Open-weight model
fact-checking-abstention-saes
by Lucas Fraga lucasfrag/fact-checking-abstention-saes
fact-checking-abstention-saes is an open-weight model from Lucas Fraga, released under other. Its published files total 2.1 GB.
Domain TopK sparse autoencoders trained on the residual stream of instruction-tuned LLMs while they read fact-checking verdict prompts (claim + evidence + instructions), released to study how models decide to abstain ("NOT ENOUGH EVIDENCE") instead of issuing…
Model Card
Domain TopK sparse autoencoders trained on the residual stream of instruction-tuned LLMs while they read fact-checking verdict prompts (claim + evidence + instructions), released to study how models decide to abstain ("NOT ENOUGH EVIDENCE") instead of issuing a verdict. Off-the-shelf SAEs reconstruct the verdict-decision position poorly (it is out of their training distribution); these SAEs are trained on the task distribution itself, so the decision position is in-distribution. Training follows Gao et al. (2024) — see each SAE's card for the full configuration and metrics. Browse the latents in the SAE viewer: lucasfrag/fact-checking-abstention-saes-viewer (top-activating claims at the…
Excerpt from the card by Lucas Fraga, licensed other.
Identity and Version
- Repository
- lucasfrag/fact-checking-abstention-saes
- Publisher
- Lucas Fraga
- Task
- Not stated by the source
- Modality
- Other
- Library
- sae_lens
- Parameters
- Not stated by the source
- Languages
- sae-lens
- Revision
- fddc3fecd2b27c5766a7d2eacd2e6456eed6ead9
- First published
- 2026-09-28
- Last updated
- 2026-10-02
Files and Weights
12 files, 2.1 GB in total. The weights are 2 files totalling 2.1 GB in safetensors.
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| deepseek-r1-distill-llama-8b/layer-15/sae_weights.safetensors | Weights | 1.1 GB | 3aba0a152b4b |
| llama-3.1-8b-instruct/layer-15/sae_weights.safetensors | Weights | 1.1 GB | b907919ca3a9 |
| deepseek-r1-distill-llama-8b/layer-15/cfg.json | Configuration | 640 B | — |
| deepseek-r1-distill-llama-8b/layer-15/export_verification.json | Configuration | 399 B | — |
| deepseek-r1-distill-llama-8b/layer-15/fcsae_training_config.json | Configuration | 1.8 KB | — |
| llama-3.1-8b-instruct/layer-15/cfg.json | Configuration | 632 B | — |
| llama-3.1-8b-instruct/layer-15/export_verification.json | Configuration | 400 B | — |
| llama-3.1-8b-instruct/layer-15/fcsae_training_config.json | Configuration | 1.8 KB | — |
| README.md | Documentation | 3.3 KB | — |
| deepseek-r1-distill-llama-8b/layer-15/README.md | Documentation | 5.4 KB | — |
| llama-3.1-8b-instruct/layer-15/README.md | Documentation | 3.8 KB | — |
| .gitattributes | Repository | 1.5 KB | — |
License and Download
- License
- other
- Access
- Open weights, no gate
- Download size
- 2.1 GB
Released by Lucas Fraga through its official repository on Hugging Face.
Built From
- Described by arXiv:1803.05355
- Described by arXiv:2103.08541
- Described by arXiv:2406.04093
- Described by arXiv:2411.14257
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 2.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About fact-checking-abstention-saes
What license is fact-checking-abstention-saes released under?
other, as its publisher declares it. Read the license text before commercial use.