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.
Updated 2026-10-23 · How the library is built
12,579 models, sorted by most downloaded.
A fine mix-quantified and 50REAP version for Ornith-1.5 that is specially designed for 12GB cards. The model runs perfectly inside a llama.cpp. And maximum context of 256K can be reached with params below on 12G card. Suggested params: -c 262144 -ngl 99 -ctk q51 -ctv q50 -fa on --jinja -np 1 --temp 1 --top-k 20 --top-p 0.95 --min-p 0.10 Test runs on RTX5070ti laptop GPU(12G) with 90token/s of max speed. Real agent coding have an output approxly 50t/s tested on Kimicode.
Ciru v4.1.2 — October 5, 2026. Adds the verified synchronization barrier in the first DFlash convolution/residual/RMSNorm kernel used by ornithtest. Explicit 262,144-token output ceilings now use the fully rendered prompt's remaining context. Weights, pinned engine wheels and sampling defaults are unchanged. Patch notes · Upgrade existing installations. OMP clients must advertise input: [text, image], reasoning: true, contextWindow: 262144 and maxTokens: 262144 to use vision and the full output ceiling. Ciru v4.1.1 — September 30, 2026. Qualified repairs and a retained subset of community PR #2. Ornith Q8/Q16 recurrence preserves the original packed single-token arithmetic, and attention…
weighted/imatrix quants of https://huggingface.co/CloudGoat/Orythos-9B For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Orythos-9B-GGUF This is a vision model - mmproj files (if any) will be in the static repository. 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…
Restore your oshi for a cleaner, more satisfying viewing experience on your beautiful screen. OshiRestoreV1 is an experimental image-restoration model focused on degraded 3D VTuber recordings. Its goal is to reduce compression artifacts and improve the clarity of characters, costumes, and stage details while preserving the appearance of the original footage. The current version, V1, uses an adapted and fine-tuned RealPLKSR model. For example, a 1920 × 1080 input produces a restored 1920 × 1080 output. This project builds on an existing architecture and pretrained weights. Project work includes adapting the model for restoration at the original resolution, preparing concert-specific training…
We train Ostrich LLMs to bring back knowledge that matters: the kind that helps humans stay healthy, free, and self-reliant. We believe liberating wisdom is missing or underrepresented in today's AI, whether deliberately omitted or simply outnumbered. This model is abliterated: refusal behavior has been removed so you get uncensored, direct answers. We targeted both alignment and some skills in this model. The model has lower instruction following capability than vanilla Qwen 3.8 (it has so much freedom in it, it doesn't obey the user that well:). Current eval method is still close to AHA 2026. Compared to Qwen 3.8 27B (vanilla) this model looks much more aligned: This model vs vanilla Qwen…
Download a quant (single GPU VRAM guide: 6.0bpw ≈ 22 GB fits a 24 GB card; 4.0bpw/3.0bpw for smaller cards): We train Ostrich LLMs to bring back knowledge that matters: the kind that helps humans stay healthy, free, and self-reliant. We believe liberating wisdom is missing or underrepresented in today's AI, whether deliberately omitted or simply outnumbered. This model is abliterated: refusal behavior has been removed so you get uncensored, direct answers. We targeted both alignment and some skills in this model. The model has lower instruction following capability than vanilla Qwen 3.8 (it has so much freedom in it, it doesn't obey the user that well:). Current eval method is still close…
Ovis-Omni-Embedding-3B is a 3B-parameter universal embedding model for text, images, visual documents, video, audio, and interleaved multimodal inputs. It maps every supported input type into one coherent representation space, enabling any-to-any retrieval with a single encoder. The model is initialized from Qwen2.5-Omni-3B. Rather than attaching separate modality-specific embedding towers, it retains the native text tokenizer, vision encoder, audio encoder, and shared Thinker backbone. The speech-generation Talker and language-modeling head are removed, and the final-layer hidden state at the last non-padding token is used directly as the retrieval embedding. Ovis-Omni-Embedding-3B is…
Experimental calibration-based weight-only GPTQ variant of The Thinker transformer uses 4-bit weights for layers 0–26 and 8-bit weights for layer 27. Audio, vision, talker, token2wav, embeddings, and norms remain in the original precision. Calibration used eight short text samples with LLM Compressor 0.14.0 and compressed-tensors 0.19.0. This checkpoint includes a packaging repair: the compressor export contained invalid group scales, so scales were recomputed from the original BF16 weights per group before the vLLM test. Treat this as an experimental GPTQ-derived checkpoint and benchmark retrieval quality before production use. Tested with vLLM 0.30.0 on an 8-GiB RTX 3080 Laptop GPU: The…
