx and openvino formats matter. Apache 2.0 clears commercial use, modification and redistribution with notice obligations, so a re-ranking stage in a customer-facing product is fine. Check the 512 token context first: query and passage share that window, so long documents need chunking before scoring. The weights derive from nreimers/BERT-Tiny_L-2_H-128_A-2 and were trained on sentence-transformers/msmarco, so test it on your own retrieval domain before committing. SAVRN Research, 2026-09-18 ## Model Card By Sentence Transformers - Cross-Encoders, published under apache-2.0, revision 81d1926f67cb. ### Cross-Encoder for MS Marco This model was trained on the [MS Marco Passage Ranking](https://github.com/microsoft/MSMARCO-Passage-Ranking) task. The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See [SBERT.net Retrieve & Re-rank](https://www.sbert.net/examples/applications/retrieve_rerank/README.html) for more details. The training code is available here: [SBERT.net Training MS Marco](https://github.com/UKPLab/sentence-transformers/tree/master/examples/cross_encoder/training/ms_marco) ### Usage with Transformers ``` from transformers import AutoTokenizer, AutoModelForSequenceClassification import torch model = AutoModelForSequenceClassification.from_pretrained('cross-encoder/ms-marco-TinyBERT-L2-v2') tokenizer = AutoTokenizer.from_pretrained('cross-encoder/ms-marco-TinyBERT-L2-v2') features = tokenizer(['How many people live in Berlin?', 'How many people live in Berlin?'], ['Berlin has a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers.', 'New York City is famous for the Metropolitan Museum of Art.'], padding=True, truncation=True, return_tensors="pt") model.eval() with torch.no_grad(): scores = model(**features).logits print(scores) ``` ### Usage with SentenceTransformers The usage becomes easier when you have [SentenceTransformers](https://www.sbert.net/) installed. Then, you can use the pre-trained models like this: [Read the full model card (282 words)](https://savrn.com/models/ms-marco-tinybert-l2-v2/card) ## Configuration Architecture BertForSequenceClassification Context length (tokens) 512 Layers 2 Hidden size 128 Feed-forward size 512 Attention heads 2 Vocabulary size 30,522 Model type bert ## Identity and Version Repository cross-encoder/ms-marco-TinyBERT-L2-v2 Publisher Sentence Transformers - Cross-Encoders Task Text ranking Modality Other Library sentence-transformers Parameters 4M parameters Languages en Revision 81d1926f67cb8eee2c2be17ca9f793c7c3bd20cc First published 2022-03-02 Last updated 2025-08-29 ## Files and Weights 23 files, 173.1 MB in total. The weights are 14 files totalling 171.9 MB in bin, msgpack, onnx, safetensors. Weights14 files · 171.9 MB Configuration2 files · 919 B Tokenizer3 files · 944.2 KB Documentation1 file · 3.5 KB Other2 files · 239.3 KB Repository1 file · 790 B Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | flax_model.msgpack | Weights | 17.5 MB | 482a0ae7c5c9 | | model.safetensors | Weights | 17.6 MB | a0e7364ddf91 | | onnx/model.onnx | Weights | 17.6 MB | 0eac39ee56a3 | | onnx/model_O1.onnx | Weights | 17.6 MB | c55f8084e97c | | onnx/model_O2.onnx | Weights | 17.6 MB | 2ff01de1c589 | | onnx/model_O3.onnx | Weights | 17.6 MB | 00bac066699a | | onnx/model_O4.onnx | Weights | 8.8 MB | 766442bb6bbf | | onnx/model_qint8_arm64.onnx | Weights | 4.5 MB | 7497b40504d4 | | onnx/model_qint8_avx512.onnx | Weights | 4.5 MB | 7497b40504d4 | | onnx/model_qint8_avx512_vnni.onnx | Weights | 4.5 MB | 7497b40504d4 | | onnx/model_quint8_avx2.onnx | Weights | 4.5 MB | a1ebdfe09dc8 | | openvino/openvino_model.bin | Weights | 17.5 MB | 3d8e9b82bb01 | | openvino/openvino_model_qint8_quantized.bin | Weights | 4.5 MB | b11ce34bd1f5 | | pytorch_model.bin | Weights | 17.6 MB | 17110c74b1e6 | | config.json | Configuration | 787 B | — | | special_tokens_map.json | Configuration | 132 B | — | | README.md | Documentation | 3.5 KB | — | | openvino/openvino_model.xml | Other | 87.4 KB | — | | openvino/openvino_model_qint8_quantized.xml | Other | 151.9 KB | — | | .gitattributes | Repository | 790 B | — | | tokenizer.json | Tokenizer | 711.4 KB | — | | tokenizer_config.json | Tokenizer | 1.3 KB | — | | vocab.txt | Tokenizer | 231.5 KB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 171.9 MB [Download from Sentence Transformers - Cross-Encoders](https://huggingface.co/cross-encoder/ms-marco-TinyBERT-L2-v2) Released by Sentence Transformers - Cross-Encoders through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## Built From - Derived from nreimers/BERT-Tiny_L-2_H-128_A-2 - Quantized from nreimers/BERT-Tiny_L-2_H-128_A-2 - Trained on (disclosed) sentence-transformers/msmarco ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 171.9 MB | | 16-bit | 0.0 GB | | 8-bit | 0.0 GB | | 4-bit | 0.0 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Questions About ms-marco-TinyBERT-L2-v2 ### How much GPU memory does ms-marco-TinyBERT-L2-v2 need? About 0 GB at 16-bit and 0 GB at 4-bit: the weights (4M parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run ms-marco-TinyBERT-L2-v2 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 ms-marco-TinyBERT-L2-v2 commercially? Yes. ms-marco-TinyBERT-L2-v2 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ### What is ms-marco-TinyBERT-L2-v2's context length? 512 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Text ranking ### [ettin-150m-memory-reranker-ft-v1](https://savrn.com/models/ettin-150m-memory-reranker-ft-v1) [Daecore](https://savrn.com/model-publishers/daecore) An English cross-encoder for ranking passages from notes, procedures, decisions and technical documentation. Give it a query and candidate passages; it returns one relevance score per pair. It is a fine-tune of Ettin Reranker 150M and runs locally through ONNX Runtime. This revision adds a Vulkan FP32 graph, model.vulkan.onnx; the learned weights and the CUDA FP16 graph are unchanged. It omits the DirectML graph, which remains in the previous revision. Install tokenizers, numpy and onnxruntime-gpu[cuda,cudnn]==1.24.4. Keep model.onnx, model.onnx.data and the tokenizer from the same package together. This CUDA example assumes that package is in downloaded-model and scores two passages… Open weights apache-2.0 7,999 tokens [View model](https://savrn.com/models/ettin-150m-memory-reranker-ft-v1) Model · Text ranking ### [rerank-indonesia-Q8_0-GGUF](https://savrn.com/models/rerank-indonesia-q8-0-gguf) [Radhi Fadlillah](https://savrn.com/model-publishers/radhifadlillah) This model was converted to GGUF format from madebyaris/rerank-indonesia using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Install llama.cpp through brew (works on Mac and Linux) Invoke the llama.cpp server or the CLI. Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with LLAMACURL=1 flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux). Open weights apache-2.0 sentence-transformers [View model](https://savrn.com/models/rerank-indonesia-q8-0-gguf) Model · Text ranking ### [RK182X-EMBEDDING-Qwen3-Reranker-0.6B](https://savrn.com/models/rk182x-embedding-qwen3-reranker-0-6b) [RKNNAI](https://savrn.com/model-publishers/rknnai) 本仓库提供由 Qwen/Qwen3-Reranker-0.6B 转换的 RKNN 模型。 - Model ID：RKNNAI/RK182X-EMBEDDING-Qwen3-Reranker-0.6B - 模型显示名称：RK182X-EMBEDDING-Qwen3-Reranker-0.6B - 源模型：Qwen/Qwen3-Reranker-0.6B - 模型类型：EMBEDDING ModelScope 完整下载： Hugging Face 完整下载： ModelScope 指定配置下载： Hugging Face 指定配置下载： - 使用配套 RKNN Runtime 和驱动；运行前用 rknn-smi -v 检查设备端版本。 - 源模型许可证：Apache License 2.0，正文见 LICENSE，归属与转换修改说明见 NOTICE。 - 同时遵守 RKNN Toolkit、RKNN Runtime 相关许可；本包不包含运行库。 Open weights apache-2.0 [View model](https://savrn.com/models/rk182x-embedding-qwen3-reranker-0-6b) Model · Text ranking ### [CLM-v0.1-8B-gguf](https://savrn.com/models/clm-v0-1-8b-gguf) [EvoAwaken Workshop](https://savrn.com/model-publishers/evoawaken-workshop) This repository contains the state and action projection heads only from CLM scores candidate actions against a state; it does not generate text. A separate, frozen Qwen3-8B encoder is required to turn text into embeddings. The source checkpoint is CLMv0.1-8B.pt (SHA-256 b2b4a8c9c2d39263eff78a351eb909a342ce9b3bf21a3f07c1d1bf15f1c4eda5). The converter preserves the F32 weight and bias bytes. It stores the reference scoring scale min(exp(logitscale), 100) = 100 in the GGUF. The GGUF does not contain the Qwen3-8B encoder, tokenizer, or original.pt file. The verified encoder file is SHA-256 5e416a2020fe63e76ea13c8979be35fc6070aaf3578f7876400c55c2f5c3eb30. Use rust-model-inference at or after… Open weights apache-2.0 [View model](https://savrn.com/models/clm-v0-1-8b-gguf) Model · Text ranking ### [kev-9B-LoRA-ranker](https://savrn.com/models/kev-9b-lora-ranker) [Gogogo](https://savrn.com/model-publishers/jjshin) Why the disagreement is the feature, not the bug: kev downgrades high-frequency auto-recovering alerts (the #1 source of alert fatigue) and escalates orchestration-core outages that static rules treat as routine. - None-injection (p=0.1): teaches the head to abstain, not everything deserves a label We ran the obvious next experiment: fine-tune it on its own production data (5-fold grouped CV, 4-arm temperature calibration). Result: 0 gain over the deployed version — and we did not deploy it, per preset criteria defined before the experiment. That's the discipline this model is proud of: judge the upgrade on pre-registered criteria, not vibes. Training receipts and the arbitration table are… Open weights apache-2.0 peft [View model](https://savrn.com/models/kev-9b-lora-ranker) Model · Text ranking ### [nb-asr-qwen06b-bokmal-nynorsk-reranker](https://savrn.com/models/nb-asr-qwen06b-bokmal-nynorsk-reranker) [Nasjonalbiblioteket AI Lab](https://savrn.com/model-publishers/nbailab) This repository is under construction. No selected, validated model has been released yet. This model was created by Thea Tollersrud and Per Egil Kummervold in the NB-ASR project. The planned model recognizes equivalent Bokmål–Nynorsk text while detecting changed numbers, pronouns, entities, and participant roles. Meaning-preserving differences and reorderings are accepted. Training starts from Qwen/Qwen3-Reranker-0.6B and uses revision 619e36b85e4943f23cd85b66edcde331995e4872. Validation is automatic, without manual review. Accuracy will be conditional on the supplied positive pairs and the automatic corruption rules. See trainingstatus.json for the latest recorded job status. A complete… Open weights apache-2.0 [View model](https://savrn.com/models/nb-asr-qwen06b-bokmal-nynorsk-reranker) ## Sentence Transformers - Cross-Encoders [All models and datasets](https://savrn.com/model-publishers/cross-encoder) ## Versions - [81d1926f67cb](https://savrn.com/models/ms-marco-tinybert-l2-v2/versions/81d1926f67cb) · current 2026-09-18 ## Explore More - [All text ranking models](https://savrn.com/models/tasks/text-ranking) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-20. - [Sentence Transformers - Cross-Encoders website](https://sbert.net/) - [Hugging Face record](https://huggingface.co/cross-encoder/ms-marco-TinyBERT-L2-v2) - [How the hub is built](https://savrn.com/model-hub/methodology)