Sakana AIDroPE
Llama-2-7b-hf-DroPE
Sakana publishes a Llama 2 7B checkpoint for its DroPE method. The method removes positional embeddings and recalibrates a pretrained model.
Research release
A training recipe does not imply every model was released
The repository links this checkpoint for inference. It also includes training configurations for several SmolLM models. Those configurations alone do not establish downloadable trained SmolLM-DroPE releases.DroPE release and recalibration workflow
Measure useful context on the chosen task
Practical guidance. Inspect retrieval accuracy as the input grows. A method designed to extend context does not supply an unconditional guarantee that every distant detail remains usable.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Access | Downloadable |
| Model identifiers | SakanaAI/Llama-2-7b-hf-DroPE |
| Base model | Meta Llama 2 7BDroPE release and recalibration workflow |
| Weight terms | No checkpoint-specific license was identified in the reviewed file listing; Llama 2 ancestry remains relevant.Llama-2-7b-hf-DroPE weights and configuration |