Pythia 1.4B Deduped
Pythia 1.4B Deduped is the deduplicated-Pile counterpart of Pythia 1.4B. It preserves the architecture while changing the training data.
Research release
A deeper deduplicated model than Pythia 1B
The 24-layer architecture shares its hidden width with the 1B model but uses more layers and attention heads. This release applies that architecture to deduplicated Pile data. The matched standard model provides the direct data comparison.pythia-1.4b-deduped model cardpythia-1.4b model configuration
Weights and training revisions
The release includes 154 saved training steps. The data order is shared with other sizes in the same suite. The main branch provides the final model; a step revision selects an earlier checkpoint. This base model has not been trained as a conversational assistant.pythia-1.4b-deduped model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Status | Research release |
| Launch date | |
| Access | Downloadable |
| Model identifiers | EleutherAI/pythia-1.4b-deduped |
| Context window | 2,048 tokens Maximum position count in the released model configuration. pythia-1.4b model configuration |
| Nominal parameters | 1.4B including embeddingspythia-1.4b-deduped model card |
| Network | 24 layers; hidden width 2,048; 16 attention headspythia-1.4b model configuration |
| Training data | Deduplicated Pilepythia-1.4b-deduped model card |
| License | Apache 2.0pythia-1.4b-deduped model card |
Related models
- Pythia 1.4B
EleutherAI ยท Pythia
A 24-layer Pythia network with a two-thousand-wide hidden state.
Launch date Research releaseDownloadable