This gemma4 model was trained 2x faster with Unsloth and Huggingface's TRL library.
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.
LoRA sobre stabilityai/stable-diffusion-xl-base-1.0, evolução incremental da PS-IMAGE-1.4: inicializada com os pesos 1.4 e continuada por +700 steps focados em texto. letras+números, captions em 5 línguas (80 EN / 80 PT / 80 ES / 80 FR / 80 DE) - 700 steps, batch efetivo 4, lr 3e-5 cosine, AdamW, min-SNR γ=5, T4, loss final 0.0057 - Checkpoints a cada 100 steps em checkpoints/step (dá p/ retomar o treino) O SDXL rende melhor em inglês. Jeito mais fiel — tradutor leve + 1.5: A 1.5 também treinou com captions PT/ES/FR/DE, então prompting direto nessas línguas funciona melhor que na 1.4 (ver textpt e textes abaixo) — mas traduzir p/ inglês continua sendo o mais fiel. Prompt (seed 24000): photo…
Evolução do PS-IMAGE-1.5 — mesmo formato de LoRA (UNet do SDXL, r16/α32, toq/k/v/out.0, ff.net.0.proj/ff.net.2), treinado por continuação (drop-in: o código do 1.5 funciona trocando o nome do repo). - Comparação A/B: exemplos15/ (1.5) vs exemplos16/ (1.6), mesmos prompts e seeds SDXL não tem entrada de imagem — "visão" não se treina como LoRA. Para editar fotos mantendo consistência use o notebook PS Image 1.6 Edicao: img2img, inpainting por máscara, ControlNet (canny/depth/pose), IP-Adapter (imagem de referência) e refino 768→1152, tudo usando este LoRA.
Evolução incremental do PS-design-2.5 — fine-tune contínuo do mesmo LoRA (r16 / alpha32, base Qwen/Qwen2.5-0.5B-Instruct), treinado em 2× Tesla T4 em 102 min. reais (SaaS, agência, saúde digital, coaching, pilates, e-commerce, portfólio), convertidos em arquivos HTML únicos e autônomos, com o CSS da própria página embutido em. Cada exemplo real foi repetido ~25× para pesar de verdade ao lado da destilação. - Auto-destilação: 55 páginas geradas pelo próprio 2.5 (30 nichos × 6 design systems), mantendo o estilo do modelo e evitando esquecimento catastrófico. Checkpoints intermediários: checkpoints/checkpoint-. Exemplos gerados: exemplos/.
PULSE is an asymmetric, variable-rate neural image codec for practical compression on resource-constrained hardware. It combines ultra-low-complexity decoding, bit-exact entropy coding, and MSE or perceptual optimization. provide deterministic entropy transport, with a content-adaptive Meta Prior. levels each, including face/text-aware perceptual fine-tuning. decompression, and optimized CPU, H100 GPU, and mobile NPU deployment tools. Each bundle contains config.json, model.pt and entropycontrolint.pt. Eight QPs (0–7) are supported. Keep the matching files together. manifest.json and SHA256SUMS record all inference-file checksums. Install the PULSE code following its installation guide…
static quants of https://huggingface.co/Indexnusrefather/Pallete-2.6B-RP-Reasoning-2609-v1b For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/Pallete-2.6B-RP-Reasoning-2609-v1b-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…
weighted/imatrix quants of https://huggingface.co/Indexnusrefather/Pallete-2.6B-RP-Reasoning-2609-v1b For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/Pallete-2.6B-RP-Reasoning-2609-v1b-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…
Peppercorn is a compact vision-language model for Vietnamese and English document OCR, with 328M parameters. It takes a document page as input and produces structured Markdown, including HTML tables and LaTeX mathematical expressions. Its name comes from the Vietnamese saying nhỏ nhưng có võ — small, but mighty: compact in size, yet capable. Compared with original design, Peppercorn uses approximately 4× fewer image tokens per page (~900 vs. ~3,600) and achieves higher throughput under our tested serving configuration. It performs particularly well on Vietnamese administrative documents and synthetic tables, while English OCR, mathematical expressions, and some real-world financial tables…
PhaGen-0404-step12800 is the frozen paper model: 152,421,601 parameters, three hierarchical stages with dimensions 512/256/196, maximum token canvas 131,072, and a nine-token character-level DNA tokenizer. This repository contains the standalone inference weights, tokenizer, model configuration and custom code. Weights are the selected April 2026 run's global step 12,800, not the final saved checkpoint or a later training implementation. Use PyTorch 2.5, Transformers 4.54.1, einops 0.8, beartype 0.22, accelerate and safetensors. The PhaGen source repository also provides an installable phagen CLI and an agent skill. Pin a Hub commit hash with revision for repeatable loading. Custom model…
Tiny model. Big decisions. — 小海豹，大决断。· 一斑见全豹，一点定全局。 斑海豹 Largha, the spotted seal: a 144.3M bilingual decision model (150M-class) for structured decisions — one forward pass per decision, on your own hardware. Not a chat model: it takes a state plus a list of typed questions (noul yes/no · choice pick-one · score 2–10) and returns calibrated answers with confidence, robust to option reordering. The official runtime is phocinae-server: a local FastAPI service (127.0.0.1 only) that loads these weights with a pure PyTorch forward (no extra runtime needed). Full spec: docs/protocol.md. answers values: noul = bool · choice = 0-based option index · score = integer 2–10. Errors: 422 (unknown…
The Photo Anime model is based on Neural Network Lora Fusion (Layered Lora Fusion). Fusion Method: This process involves a layered merging of various Lora modules, focusing on highly minute and precise stratification of different Lora characteristics. Refinement: After the layered fusion, a "surgical adjustment" is performed using the underlying neural network. Result: This technique achieves unprecedented levels of imagination, artistic coherence, practicality, speed, and ease of use. User Experience: Its efficacy is demonstrated by the model's ability to imbue the generated image with "soul" using only the simplest prompt: 1girl. Cultural commentators and artists find it difficult to…
One forward pass. One calibrated decision. PhysicalEye-Decide-35B answers a question about an image by choosing among the options you give it, and returns a probability for every option. It does not write text. It reads the image, the question and the options once, and outputs a decision. Use it for visual inspection, yes/no checks, option selection and scoring on production lines, in apps and in robots. - A small decision readout sits on top of the backbone and scores each option directly. One request carries up to 4 images, an optional context (state) and one or more questions. Three question types are supported: choice (pick one option), noul (yes or no) and score (a value on a scale).…
PicoPolite240M is a small-scale language model created using extremely limited computational resources (a single GPU with 32 GB memory) to research training methods for LLMs and SLMs. We will publish a book detailing the entire learning methodology. Furthermore, we will release the complete synthetic dataset created for this purpose. Using the data below, I trained the model for one epoch with a token window of 2000 and fp32 precision; the RoPE θ was set to 3200. Subsequently, using the data below, I trained it for one epoch with a token window of 6400 and bf16 precision; the RoPE θ was set to 10000. Afterward, I applied LongRoPE to extrapolate the token window. This model utilizes google…
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.