Open-Weight Models

Model · Text generation

semsimula-ladder-owt-d384-l2-attention-potential

Dimitar Gueorguiev

One arm of the SemSimula mechanism ladder — OpenWebText, d=384, L=2: five models that differ by exactly one mechanism each, so that every gap in the ordering prices one component by construction rather than by attribution. Italic entries are pre-registered predictions, recorded before the run. The ordering is itself the prediction; any inversion is a result. Every arm above sees exactly 532,480,000 training tokens — 32,500 steps × 32 × 512 = 532,480,000 tokens — on the same corpus, tokenizer and validation batches. No arm is trained longer than another, so a gap in the table is a gap between mechanisms and never a difference in budget. The models below integrate a damped mechanical system…

Model · Text generation

semsimula-ladder-owt-d384-l2-gpt2-matched

Dimitar Gueorguiev

One arm of the SemSimula mechanism ladder — OpenWebText, d=384, L=2: five models that differ by exactly one mechanism each, so that every gap in the ordering prices one component by construction rather than by attribution. Italic entries are pre-registered predictions, recorded before the run. The ordering is itself the prediction; any inversion is a result. Every arm above sees exactly 532,480,000 training tokens — 32,500 steps × 32 × 512 = 532,480,000 tokens — on the same corpus, tokenizer and validation batches. No arm is trained longer than another, so a gap in the table is a gap between mechanisms and never a difference in budget. The models below integrate a damped mechanical system…

Model · Text generation

semsimula-ladder-owt-d384-l2-none

Dimitar Gueorguiev

One arm of the SemSimula mechanism ladder — OpenWebText, d=384, L=2: five models that differ by exactly one mechanism each, so that every gap in the ordering prices one component by construction rather than by attribution. Italic entries are pre-registered predictions, recorded before the run. The ordering is itself the prediction; any inversion is a result. Every arm above sees exactly 532,480,000 training tokens — 32,500 steps × 32 × 512 = 532,480,000 tokens — on the same corpus, tokenizer and validation batches. No arm is trained longer than another, so a gap in the table is a gap between mechanisms and never a difference in budget. The models below integrate a damped mechanical system…

Model · Text generation

semsimula-ladder-owt-d384-l2-none-norc

Dimitar Gueorguiev

One arm of the SemSimula mechanism ladder — OpenWebText, d=384, L=2: five models that differ by exactly one mechanism each, so that every gap in the ordering prices one component by construction rather than by attribution. Italic entries are pre-registered predictions, recorded before the run. The ordering is itself the prediction; any inversion is a result. Every arm above sees exactly 532,480,000 training tokens — 32,500 steps × 32 × 512 = 532,480,000 tokens — on the same corpus, tokenizer and validation batches. No arm is trained longer than another, so a gap in the table is a gap between mechanisms and never a difference in budget. The models below integrate a damped mechanical system…

Fine-tune of convaiinnovations/laya-multilingual (Apache-2.0) for prompt-injection and jailbreak detection. The base model is mmBERT, vocab 256k. Code: 3p3r/sentinel-laya. The English checkpoint is a different model: 3p3r/sentinel-laya. This revision starts from the inject-label checkpoint and adds a round on 3p3r/short-role-attacks, a synthetic English and German set of short role swaps and short orders, each with a benign twin. The five evaluation sets below were not part of training. Noul temperature is 2.2534, kept from the first revision. Threshold is 0.5. Positive class: jailbreak or prompt injection. Metric: Binary F1. Same prompts as the English card, one RTX 3090, batch size 64.…

Model · Tabular regression

sentinel-mumbai-hybrid-pinn-keras

Abdullah Ashraf

Sentinel-Mumbai Hybrid PINN is an attention-augmented Physics-Informed Neural Network (PINN) developed in Keras 3 and optimized for high-throughput GPU inference (Tesla T4 mixed-precision float16). The network incorporates a 24-hour temporal sequence lookback and enforces hydraulic mass conservation via a partial differential equation (PDE) continuity constraint during gradient descent: $$\text{PDE Residual} = \frac{\partial \hat{y}}{\partial t} - (\text{Rainfall} - \text{Infiltration}) = 0$$ Where $\text{Infiltration} \approx 0.0$ on impervious urban concrete surfaces (landcover class 80). Released under the MIT License.

