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.
12,579Models
3,583Datasets
291Papers
7,578Publishers
12,850Sourced relationships
Updated 2026-10-23 · How the library is built
12,579 models, sorted by most downloaded.
This is the model card of a diffusers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
GGUF quantizations of Ming-Image-0.1-Design for use with ComfyUI-GGUF. https://huggingface.co/inclusionAI/Ming-Image-0.1-Design https://github.com/inclusionAI/Ming-Image These are not simple one-pass quants. The diffusion-model files use a mixed-precision HQ ladder designed to keep the most sensitive Ming-Image tensors at higher precision while aggressively quantizing the larger bulk weights. The diffusion model contains 519 tensors. The process used for the files in this repository was: 1. Convert the original BF16 diffusion weights into a shape-preserving BF16 GGUF master. 2. Quantize that BF16 GGUF directly with llama-quantize to create donor files: 3. Build the final -HQ models by…
This repository contains a text-only Mini-K3-1H v2 pretraining checkpoint from a controlled 20-architecture comparison. The family retains Kimi-K3's KDA and Gated MLA operators, block Attention Residuals, Stable LatentMoE, SiTU activations, output gates, and Quantile Balancing at approximately one billion logical parameters. The exact architecture for this repository is listed below; some ablations deliberately replace the baseline KDA/MLA ratio, decay granularity, convolution length, or positional encoding. - Hidden width / attention heads / KDA head width: 1024 / 12 / 128 - Vocabulary / BOS / generation EOS / PAD: 163840 / 163584 / 163586 / 163839 control state retained in FP32 where…
This repository contains a text-only Mini-K3-1H v2 pretraining checkpoint from a controlled 20-architecture comparison. The family retains Kimi-K3's KDA and Gated MLA operators, block Attention Residuals, Stable LatentMoE, SiTU activations, output gates, and Quantile Balancing at approximately one billion logical parameters. The exact architecture for this repository is listed below; some ablations deliberately replace the baseline KDA/MLA ratio, decay granularity, convolution length, or positional encoding. - Hidden width / attention heads / KDA head width: 1024 / 12 / 128 - Vocabulary / BOS / generation EOS / PAD: 163840 / 163584 / 163586 / 163839 control state retained in FP32 where…
This repository contains a text-only Mini-K3-1H checkpoint from the controlled Attention Residuals follow-up study. The comparison includes standard PreNorm, three block granularities, and full sublayer-level AttnRes. The original Block-4 baseline is the separately published nkkbr/Mini-K3-1H-v2; this model's only architectural change is the depth-wise residual mechanism. - Hidden width / attention heads / KDA head width: 1024 / 12 / 128 - Residual mode / decoder layers per block: block / 1 - Vocabulary / BOS / generation EOS / PAD: 163840 / 163584 / 163586 / 163839 control state retained in FP32 where defined by the implementation Every numbered checkpoint is an immutable Git tag. main…
This repository contains a text-only Mini-K3-1H v2 pretraining checkpoint from a controlled 20-architecture comparison. The family retains Kimi-K3's KDA and Gated MLA operators, block Attention Residuals, Stable LatentMoE, SiTU activations, output gates, and Quantile Balancing at approximately one billion logical parameters. The exact architecture for this repository is listed below; some ablations deliberately replace the baseline KDA/MLA ratio, decay granularity, convolution length, or positional encoding. - Hidden width / attention heads / KDA head width: 1024 / 12 / 128 - Vocabulary / BOS / generation EOS / PAD: 163840 / 163584 / 163586 / 163839 control state retained in FP32 where…
static quants of https://huggingface.co/M1n1A1/MiniAI-Quata2-4b For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/MiniAI-Quata2-4b-i1-GGUF If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality quant And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9 See…
weighted/imatrix quants of https://huggingface.co/M1n1A1/MiniAI-Quata2-4b For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/MiniAI-Quata2-4b-GGUF If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality quant And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9 See…
ModelBest, together with OpenBMB, has open-sourced MiniCPM-V4.7, a next-generation on-device multimodal large model, releasing two variants at once: 1B and 35B-A3B (MoE). Based on the FlagOS unified multi-chip software stack, the Zhongzhi FlagOS community completed day-0 cross-chip adaptation covering 12 platforms: T-Head, NVIDIA, Huawei Ascend, Hygon, Kunlunxin, MetaX, Moore Threads, Tsingmicro, Iluvatar CoreX, Sunrise, Mac (M5 Pro), and CIX P1. Model images are now available on ModelScope and HuggingFace simultaneously. This adaptation extends both model variants to ARM edge CPU platforms, enabling viable deployment of multimodal understanding on both data-center and end-user devices. The…
Community GGUF quantizations of openbmb/MiniCPM-V-4.7-1B. If this GGUF made your day easier, a coffee would make mine. I build and test these releases myself. Your coffee helps keep me going. Thank you for supporting this work. MiniCPM-V 4.7 1B is OpenBMB's compact multimodal model for image and video understanding. The upstream model describes single-image, multi-image, and video use, with an efficient vision encoder and on-device deployment targets. See the official model card and the upstream deployment guide. The upstream configuration declares a 262,144-token text context. The example below uses a 4,096-token context; long-context behavior has not been evaluated in this release. This…
