YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Phi-4-mini-instruct โ TTRL single-model GRPO on MATH345 (LLM side)
TTRL: majority vote over K samples as pseudo-label (no ground truth). 136 steps = 1 epoch, 128 prompts/update, K=12, beta=0, lr 3e-6, bnpo loss, adam_beta2 0.95, eval every 10 steps. best/ = best-by-val (step 120); endpoint/ = step 136; training/ = train.log, trainer_state, best_metric.
Consolidated from the earlier split repos (-best / -endpoint). Local source: trl-projects-llm/work_dirs/stage/ttrl__microsoft_Phi-4-mini-instruct__20260826_221628
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐ Ask for provider support