Downloads · 30 days
0
itigges22/atlas-lens-gemma4-12b
atlas-lens-gemma4-12b is a machine learning model from itigges22. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
Cost-field C(x) (and optionally metric tensor G(x)) trained against the costfield embedding space. Loaded by the ATLAS geometric-lens service to score code candidates without execution.
Downloads · 30 days
0
Access
Public
Updated Jul 2, 2026
Repo size
94.5 MB
Likes
0
Public
Click a slice to open those files.
.pt75.3 MB · 80%
From the Hugging Face model README
Cost-field C(x) (and optionally metric tensor G(x)) trained against the cost_field embedding space. Loaded by the ATLAS geometric-lens service to score code candidates without execution.
cost_field.ptcost_field.safetensorsgx_xgboost.jsongx_weights.json# Drop these into your ATLAS checkout
mkdir -p geometric-lens/geometric_lens/models/
huggingface-cli download <this-repo> cost_field.pt \
--local-dir geometric-lens/geometric_lens/models/
# Verify ATLAS picks them up
atlas lens check
# expected: verdict: compat
| Field | Value |
|---|---|
| Base model | cost_field |
| Input embedding dim | 3840 |
| cost_field.pt SHA256 | 82b2d960b71b84a90ecc69189c67bc72a9b7cbc5c79b7b207dcaf2634b0af509 |
| cost_field.pt size | 7.76 MB |
Trained locally via atlas lens build against cost_field's
self-embeddings. Architecture: 3840 -> 512 -> 128 -> 1 (SiLU, SiLU,
Softplus). Contrastive ranking loss on labeled pass/fail code samples.
apache-2.0. The artifact derives from a cost_field forward-pass — verify the base model's license is compatible with redistribution before publishing.
To get ATLAS users this support automatically via atlas model list,
open a PR against https://github.com/itigges22/ATLAS using the body
atlas lens publish produced. PC-059 (#101) tracks the manual-review
flow; PC-060 (#102) tracks the eventual auto-merge pipeline.