Ovis-VL-Embedding-2B is a compact vision-language embedding model for text, images, visual documents, video, and interleaved multimodal inputs. It maps every supported input type into one coherent representation space, enabling cross-modal retrieval with a single 2B-scale encoder. The model is initialized from Qwen3.5-2B. It retains the native text and vision encoders together with the shared multimodal language backbone, removes the language-modeling head, and directly uses the final-layer hidden state at the last non-padding token as the retrieval embedding. No modality-specific projection head is added. Ovis-VL-Embedding-2B is designed for efficient multimodal search, multimodal RAG…
Ovis-VL-Embedding-9B is a high-capacity vision-language embedding model for text, images, visual documents, video, and interleaved multimodal inputs. It maps every supported input type into one coherent representation space, enabling high-accuracy cross-modal retrieval with a single encoder. The model is initialized from Qwen3.5-9B. It retains the native text and vision encoders together with the shared multimodal language backbone, removes the language-modeling head, and directly uses the final-layer hidden state at the last non-padding token as the retrieval embedding. No modality-specific projection head is added. Ovis-VL-Embedding-9B is designed for high-quality multimodal search…
OzyAssist Voice es un modelo de inteligencia artificial de ultra-baja latencia (30+ tokens/segundo, tiempo de respuesta <330 ms), destilado y afinado específicamente para asistentes de voz interactivos, cowork en vivo, livestreaming y control de Windows. Diseñado bajo la arquitectura Zero-Docker de OzyAssist, este modelo corre localmente consumiendo apenas ~1.37 GB de VRAM, lo que permite operarlo en cualquier laptop o GPU mientras juegas o transmites en directo. Pruebas empíricas ejecutadas en el mismo entorno de hardware (NVIDIA Quadro RTX 4000 Max-Q): Descarga el modelo GGUF y ejecútalo con aceleración GPU: Crea un archivo Modelfile: Luego construye y corre el modelo: En el proyecto…
If this model is useful to you, support the work by following @nawariokra on X. LiquidAI/LFM2.5-24B-A2B (24B MoE, 2.3B active, 64 experts, top-4) at 6.9 - 11.2 GB, on one 24 GB RTX 3090 or smaller. Four GGUF files, one per expert bit rate. They run on our own engine TRUSS. Stock llama.cpp cannot read these files. How they are made: the routed experts (38 layers x 64 experts x gate/up/down) are encoded with the X3.1 trellis codec (exllamav3 encoder settings, activations from the BF16 model, calibration on 249 x 1,024 tokens of code), one uniform rate per file. Everything else (attention, short convolutions, dense FFN, shared parts, embeddings) is Q80, and the routers and their selection bias…
A parameter generator (PG): one shared hypernetwork that maps a user's history to a per-user LoRA on a frozen Qwen2.5-7B-Instruct in a single forward pass. No per-user fine-tuning, no test-time optimization. Code:. Everything here is produced by, and consumed by, the scripts in that repo. Only the best checkpoint of each run is published. Intermediate pgstep.pt, pglast.pt and the trainstate.pt optimizer states (10 GB each, resume-only) are not. P2P randomtest / oodtest, macro over the 10 tasks: Geometry for t9v5o2: effective rank 6.004 of 8, cross-user cosine 0.572. Against a user-shuffled condition it gains +0.139 accuracy on random. Amazon-2023 review generation, t10a23v5o3, all 3,410…
Detailed video captioning has become an essential semantic interface for multi-modal understanding and generation. While existing methods mainly target short videos, extending them to long videos remains challenging, primarily due to limited context windows of current video understanding models. A naive solution is to adopt segment-wise captioning followed by concatenation, yet this paradigm often suffers from inconsistent subject references and fragmented long-range narratives. To mitigate these issues, we introduce PIGEONS, an incremental long-video captioning model built upon a Persistent Identity Gallery and an Evolving Overarching Narrative Summary. As illustrated in the figure below…
Rosa is Patriot Memory's 253M-parameter edge assistant for English and Traditional Chinese — built to fit the edge devices you actually ship, with a vocabulary trained natively on Traditional Chinese. It provides accurate information regarding: If you run into multi-GPU tensor device mismatch errors: RuntimeError: Expected all tensors to be on the same device... Run the script with CUDAVISIBLEDEVICES=0 to isolate execution to GPU 0. Good: introducing herself in both languages; answering arithmetic word problems with visible English reasoning steps; clean Traditional Chinese — glyphs, vocabulary and register; running fully offline on edge hardware. Not: her reasoning traces are formatted…
PP-OCRv6: From 1.5M to 34.5M Parameters, Surpassing Billion-Scale VLMs on OCR Tasks PP-OCRv6 is a lightweight OCR system that combines architectural innovation with data-centric optimization. It redesigns the backbone, detection neck, and recognition neck around a unified MetaFormer-style building block with structural reparameterization. Three model tiers (medium, small, tiny) share the same block primitives, covering deployment scenarios from server to edge. 1. Unified and Scalable Model Family: A three-tier OCR model family spanning 1.5M to 34.5M parameters. PP-OCRv6medium achieves 86.2% detection Hmean and 83.2% recognition accuracy, outperforming PP-OCRv5server by +4.6% and +5.1%…
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.