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 2026-10-22 · How the library is built
12,283 models, sorted by most downloaded.
Completed one joint R2R + RxR epoch, 3,852 optimizer updates, seed429, from Qwen3.5-4B base revision851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a. 30,815 episodes; effective batch8 (4H100 x1episode xGAS2), full episodes/fullBPTT. Actual one-epoch cosine schedule,116LRwarmup steps; auxiliary ramp771updates. G1/epoch-1/ is the inference export: native Window8 text policy with canonical action-text history, without a dual lane or episodic bank. Frozen online FP32 VGGT-Omega supervises sampled visual tokens during training only. The student projector follows SpatialForcing's affineLayerNorm2560 -> Linear4096 -> GELU -> Linear2048 architecture. No feature pre-extraction or PCA is used. All retained…
Private artifacts of the SimpleMemVLN episodic-memory campaign (plans/episodicmem.md, branch c2episodic on GitHub). code/c2episodic406b08f.bundle: git bundle of branch c2episodic (git clone c2episodic406b08f.bundle -b c2episodic). Training source commit e28bd01f10612851cf9ef303fbddbd43c2cae58a. / /export-e1: epoch-1 serving export; / /checkpoint-3852: complete training state for exact same-profile epoch-2 resume; / /campaign: logs and manifests. Epoch 1 is a mid-schedule snapshot of a two-epoch (7,704-update) schedule. No navigation result is implied by any file here.
Roy C March 2026 We present a methodology for training small language models on CPU at FP32 precision that achieves capability-per-dollar efficiency orders of magnitude beyond GPU-based training. Across 15 models spanning four novel architecture families — Mixture of Attentions (MoA), cross-architecture fusion (Qemma), swarm intelligence (SAGI), and metric-space causal language models (DiscoverLM) — total compute cost was $24 on a single AMD EPYC 9454P processor. We introduce seven methodological pillars: (1) FP32 precision preservation, with experiments demonstrating 5,810x single-operation error and 23,225x compounding error ratio for FP16 at network depth; (2) sparse cognitive…
A lightweight neural network that upscales 128×128 RGB images to 256×256. No deep learning frameworks or image libraries required. Python 3.10+. Both comparisons use the best weights from 100 training epochs. Originals are displayed at 2× size. Held-out validation PSNR: 26.75 dB, compared with 25.05 dB for nearest-neighbor upscaling. Install and upscale using the included weights: Saves the 256×256 output and a labeled outputs/upscaled-comparison.png. Inputs must be non-interlaced, 8-bit PNGs; convert WebP images to PNG first. The included superressamples/ dataset contains 199 image pairs, with matching filenames in 128x128/ and 256x256/. Train a new model: Training uses random aligned…
Superfast Video is a proprietary, closed-weight text-to-video model developed by LINEARGAME PTE. LTD. and made available under the YorollAI brand. It generates 1344×768, 24 fps videos with audio through a hosted API. downloadable checkpoints are not distributed through this repository. The Hugging Face Space is a hosted evaluation interface. It sends requests to the official Superfast API; inference does not run from model files stored in this repository. Superfast Video is intended for creative video generation, concept visualization, prototyping, marketing creatives, social video, entertainment, and other lawful uses permitted by the applicable service plan and the Yoroll Terms of Use.…
next Supernova NepaliFast V5 is a custom tokenizer architecture designed primarily for Nepali, English, Devanagari, and multilingual Unicode text. V5 is not simply a new vocabulary placed on top of an existing BPE, Unigram, or WordPiece engine. The core V5 system includes a custom main tokenizer, custom trie-based matching, a dedicated Unicode fallback expert, grapheme-aware routing, exact reconstruction validation, a global token-ID namespace, and deterministic decoding. Supernova V5 uses a tokenizer-level Mixture-of-Experts (MoE) architecture. This document provides an authoritative security audit and bytecode forensic verification for all compiled pickle (.pkl) serialized files in the…
This directory contains the completed, optimized Rust Native Tokenizer Engine (supernovarust) for Supernova NepaliFast V5. It is built using PyO3 and Maturin, delivering 15M+ characters per second (running 3.86x faster than OpenAI's highly optimized tiktoken engine). OpenAI Tiktoken (cl100kbase): ~3.9M chars/sec (~3.96 ms average on sample document) Follow these steps to compile and install the high-performance native module directly from this source code. Run this inside the supernovarust directory containing Cargo.toml: After building, install the newly created.whl package using pip: Once compiled and installed, you can use the Rust Engine in your Python workspace as follows
TopP: 0.8 ~ 0.95 RepetitionPenalty: 1 ~ 1.1 Space Time 2. 邮箱: [email protected]
static quants of https://huggingface.co/SpaceTimee/Suri-Qwen-3.8-27B-Neuro For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion. 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…
This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored 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).
This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored 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).
This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored 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).
This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored 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).
This model was converted to GGUF format from SpaceTimee/Suri-Qwen-3.8-27B-Uncensored 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).
ONNX export of SurjoLabs/Surjo-Image-Preview for the browser (onnxruntime-web, WebGPU), together with its text encoder LiquidAI/LFM2.5-Encoder-230M and VAE decoder fp16w means weights are stored in fp16 and computed in fp32 (image PSNR about 60 dB against PyTorch fp32). An int4 text encoder was tried and dropped: it changed the pictures too much. Sampling: start from standard normal noise, t = i/steps, dt = 1/steps, 50 steps, cfg 3.0. onnx/parity.json has the image PSNR against the PyTorch fp32 pipeline.
basemodel: - ukisai/Swift-1.5-Qwen3.8-27b
quantised to OCP MXFP4 (fp4e2m1 weights, e8m0 scales, group size 32) with GPTQ, in the compressed-tensors mxfp4-pack-quantized format. Weight-only: activations stay in 16-bit. This is the source checkpoint of the Radiance container and the checkpoint the vllm-sly-radiance image is benchmarked with (together with the DFlash2 drafter syvai/Qwen3.8-27B-DFlash2-W4A16). 1. Qwen/Qwen3.8-27B (Apache-2.0) 2. ukisai/Swift-1.5-Qwen3.8-27b (Swift Open License v1.0) 3. akumaburn/Swift-1.5-Qwen3.8-27b-heretic (abliterated, BF16) 4. This checkpoint: llm-compressor GPTQModifier, scheme MXFP4A16, static act-order, sequential pipeline, 89 calibration samples, max sequence length 2048 (recipe.yaml) The…
A single Radiance container (.rad, 16.79 GiB) with the target model only, quantised for AMD RDNA4 (Radeon AI PRO R9700, gfx1201, 32 GB). It is not a Transformers/vLLM checkpoint. It is the drafter-free sibling of which carries the DFlash2 drafter in the same file; use this one with the model's own MTP block, with a different drafter, or without speculative decoding. The GPTQ source checkpoint is 1. Qwen/Qwen3.8-27B (Apache-2.0) 2. ukisai/Swift-1.5-Qwen3.8-27b (Swift Open License v1.0) 3. akumaburn/Swift-1.5-Qwen3.8-27b-heretic (abliterated, BF16) 4. GPTQ to MXFP4 (llm-compressor, MXFP4A16, static act-order, 89 calibration samples, lmhead/vision/MTP excluded — gptq-recipe.yaml) 5.…
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.