Jun1801/telcollm-oran-training

TelcoLLM ORANBench training data Training inputs for a Qwen3-8B run against ORANBench. The code lives in git; this repo carries only the data, which is too large for it. Generated 2026-09-03. Contents path size what it is train_ready_v1/ 162 MB cleaned Product A and B, with true token counts -- the inputs to every mix replay/ 19 MB in-domain and general replay, already through the split firewall gate splits_v2/ 0 MB the document-family split every… See the full description on the dataset page: https://huggingface.co/datasets/Jun1801/telcollm-oran-training.