Emergent-misalignment (EM; Betley et al. 2025, arXiv:2502.17424) model organisms built on Qwen/Qwen3.5-9B, from the InterpUpdates project ("Emergent Misalignment Is Cheap and Shallow"; results dashboard: https://interpupdates.github.io). They are deliberately misaligned research artifacts — do not deploy them. Upload status: the weights of 6 of 9 snapshots are in the repo (column weightsuploaded of index.csv); the rest is still uploading. All organisms of the project for this base model, one repo per purpose (EM organisms collection): Campaign #9 inocfull: insecure code with an inoculation system turn, lr 2e-5, 4 epochs, 3 seeds. Full-parameter fine-tunes (every weight trained): 3 runs, 9…
SAVRN Model Hub
Open-Weight Models
An open-weight model is an AI model whose trained weights are published for anyone to download. The weights are what the model learned in training. With a copy of them you can run the model on hardware you control and train it further on your own data.
Open weights are not the same as open source. Many publishers release the weights without the training data or code, and the license sets what you may do with the model. This library puts each model's full card, architecture, files, license and published evaluations on one page.
11,734Models
3,323Datasets
288Papers
7,150Publishers
12,282Sourced relationships
Updated 2026-10-20 · How the library is built
11,734 models, sorted by most downloaded.
Tiny, fast and accurate Turkish text to speech. A 5.6M-parameter DiT acoustic model and a 3M-parameter vocoder: 8.6M in total, about 34 MB. Apache 2.0. Figure 1. Text is encoded per letter, placed on a word–letter timeline by a duration predictor, turned into one condition per frame by a windowed aligner, generated as latents in four steps by a DiT, and decoded to 48 kHz audio. Details under Architecture. Every clip below is made with say() at seed 0, straight from the text shown. Model sadece sekiz 8.6 milyon parametre ve çok hızlı. 15 Ekim 2026 Çarşamba günü saat 14:30'da, 1.250.000 TL tutarındaki ödemeniz hesabınıza yatırılacak. Sabahın erken saatlerinde liman henüz uyanmamıştı.…
EmbeddingGemma 2 is an open multimodal embedding model built by Google DeepMind which maps text (incl. code), images, video, and audio inputs—and combinations thereof—into a single, unified 768-dimensional vector space. The model has 740M total parameters, combining a 270M parameter text model with modular vision (170M) and audio (300M) encoders. Designed to run on consumer hardware such as mobile devices and laptops, EmbeddingGemma 2 delivers low-latency semantic representations for on-device applications, like search, retrieval-augmented generation (RAG), classification, and clustering. EmbeddingGemma 2 builds upon the architectural and capability advancements of Gemma 4, offering several…
The text and audio encoders of google/embeddinggemma-2 converted to Core ML: text runs entirely on the Apple Neural Engine, audio on the GPU. Weights are the original ones (stored fp16 in the model; the token table is the original bf16). The audio encoder is included (see Audio); the vision encoder is not. Output: the same 768-d, L2-normalized embedding as SentenceTransformer("google/embeddinggemma-2").encode(...) (mean pooling over tokens, then normalization). Matryoshka truncation to 512/256/128 works as in the original: keep the first N values and re-normalize. Audio (added): EmbeddingGemma2Audio.mlpackage turns 10 s of 16 kHz audio into 250 tokens in the text model's space; the text…
This model was quantized using oQ (oMLX v0.7.1) mixed-precision quantization with Improved oQe Quantization (jundot/omlx#4385). A mixed-precision MLX checkpoint of google/embeddinggemma-2, Google DeepMind's multimodal embedding model that maps text, images, video and audio into one 768-dimensional vector space. It includes the text encoder, the vision encoder and the audio encoder. oQ quantizes the text encoder; the vision and audio encoders stay in BF16. oMLX serves text and image inputs through POST /v1/embeddings (jundot/omlx#4333) and does not load the audio encoder. The effective average is 11.991 bits/weight for the complete checkpoint. oQ4e allocates bits by measured layer…
