agu18dec/

longcode-50k-t5gemma-2b-inference

50k code samples with long context for T5-Gemma 2B inference

Samples: 50,000 Model: T5-Gemma 2B (google/t5gemma-2b-2b-prefixlm-it) Format: JSON with teacher logits for knowledge distillation

Each sample contains:

encoder_input_ids: Long context input tokens (~52k tokens avg) decoder_input_ids: Output sequence tokens (~296 tokens avg) reasoning_mask: Mask for reasoning tokens teacher_token_ids: Teacher model's predicted… See the full description on the dataset page: https://huggingface.co/datasets/agu18dec/longcode-50k-t5gemma-2b-inference.