Model · Tabular regression

sentinel-mumbai-pinn-v1

Abdullah Ashraf

Sentinel-Mumbai PINN v1 is a high-capacity Physics-Informed Neural Network (PINN) engineered for hyper-local urban flood depth forecasting and storm surge dynamics across Mumbai / Salsette Island. Trained across multi-decadal telemetry (rainfall, cumulative precipitation, tidal surge oscillations, elevation, urban impermeability, and drainage networks), the model embeds physical boundary laws directly into its optimization objective, preventing unphysical mass disappearance during coastal hydraulic lock events. The model optimizes a composite objective balancing empirical data fidelity against hydraulic conservation of mass: $$\mathcal{L}{\text{total}} = \alpha \cdot…

Model · Tabular regression

sentinel-v7-deep-flood-lstm

Abdullah Ashraf

Sentinel-V7 Deep Hydrological Bi-LSTM is a specialized recurrent neural sequence model engineered for real-time flood risk scoring and inundated area forecasting. It maps multi-sensor remote sensing inputs (Copernicus Sentinel-1 SAR radar backscatter, digital elevation models, river network proximity, ECMWF GEOGloWS river discharge, and precipitation metrics) to a continuous flood vulnerability metric. The model is packaged in Keras 3 format along with its fitted Scikit-Learn MinMaxScaler (scaler.joblib), ensuring seamless deployment on edge systems and cloud microservices. Released under the MIT License.

Experiment recovery package. Neutral track names preserve independent training arms; the restore manifest maps them back to exact local identities. Complete states contain adapter, optimizer, reference and random-generator state. The separate reference initializer is inference-only and has no optimizer state. The context archive contains data, code, episode history and recovery instructions. The data/ directory indexes the exact SFT/RL datasets and their original schemas; environment/ documents the pinned environment, model and rebuild commands. Credentials, base-model cache and the Python environment are excluded. Original project contributions, including adapters, experiment code and…

Selected adapter, optimizer, reference, RNG and cursor states with source/data receipts. Restore only from a verified pinned commit and check every SHA256. Base weights and private credentials are supplied separately. Project-authored backup artifacts use Apache License 2.0; upstream artifacts retain their applicable licenses. https://www.apache.org/licenses/LICENSE-2.0

Model

seti-tools

Charles Norton

Technosignature searches on public astronomical archives: the code, the papers, and a database of what every run searched and found. Python 3.11+ with requirements.txt (torch for the GPU paths only). Paths come from SETIDATA, SETICACHE, SETIRESULTS and SETICOVERAGE, or a config file outside the repository (keys in localconfig.example.json). python -I selftest.py checks the package, python -I searches/.py lists a family's approaches, python -m setitools.coverage card lists every run that covered a target, and table regenerates RESULTS.md. The code (.py and.jl files) is under the MIT License; everything else, the papers, documentation and coverage store included, is under CC BY 4.0. Both are…

Model

sft-5g-rca-qwen3-1.7b

Mark Anthony

This model is a fine-tuned version of Qwen/Qwen3-1.7B. It has been trained using TRL. This model was trained with SFT.

Multi-turn chess SFT checkpoints trained after the YaRN positional-encoding fix in sfttrainer.py (2026-09). Before the fix, the blocksize / ropescaling change was applied after frompretrained(), so runs with a 3072-token context trained with plain RoPE extrapolation while their saved config claimed YaRN. These runs load the config first and build the model from it, so training and inference both use YaRN (factor 3, 1024 → 3072). Thanks to Jens Tuyls for spotting the bug. Folders mirror the local save tree. Each model lives at /final/, where modelid = C{compute}{size}alpha{α}beta{β}. Alongside final/ are combinedeval.json (when evaluated) and configsnapshot/ (training config). Optimizer…

Model · Text generation

sft_gene_pathway_v9_9_100k_a256bs32_9b_celltext

WangLab UofT

Qwen3.5-9B fine-tuned to reason mechanistically about genetic (CRISPR) perturbations and predict how a queried pathway responds — upregulated, downregulated, or unchanged — in a given cell. This is the cell-text arm of BioReasonCell: the v9.9 SFT recipe, with one change. The model gets no STATE cell embedding. It reads the cell as text instead, a Cell Context: block in the prompt, so it is a plain text model and every input it needs can be built from a new screen's h5ad. The block follows the queried pathway, and the prompt's opening reads "Given a cell's context, a genetic perturbation, …": Top Expressed Genes are the 50 most abundant genes of the line's non-targeting control cells…

Model

shelling

Commander Turtle

Just an archive of forks / snapshot references. All files independently trained rely on overhead initial licensing. /repo.tar -> Git head snapshot kept up-to-date for the AGPL project Odyssues Worthwhile lookup: vllm-windows wheels. Try another if one doesn't work. WSL isn't so bad either Just edit serve command as needed: Non-negotiable moe-backend and flashinfer sampler env var - discovered from (this model - link to highlight) - where it's shown the env vars needed for compressed-tensors on consumer blackwell (tl;dr sm120 = 5090... sm121 = RTX6000+, datacenter blackwell) Second model - not as scientific - Some abliterated-s had been made and worth researching if needed (bigger…

This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.20.0

Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019). - PEFT 0.20.0