Datasetpublic

eve-esa/EVE-Instruct-Dspark-training-data

EVE-Instruct D-Spark Training Data Prepared speculative-decoding training data for eve-esa/EVE-Instruct. Source: eve-esa/synth Source splits: qa and long_qa only Rows: 614,960 Format: EVE system prompt + question (input) + answer (output) Source context and file_path fields omitted Maximum sequence length: 4096 tokens Minimum trainable assistant tokens: 16 Columns: input_ids, loss_mask, seq_len loss_mask is 1 only for assistant response tokens (including EOS). token_freq.pt… See the full description on the dataset page: https://huggingface.co/datasets/eve-esa/EVE-Instruct-Dspark-training-data.

sourceHugging Faceotherupdated 2mo agoView on Hugging Face

0likes65downloads

Dataset Card

EVE-Instruct D-Spark Training Data

Prepared speculative-decoding training data for eve-esa/EVE-Instruct.

—Source: eve-esa/synth
—Source splits: qa and long_qa only
—Rows: 614,960
—Format: EVE system prompt + question (input) + answer (output)
—Source context and file_path fields omitted
—Maximum sequence length: 4096 tokens
—Minimum trainable assistant tokens: 16
—Columns: input_ids, loss_mask, seq_len

loss_mask is 1 only for assistant response tokens (including EOS). token_freq.pt contains the token-frequency distribution used to construct the draft vocabulary mapping.