Antoine Chaffin, Benjamin Van Durme, Marc Marone, Orion Weller, Kathryn Ricci, Dawn Lawrie
We lifted 10 functions out of this paper's own repositories and ran 0 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| jhu-clsp/ettin-encoder-vs-decoder | canonical | 0 of 10 |
| Function | Status | Where it lives |
|---|---|---|
| copy_data_folder | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/sample_down_or_up_to_meet.py code served (permissive licence) · get_code("907d1d57fc3bfcb9") |
| count_jsonl_lines | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/jsonl_to_mds.py code served (permissive licence) · get_code("e051b521b313e36e") |
| count_samples | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/count_instances.py code served (permissive licence) · get_code("52f254918710203e") |
| create_model_key | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/bias_eval/batch_eval.py code served (permissive licence) · get_code("805a542cd76d0dc6") |
| create_visualization | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/bias_eval/create_plot.py code served (permissive licence) · get_code("be4df63ffad5921c") |
| find_all_stats_files | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/sample_down_or_up_to_meet.py code served (permissive licence) · get_code("9c036669e7a39b8c") |
| get_tokens_from_stats | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/sample_down_or_up_to_meet.py code served (permissive licence) · get_code("d1c8fcebd8f6e32a") |
| load_existing_results | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/bias_eval/batch_eval.py code served (permissive licence) · get_code("ae71360e40902e60") |
| parse_model_data | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/bias_eval/create_plot.py code served (permissive licence) · get_code("a6cd85fd2f6cd892") |
| recalculate_percentages | Not yet run | jhu-clsp/ettin-encoder-vs-decoder/pretraining/data_processing/bin/sample_for_context_extension_random.py code served (permissive licence) · get_code("3be0f2f6856a628f") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
The large language model (LLM) community focuses almost exclusively on decoder-only language models, since they are easier to use for text generation. However, a large subset of the community still uses encoder-only models for tasks such as classification or retrieval. Previous work has attempted to compare these architectures, but is forced to make comparisons with models that have different numbers of parameters, training techniques, and datasets. We introduce the SOTA open-data ETTIN 1 suite of models: paired encoder-only and decoder-only models ranging from 17 million parameters to 1 billion, trained on up to 2 trillion tokens. Using the same recipe for both encoder-only and decoder-only models produces SOTA recipes in both categories for their respective sizes, beating ModernBERT as an encoder and Llama 3.2 and SmolLM2 as decoders. Like previous work, we find that encoder-only models excel at classification and retrieval tasks while decoders excel at generative tasks. However, we show that adapting a decoder model to encoder tasks (and vice versa) through continued training is subpar compared to using only the reverse objective (i.e. a 400M encoder outperforms a 1B decoder on MNLI, and vice versa for generative tasks). We open-source all artifacts of this study including training data, training order segmented by checkpoint, and 200+ checkpoints to allow future work to analyze or extend all aspects of training. 2
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2507.11412")
get_code_for_paper("2507.11412")
have("2507.11412")
Connect an agent — have() is free.