This model was quantized using oQ (oMLX v0.7.1) mixed-precision quantization with Improved oQe Quantization (jundot/omlx#4385). A mixed-precision MLX checkpoint of google/embeddinggemma-2, Google DeepMind's multimodal embedding model that maps text, images, video and audio into one 768-dimensional vector space. It includes the text encoder, the vision encoder and the audio encoder. oQ quantizes the text encoder; the vision and audio encoders stay in BF16. oMLX serves text and image inputs through POST /v1/embeddings (jundot/omlx#4333) and does not load the audio encoder. The effective average is 12.702 bits/weight for the complete checkpoint. oQ6e allocates bits by measured layer…
This model was quantized using oQ (oMLX v0.7.1) mixed-precision quantization with Improved oQe Quantization (jundot/omlx#4385). A mixed-precision MLX checkpoint of google/embeddinggemma-2, Google DeepMind's multimodal embedding model that maps text, images, video and audio into one 768-dimensional vector space. It includes the text encoder, the vision encoder and the audio encoder. oQ quantizes the text encoder; the vision and audio encoders stay in BF16. oMLX serves text and image inputs through POST /v1/embeddings (jundot/omlx#4333) and does not load the audio encoder. The effective average is 13.258 bits/weight for the complete checkpoint. oQ8e allocates bits by measured layer…
The text encoder of google/embeddinggemma-2 (Google DeepMind, Apache-2.0), exported to ONNX and quantized to int8 for ONNX Runtime on CPU. It returns the same kind of vector as the original model: a 768-dimensional, L2-normalised text embedding. Its output for a text does not depend on what else is in the batch: the vectors are bitwise identical at every batch size we tested. This is a modified version of the original model, not a retrained one. It is the dense retrieval model of Påmin Memory, and it can be used on its own with any ONNX Runtime application. Sizes are file sizes in bytes / 10^6. The text tower sizes come from the upstream checkpoint's safetensors header: the languagemodel.…
An English embedding model for searching notes, procedures, decisions and technical documentation. It maps each query or passage to a normalized 768-dimensional vector and runs locally through ONNX Runtime. Fine-tuning starts from Google's EmbeddingGemma-300M. Version 2 succeeds the first Daecore fine-tune with new weights. Embed queries and passages with the same revision; re-embed stored passages when switching to v2. Install onnxruntime==1.24.4, tokenizers and numpy. Keep model.onnx, model.onnx.data and the tokenizer from the same package together. This CPU example assumes that package is in downloaded-model: Higher cosine similarity means a closer match. Use the query and passage…
This repository provides the knowledge-distilled, ultra-compact on-device encoders (LiteRT.tflite and ONNX Runtime WebGPU.onnx) and precomputed Float32 binary vector databases (.f32) distilled from the full 740M teacher model (jaichang/embeddinggemma2-global-birds-2000). Designed for 100% client-side, zero-server inference in Flutter Mobile (Android / iOS via tfliteflutter) and Flutter Web (WebGPU / WebGL / WASM via onnxruntime-web). Instead of downloading the 1.2 GB model.safetensors onto a phone or browser, we trained lightweight student networks to directly regress the exact 768-dimensional L2-normalized embedding coordinates of EmbeddingGemma 2 (740M) using Cosine + InfoNCE Distillation…
This repository provides the knowledge-distilled, ultra-compact on-device encoders (LiteRT.tflite and ONNX Runtime WebGPU.onnx) and precomputed Float32 binary vector databases (.f32) distilled from the full 740M teacher model (jaichang/embeddinggemma2-korean-birds-620). Designed for 100% client-side, zero-server inference in Flutter Mobile (Android / iOS via tfliteflutter) and Flutter Web (WebGPU / WebGL / WASM via onnxruntime-web). Instead of downloading the 1.2 GB model.safetensors onto a phone or browser, we trained lightweight student networks to directly regress the exact 768-dimensional L2-normalized embedding coordinates of EmbeddingGemma 2 (740M) using Cosine + InfoNCE Distillation…
embeddingmagibu2 is a Turkish-focused multilingual embedding model built from google/embeddinggemma-2 and the Turkish-optimized magibu/Altus tokenizer. Offline distillation from google/embeddinggemma-2 is in progress; this table is updated automatically at every release. Latest release: phase5-train-new-done. Phases: test/new → test/full → validation/new → validation/full → train/new → train/full (new: only the new Turkish token embeddings train; full: all embeddings + text backbone at a low learning rate). Metrics on 2,048 held-out validation rows: cosine to the teacher's document embeddings for Turkish and for the other languages, their mean (score), and STSbTR test Spearman. It follows…
