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 2026-10-21 · How the library is built
12,010 models, sorted by most downloaded.
Continued pretrain of jjjlimaus/chrono4-2022-dense-qa50 on jjjlimaus/chrono-2022-ql-mix-863-375. - Public and gated (manual review) - fp8=false (fp32 masters + bf16 autocast) - maxseqlen / packto = 512, batch=22/GPU, lr=5e-5 AdamW(fused)
Forget-2023 retain-heavy pass of jjjlimaus/chrono4-2022-ql-mix-863-375 on jjjlimaus/chrono-2023-forget-retainheavy. - loss = retainCE − 0.25 · forgetCE - lr=2e-5, 8×B22 pack=512, fp32 masters + bf16 autocast, no FP8 - save every 100, planned 800 steps (retain may EOF earlier) - Public and gated (manual review)
2023-cutoff ChronoLLM (SN38) checkpoint trained in 3 chained stages (pretrain → sft → forget) from Every commit message is - (step counter restarts at 0 in each stage), so any revision of this repo maps to exactly one stage and step. - Optimizer carry-over. AdamW (fused) moments and fp32 master weights from the end of pretraining initialise SFT, and SFT's end state initialises forget-SFT. No optimizer reset, no re-warmup between stages. - Pretrain (CPT). Plain causal LM on 2023 news/wiki documents. Equal token quota for each month of 2023 (422,916,666 tokens/month incl. a 1.5% unread buffer), random draw inside each month, globally shuffled. Training stops at 55,486 steps. - SFT. Loss on…
2023-cutoff ChronoLLM (SN38) checkpoint: one SFT stage on a shuffled 7:3 mix of quality answers (keltknight distillation + OpenAI gold x3) and 2023 extractive Q&A, from (commit pretrain-55486: anacoluthe89/chrono4-2022@fae9ed37 + 5B-token month-balanced 2023 CPT, no SFT). Every commit message is sft-, so each revision maps to exactly one training step. - Subset 1 (2,399,928,714 tokens, 28,834,968 rows), every row: 1. keltknightdistill: 20,880,051 rows, 1,681,389,810 tokens. Answers by keltknight/r13-m116@66127522f156c4375c49b28df292ca76e2ce20d0 (validator generation path: BOS + raw prompt, greedy, 100 new tokens, EOS stop) for every unique prompt of sn38-quality-gold-openai-merged…
These are the 48 original simulation checkpoints used by the paper: Chronos and the timestamp-blind diffusion-transformer (DT) baseline, three training seeds, and eight tasks. Each checkpoint is approximately 560 MB. This repository is a public release candidate while dataset publication is completed. The Conveyor/Pivoting training and evaluation campaign is finished. It contains the original selected weights, not retrained replacements. manifest.json records their exact sizes and SHA-256 hashes, fixed evaluation seeds, selected scene batches, and paper reference counts. All 48 remote checkpoint hashes have been verified against the paper manifest. The compact release model also matches the…
chsm8-sft (v2.1) further trained with DPO so that it follows a natural-language style prompt more strongly ("trade queens as early as possible", "push your pawns aggressively up the board"...). Same architecture and usage as chsm8-sft: decoder over factorized move tokens, cross-attention to a frozen google/embeddinggemma-300m encoding of the prompt (all token states as keys) in the top 6 of 12 layers. Code: github.com/nov3o/llmchess (src/dposft.py, src/evalsft.py). Both use the identical loss and hyperparameters (below); they differ in run length, batch size and LR schedule length. Which to pick. model.pt is the main model because it follows style prompts most strongly (66.5% early queen…
This repository contains the CI-Net processing, training, inference, and validation code. The directories under code follow the processing order: 1. datapreparing: read and align satellite and radar inputs. 2. labeling: create cloud labels. 3. finalpreprocess: convert the prepared fields into model input files. 4. training: train the model and run inference. 5. validation: create validation targets and calculate validation metrics. Each code directory contains Python files in src and its configuration and shell entry points in run. rawdata contains the input example included with this repository. result contains prepared data, model files, inference outputs, and validation outputs. See…
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.