Fresh joint-context ARC model: all demonstrations and the query share attention. Not an upstream checkpoint or paper reproduction. Check metrics.jsonl for completed optimizer updates; compilation is not training. Checkpoints are saved on the remote Trainium host and uploaded here, not downloaded to the workstation. Dataset policy and sources are in manifest.json. Spatial ConvGLU: False. Dataset policy: all-public-kaggle-only. With all-public training, local metrics are training diagnostics, not a benchmark. Snapshots are 300-DPI PNGs plus vector SVGs. Epochs: 100; sampling: shuffled without replacement. Epoch checkpoints preserve the permutation and cursor.
Independent publisher
Ritwika Kancharla
ritwika96
Deep RL, Transformers
Models in Library2
Datasets in Library2
Models on Hugging Face20
Followers2
Compare the published evidence
This profile connects the 4 model and dataset records shown below to their upstream cards and sources. The library counts describe SAVRN's coverage, rather than this publisher's entire output.
Published tasks: Text generation.
Reported licenses: apache-2.0, mit. Open a record to review its specific license and conditions.
Compare model specifications, source dates, and hardware estimates on the individual records. Publisher identity verification is separate from a model's quality, deployment readiness, or suitability for your work.
Models
Fresh joint-context ARC model: all demonstrations and the query share attention. Not an upstream checkpoint or paper reproduction. Check metrics.jsonl for completed optimizer updates; compilation is not training. Checkpoints are saved on the remote Trainium host and uploaded here, not downloaded to the workstation. Dataset policy and sources are in manifest.json. Spatial ConvGLU: True. Dataset policy: all-public-kaggle-only. With all-public training, local metrics are training diagnostics, not a benchmark. Snapshots are 300-DPI PNGs plus vector SVGs.
Datasets
Public metrics for the fresh Qwen3.5-9B rank-128 direct-grid SFT experiment. snapshot.json contains loss curves, epoch/step progress, generated exact-grid evaluations and a trace index. traces/ contains saved exposed reasoning, final predictions and public task grids for the fixed evaluation sample. The publisher selects explicit fields and excludes credentials, service checkpoint addresses and private process configuration. Data syncs approximately every 60 seconds while the local publisher runs. publishedat indicates freshness; a saved snapshot does not prove the trainer is still running. The training probe uses 20 tasks from the training set. The held-out probe uses 20 tasks reserved for…
1,915,399 published training episodes from 239,520 fresh base episodes. each expanded into up to eight distinct rotations/reflections. These counts describe the files in this build, not the earlier 24-row format preview. (the gallery currently displays the earlier corpus). Select a task family in the viewer's subset menu. Each family contains its own combined episode records; the selected family's row count is not the total corpus count above. Family configurations limit conversion size because Hugging Face may truncate non-Parquet previews beyond 5 GB per configuration. They do not split programs from grid targets. Each row displays a full-resolution, lossless PNG, readable Python program…