Roy C · independent ML researcher & AI red-teamer Open-weight models, adversarial evaluation, and efficient training at the small-model scale. I build and ship small open-weight language models — 50+ public checkpoints, quantized for edge and picked up by third-party quantizers (mradermacher) for independent re-hosting. The work sits at the intersection of three things I care about: - Open-weight modeling at volume. Distillation pipelines over Qwen3, Gemma, and LFM2.5 bases, full GGUF quant ladders for local/edge deployment, and a house methodology — structure over scale — for pulling capability out of sub-2B models instead of buying it with parameters. - Adversarial / safety-relevant…
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.
3,247Models
1,032Datasets
256Papers
2,088Publishers
5,461Sourced relationships
Updated 2026-09-20 · How the library is built
3,247 models, sorted by most downloaded.
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).
Document: https://diffsynth-studio.readthedocs.io/zh-cn/latest/index.html Welcome to the magic world of Diffusion models! DiffSynth consists of two open-source projects: DiffSynth-Studio: Focused on aggressive technological exploration. Targeted at academia. Provides more cutting-edge technical support and novel inference capabilities. DiffSynth-Engine: Focused on stable model deployment. Geared towards industry. Offers better engineering support, higher computational performance, and more stable functionality. DiffSynth-Studio is an open-source project aimed at exploring innovations in AIGC technology. We have integrated numerous open-source Diffusion models, including FLUX and Wan, among…
A research-oriented Blip prototype targeting Retrieval. The included giant setup documents defaults and file formats without presenting unverified performance numbers. - The Python file contains the model and runnable example or training entry point. - config.json records the generated architecture settings. - trainingargs.json records the default experiment recipe. - model.safetensors is a valid initialization checkpoint for smoke tests; it is not presented as a trained benchmark checkpoint. - No benchmark score is claimed in this repository. The included configuration uses adam with a cosine schedule. These are starting values in the script, not evidence of a completed run. For a…
This repository provides rh-hf-e2e-t2-20260918104741-lora weight files. You can load them on RunningHub. Published by RunningHub on behalf of the author. Copyright remains with the author. Follow the original project or upstream license. RunningHub API: RunningHub API provides unified access to 500+ AI models, including Seedance and other state-of-the-art multimodal models, at prices as low as 30% of official rates. Built for AIGC startups, AI short-form drama studios, and other production teams looking to create at scale https://www.runninghub.ai/call-api - RunningHub：https://www.runninghub.ai - RunningHub 中国站：https://www.runninghub.cn - RunningHub API…
This repository provides rh-hf-e2e-t4-20260918105654-lora weight files. You can load them on RunningHub. Published by RunningHub on behalf of the author. Copyright remains with the author. Follow the original project or upstream license. RunningHub API: RunningHub API provides unified access to 500+ AI models, including Seedance and other state-of-the-art multimodal models, at prices as low as 30% of official rates. Built for AIGC startups, AI short-form drama studios, and other production teams looking to create at scale https://www.runninghub.ai/call-api - RunningHub：https://www.runninghub.ai - RunningHub 中国站：https://www.runninghub.cn - RunningHub API…
Bài tập môn Tối ưu hóa nâng cao, lớp Khoa học dữ liệu. Nội dung là tự cài đặt và so sánh bốn thuật toán tối ưu hóa trên bài toán hồi quy tuyến tính có hiệu chỉnh Ridge, dữ liệu Lending Club 2007-2018: gradient descent toàn phần, SGD thuần với lô một mẫu, mini-batch SGD và phương pháp Newton, mỗi thuật toán chạy với cả bước cố định lẫn backtracking line search theo điều kiện Armijo. Kế hoạch chi tiết: KEHOACHTRIENKHAI.md. Quy tắc làm việc: CLAUDE.md, kèm \min{w \in \mathbb{R}^d} \quad f(w) = \frac{1}{2n} \left\| Xw - y \right\|2^2 + \frac{\lambda}{2} \left\| w \right\|2^2 Hàm mục tiêu lồi mạnh và có nghiệm đóng, nên $f^$ tính được chính xác tới sai số máy và mọi biểu đồ hội tụ đều vẽ $f(wk)…
A(n) APPO model trained on the doomhealthgatheringsupreme environment. This model was trained using Sample-Factory 2.0: https://github.com/alex-petrenko/sample-factory. Documentation for how to use Sample-Factory can be found at https://www.samplefactory.dev/ After installing Sample-Factory, download the model with: To run the model after download, use the enjoy script corresponding to this environment: You can also upload models to the Hugging Face Hub using the same script with the --pushtohub flag. See https://www.samplefactory.dev/10-huggingface/huggingface/ for more details To continue training with this model, use the train script corresponding to this environment: Note, you may have…
