encoder-decoder
encoder-decoder-floresp-scoresencoder-decoder-trial-stat
Encoder/decoder trial: encoder-marginal report
Dataset: G-reen/encoder-decoder-trial-stat
Rows analysed: 122,933 (every kept (encoder, decoder, source row) triple; source G-reen/cc-re-2021-filtered shard 0, 2000 rows of at most 4000 words)
Prompt file: prompts/indirect_reference_dataset_train.json (turn 0 encodes the document, turn 1 reconstructs it from the encoding alone)
Encoders: 9 (granite-4.2-30b-nvfp4 [0], Ornith-1.5-35B-A3B-NVFP4 [1], Llama-3.3-70B-Instruct-NVFP4 [2]… See the full description on the dataset page: https://huggingface.co/datasets/G-reen/encoder-decoder-trial-stat.encoder-decoder-trial-rewritten
Encoder/decoder trial: decoded texts
Turn-1 outputs: each config shard_<index> is one decoder's reconstruction of every encoder's encodings from G-reen/encoder-decoder-trial-encodings. encoder_model / decoder_model name the pair; response_0 is the encoding the decoder saw. Rows failing post-processing are in shard_<index>_trashed_data; per-run counters are in runs/.
encoder-decoder-trial-encodings
Encoder/decoder trial: encodings
Turn-0 outputs of the indirect-reference prompts: each config encoder_<index> is one model's encoding of the same 2000 rows (source_row of G-reen/cc-re-2021-filtered shard 0, at most 4000 words). prompt records the template each row was given, identically across encoders. Per-run counters are in runs/.
encoder-decoder-trial-human-editlensEncoderDecoderLora