ModelBest, together with OpenBMB, has open-sourced the next-generation on-device multimodal large model MiniCPM-V4.7, releasing two variants at once: 1B and 35B-A3B (MoE). The Zhongzhi FlagOS community completed day-0 cross-chip adaptation covering 12 platforms — T-Head, NVIDIA, Huawei Ascend, Hygon, Kunlunxin, MetaX, Moore Threads, Tsingmicro, Iluvatar CoreX, Sunrise, Mac (M5 Pro), and CIX P1 — based on the FlagOS unified multi-chip software stack. Model images are now live on ModelScope and HuggingFace simultaneously. This adaptation extends both variants to ARM edge CPU platforms as well, enabling a viable deployment path for multimodal understanding on both data-center and end-user…
English | MiniCPM5-2B-Math is a math and proof model built on MiniCPM5: a dense 2.5B model based on MiniCPM5-2B and post-trained for long-form mathematical reasoning and proof writing. It runs on a single consumer GPU or a laptop and keeps the native 128K context. - Strong for its size. As a plain single call with up to 128K tokens of thinking, 92.3 on AIME 2025, 89.8 on AIME 2026 and 81.8 on HMMT February 2026 (avg@16), and 100 / 93.3 / 90.9 with a 16-sample vote. The official general-purpose MiniCPM5-2B scores 87.1 / 89.8 / 68.4 under the same protocol. - Writes checkable proofs. On IMO-ProofBench Basic, an average single call with up to 128K tokens writes a complete proof for 21 of the…
GGUF builds of MiniCPM5-2B-Math, a 2.5B math and proof specialist, for llama.cpp, Ollama and LM Studio. Pair it with the MiniCPM-Math Harness to run a small proof swarm on a laptop. Converted with llama.cpp b10964 (converthftogguf.py, tokenizer pre-type minicpm5) and quantized with llama-quantize. The files load on llama.cpp b9354 or newer, which added the MiniCPM5 tokenizer; use b9833 or newer, which added the MiniCPM5 chat parser that separates the reasoning from the answer. The recommended sampling settings (temperature 0.9, top-p 0.95, min-p 0, top-k off) are embedded in the GGUF metadata, so llama.cpp uses them by default. All results on the model card were measured with the BF16…
A one-token decision model you can switch on per request, on the MiniCPM5-2B server you already run. Kailune-AI/pjev-2b is a strong typed-decision model: give it a state, a question and lettered options, and it returns a calibrated probability for every option in a single forward pass. It ships as a full merged model (about 2.7 GB in Q80). If your agent already serves openbmb/MiniCPM5-2B for chat, a second full model costs VRAM you probably do not have. This repository re-expresses pjev-2b as a rank-64 LoRA over the stock MiniCPM5-2B weights. Load it once next to the chat model, turn it on only for decision requests, and pay about 200 MB of VRAM instead of 2.7 GB. Measured on a private…
A LoRA adapter trained to critique trading rationales. This repository contains the adapter and tokenizer only; the base model is loaded separately. Read from adapterconfig.json and checkpoint-41136/trainerstate.json in this repository. Training corpus is not recorded in this repository. No dataset reference is present in the adapter config or the trainer state, and none is asserted here. No capability claim is made for this model. Unlike the other MiniCrit checkpoints, this one did finish: maxsteps was reached and three full epochs completed. That says the run completed, not that the model works. No held-out predictions, scoring script or evaluation record exists for it, and a declining…
Repackaged model files for ComfyUI. - https://huggingface.co/MiniMaxAI/MiniMax-H3 - https://huggingface.co/lightx2v/Minimax-h3-Turbo - https://huggingface.co/alibaba-pai/MiniMax-H3-Fun-Controlnet-Union - https://huggingface.co/Kijai/MiniMax-H3-experimental The Qwen3-VL-32B nvfp4awq quant is converted from: https://huggingface.co/cybermotaz/Qwen3-VL-32B-Instruct-NVFP4 This nvfp4 text encoder does not require Blackwell GPU to use. For diffusion models prefer int8convrot if you are able to use pytorch with cu130. fp8scaled should only be used if you cannot use int8convrot. Place the files in the following folders: To use the embeddings in a text prompt, invoke them in the CLIPTextEncode node…
A converted copy of the light video VAE decoder of LynnReal-Omni (LynnReal-AI), made for the MiniMax H3 WebUI on Google Colab. It decodes ordinary MiniMax H3 latents with 26 transformer blocks instead of 36. Unofficial — not affiliated with, endorsed by, or supported by MiniMax or LynnReal-AI. The encoder, latent space and temporal structure are those of the official MiniMax H3 VAE; only the decoder differs. Take the encoder (and the audio VAE) from the official repository. diffusers — load the official AutoencoderKLMiniMaxH3, keep the first 26 entries of vae.decoder.transformerblocks, then load this file: TensorRT — compile the ONNX file with a fixed input profile 1×24×7×16×16 (FP16), then…
An MXFP8 (OCP microscaling FP8) weight-quantized build of MiniMaxAI/MiniMax-H3 — the omni-modal (video + native stereo audio) generative DiT — produced with both diffusion transformers and the Qwen3-VL text encoder are quantized to MXFP8. (The companion repo MiniMax-H3-mxfp8-native quantizes only the two DiTs and keeps the text encoder in BF16 — see "Relationship to the other variants".) The large AdaLN branch of each DiT is deliberately kept in BF16. - Each DiT ~62 GB (bf16) → ~43.65 GB (≈0.71×, −29 %); AdaLN (≈13 B params/DiT) stays BF16. - Text encoder (Qwen3-VL, 64 decoder layers) ~63 GB (bf16) → ~34 GB (≈0.54×, −46 %): the model.languagemodel.layers linears are MXFP8 (448 × F8E4M3 +…
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.