glm5-reap50-tunecomp-sft

# glm5-reap50-tunecomp-sft Supervised chat finetuning dataset used for the GLM-5 REAP-50 TuneComp LoRA recovery run. ## Contents - `training_data.jsonl`: 754 chat samples - `manifest.json`: row counts and provenance ## Schema Each JSONL row contains: - `id` - `category` - `messages` - `prompt` - `response` - `content_length` - `reasoning_length` `messages` is a 3-message chat list with `system`, `user`, and `assistant` roles. ## Provenance - Teacher source: remote `glm-5` responses collected via Z.AI API - Transformation: converted into SFT-style chat training data - Student base used in training: `GLM-5-REAP-50pct-FP8-scalefix` ## Notes This is the dataset artifact from the March 28, 2026 B200 TuneComp run.