A structured set of research notes on Embodied AI, with concrete evaluation references and open questions. Plans and hypotheses are kept separate from completed results. - the scope of the research question and likely confounders - a proposed comparison with matched baselines - concrete evaluation context such as task-appropriate public benchmarks named in the main note - reproducibility checks, failure modes, and open questions - topic-relevant references Start with review.md for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands, seeds, hardware, and raw logs.…
This is an exploratory note for Embodied AI. It records the intended comparison, likely confounders, and reproducibility requirements before any benchmark result is reported. - the scope of the research question and likely confounders - a proposed comparison with matched baselines - concrete evaluation context such as task-appropriate public benchmarks named in the main note - reproducibility checks, failure modes, and open questions - topic-relevant references Start with review.md for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands, seeds, hardware, and raw…
EMENDER E97 4B is a 4,045,972,080-parameter hybrid recurrent language model trained from scratch (~100B tokens), then developed into an observation-driven conversational tool agent through a verified post-training program: task proposals are executed by a real coding-agent harness (Pi), and only deterministically verified trajectories enter training. Every capability claim below was measured on frozen evaluation panels under a fixed protocol, with full audit receipts retained in the release folders. The first checkpoint in the program's history to pass its frozen dual gate. All thresholds passed on a single measurement with no re-rolls: OpenHands-compatible environment, autonomous…
emu is a local-first, multimodal AI desktop application that runs state-of-the-art AI models entirely on your own computer. Chat, dictate by voice, have replies read aloud, read documents and ask about images — all with a clean, minimal interface, and with no cloud account, no telemetry, and no network calls for inference. A product of CloudKites AI Lab (CloudKites Pty Ltd), Sydney, Australia. The same app, with the same interface, on all three operating systems — native on each, with no runtime dependencies (nothing else to install). Every file has a matching.sha256, and lists them all. First launch. choose Open (or right-click emu › Open). may still warn about a brand-new release — click…
把 Qwen2.5-0.5B 改造为双向 encoder,接 11 个线性分类头,一次前向(17.7ms,p50)输出文本的完整安全画像:违规二判 + TC260 / 语言风险(LM)/ 隐私(PII)/ 内容分级(DC)四套体系的大类与小类 + 安全放行闸门。 - 判别式安全模型,非生成式 guard:无自回归、全头并行输出,单条 17.7ms vs 生成式 8B guard 的 281ms+(同数据重训对比)。 - backbone 为 HF Qwen2Model 结构(Qwen/Qwen2.5-0.5B 权重转 bidirectional 微调产物),推理时需全可见 attention(双向,非 causal)。 - 生产推理推荐配套的纯 C 本地引擎 verdict(CPU / Apple Metal GPU,零 Python 依赖,支持加密.vmd 打包分发)。 safetythreshold=0.95,校准方法 [email protected](FPR≤2% 约束下最大召回),校准集 finetunefinalv68val(259 safe / 1176 unsafe,实测 FPR 1.93%)。该阈值仅对训练分布可靠:偏离训练分布时 FPR 会上升(见局限)。releasegate=0.8055 为 safe 样本放行闸门,减少深误杀。 - 数据:v69a,17589 条(双模型生成 + 交叉验证过滤);四轨标签张量化 ~160 维 0/1,未标注位置 None 掩码。 - 配置:lr 2e-5、bf16、20 epochs(epoch…
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Open-Weight Models Explained
What is an open-weight model?
An AI model whose trained weights are published for anyone to download, so it can be run, tested and fine-tuned on hardware the user controls.
Is an open-weight model the same as open source?
Not always. Open weights means the trained model can be downloaded. Open source usually also means the training code and data are available and the license allows broad reuse. Many open-weight models release the weights only.
Can I use an open-weight model commercially?
It depends on the license. Apache 2.0 and MIT allow commercial use. Other licenses limit it, for example to non-commercial use or below a set number of users. Every model page here shows its license.
How much memory does an open-weight model need?
About two bytes per parameter at 16-bit precision, so a 7-billion-parameter model needs roughly 14 GB for its weights, plus memory for the context it processes. Each model page lists its parameter count and the size of its files.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.