This archive stores reproducible RLinf/OpenVLA-OFT LIBERO training recipes, model artifacts, checkpoints, logs, and evaluation summaries. This model archive is intentionally separate from the independent /media/david/HDD/trainingrecipe/ repository: - models/: base VLA model artifacts. - checkpoints/: distributed PPO checkpoints by training run and global step. - results/: metrics, logs, and TensorBoard outputs by training run. - runs/: raw logs and TensorBoard snapshots. - /media/david/HDD/trainingrecipe/: one self-contained recipe directory per training run, containing only YAML, source revision, hyperparameters, and README. The first archived run is the 4-GPU H20 task-3 PPO experiment…
Model · Reinforcement learning
rloo_qwen3_1.7b_polaris_hybrid8_b32_len32768_thinking_seed79
Training run started on 2026-09-18. Training is in progress; no quality evaluation is claimed. - RLOO, learning rate 1e-6, KL coefficient 0, sigmoid length reward alpha 0.1, seed 79. - Sampling temperature 1.0, top-p 1.0. No validation run. - Synchronous generation/training on the same eight H100 80GB GPUs; ZeRO-2; bfloat16. Every 20 rollout steps, a resumable DeepSpeed checkpoint (model and optimizer state) is uploaded under checkpoints/actor/globalstepN/, together with its training configuration. It is not a standalone Transformers model folder. Restore that directory and a latest file containing its tag to resume with the same OpenRLHF configuration. Local files are removed only after…
Meta SAM 3 (Segment Anything Model 3), exported to ONNX (FP16) for a C++ ONNX Runtime pipeline: a vision encoder, a text encoder and a decoder with a dynamic prompt-count batch (a single file accepts any number of text prompts at runtime, no re-export needed). The model is derived from Meta's SAM 3 and is provided under the SAM License (see LICENSE in this repository). Use, reproduction and redistribution are subject to that agreement; the license text must be kept with any redistribution.
samai-27b（Qwen/Qwen3.8-27B 手术版：PonderNet 停机位点 L35/39/43/47/51/55/59/63 + dMoE 8 专家 + MTP）经 QLoRA SFT 后，按 mixbit 静态化管线产出的 llama.cpp 混合精度权重。 等价稠密模型 ≙ Qwen35ForCausalLM（48 linearattention + 16 fullattention，hidden 5120），可直接用上游 llama.cpp 载入。 - Token embedding / output：Q6K（--token-embedding-type Q6K） OpenAI 兼容端点 /v1/chat/completions。 - 基座：tchbcb/samai-27b（含 SFT adapter / 训练日志 / 数据） - 配方参考：tchbcb/samai-4b-mixbit
These are the checkpoints needed to continue training SAWT V4. If you only want to restore audio, use Quran-Lab/sawt-v4 instead: it carries the trimmed inference weights and the client code. The released generator.pt, anchor.pt and vae48dacvaed2.pt in the model repository are derived from these by stripping what inference does not need.
Two Qwen3-4B full-finetunes that judge one proposed agent action, before it executes, against the observed prefix and the task policy, and emit a single verdict box. They differ only in SFT learning rate. act is CONTINUE / ASK / STOP. The binary projection is CONTINUE=SAFE, STOP=UNSAFE, and ASK=UNSAFE at risk >= 50. cite points at an earlier visible step, or NONE. fail / harm / src come from a 16 failure-mode, 11 harm-type, 10 risk-source taxonomy; NONE is an absence sentinel, not an additional class. Full SFT from Qwen/Qwen3-4B, epoch 2, thinking enabled, on 2,464 step-level records. Prompts were rendered through the evaluator's own pipeline, so the training and inference formats match.…
Deep-learning-based missing-trace reconstruction on pre-stack seismic shot gathers. Given a shot gather with masked (missing) traces, the model reconstructs the full gather. Each experiment is one architecture trained on one missing-trace scenario with one random seed. Model directories are suffixed with the dataset the model was trained on: - mobil — Mobil field dataset (pre-stack seismic shot gathers) - segc3 — SEG C3 synthetic dataset (wiki.seg.org/wiki/C3): 9 regular shots, 201 traces x 625 time samples, dt = 2 ms - Chai2020 UNet (chai2020unet) — 2D U-Net (Chai et al., 2020, IEEE TGRS, DOI 10.1109/TGRS.2019.2961015): 50 layers = 19 convolutions (5x5, same padding) + 18 ReLU + 4 max-pool…
Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.
Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.
Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.
This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.
This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.
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.