GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
promotion
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
promotion
Model Tree
Qwen/Qwen2.5-1.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
Research checkpoint for the RONPO AAAI revision experiments.
Qwen/Qwen2.5-1.5B-Instruct/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p5_lr2p5e7{"base_model": "Qwen/Qwen2.5-1.5B-Instruct", "candidate": "r2_a0p5_lr2p5e7", "disclosure": "The sealed-test primary metric did not exceed the strongest baseline; public release is for reproducibility.", "effective_batch_size": 16, "final_test_gate": {"best_other_avg": {"model": "ipo_b0p05", "value": 0.5300850005739601}, "best_other_min_objective": {"model": "dpo_b0p01", "value": 0.5008064516129033}, "best_other_worst": {"model": "simpo_b2_g0p6", "value": 0.3150131166406124}, "num_prompts": 620, "paired_avg_difference_vs_best_other": {"ci95": [-0.04167766235267542, 0.01144110888198208], "value": -0.015930326311151405}, "paired_worst_difference_vs_best_other": {"ci95": [-0.055622629552722985, 0.011453373395125225], "value": -0.022017087033326416}, "passed_all_three": false, "passed_primary": false, "primary_objectives": ["helpfulness", "brevity"], "ronpo": {"avg": 0.5141546742628088, "min_objective": 0.5008064516129033, "worst": 0.292996029607286, "worst_ci95": [0.2699515108111762, 0.31703314686935]}, "winner_config": "r2_a0p5_lr2p5e7"}, "method": "RONPO full objective-response adversary", "optimizer_steps": 900, "seed": 42, "validation_selection": {"baselines": {"base": {"avg": 0.5180528085549307, "worst": 0.31452319037933574}, "dpo_b0p01": {"avg": 0.5284139789951168, "worst": 0.3207707964187358}, "inpo_eta0p0075": {"avg": 0.5185256846638394, "worst": 0.3167592763545738}, "ipo_b0p05": {"avg": 0.5257412197139942, "worst": 0.3243227234065469}, "simpo_b2_g0p6": {"avg": 0.5406167294313663, "worst": 0.3375748553984777}, "sppo_eta0p0075": {"avg": 0.5163700931828361, "worst": 0.3158720253020843}}, "best_other": {"avg": 0.5406167294313663, "name": "simpo_b2_g0p6", "worst": 0.3375748553984777}, "passed_primary_validation": true, "primary_objectives": ["helpfulness", "brevity"], "ronpo_candidates": [{"avg": 0.5389850032443516, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p5_lr2p5e7", "name": "r2_a0p5_lr2p5e7", "worst": 0.34022008377140195}, {"avg": 0.5414964972091004, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p5_anchor5e2", "name": "r2_a0p5_anchor5e2", "worst": 0.3402188976400124}, {"avg": 0.5362902709541706, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p3_lr5e7", "name": "r2_a0p3_lr5e7", "worst": 0.3324485763173458}, {"avg": 0.5301327076826696, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p2_lr5e7", "name": "r2_a0p2_lr5e7", "worst": 0.33237054087082274}, {"avg": 0.5359531842805365, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p1_lr5e7", "name": "r2_a0p1_lr5e7", "worst": 0.32402568220783473}, {"avg": 0.5327010736222885, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/ronpo_full", "name": "ronpo_full", "worst": 0.32385778295185924}, {"avg": 0.5238942543217426, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a1_lr1e7_anchor5e2", "name": "r2_a1_lr1e7_anchor5e2", "worst": 0.3234747800367659}, {"avg": 0.5229780602487047, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/ronpo_anchor1e2", "name": "ronpo_anchor1e2", "worst": 0.3226719777182472}, {"avg": 0.5276782368841395, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p4_anchor2e2", "name": "r2_a0p4_anchor2e2", "worst": 0.3149687548384218}, {"avg": 0.5183157343672885, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/ronpo_alpha0p5", "name": "ronpo_alpha0p5", "worst": 0.3116347886997874}, {"avg": 0.520138578697858, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a1_lr2p5e7_anchor2e2", "name": "r2_a1_lr2p5e7_anchor2e2", "worst": 0.307262927780274}], "selection_rule": "max primary prompt-worst normalized score, then primary prompt-avg", "selection_split": "same 500 prompt-disjoint validation prompts held out before training", "test_opened": false, "winner": {"avg": 0.5389850032443516, "model": "/NHNHOME/WORKSPACE/26msit001_A/BASE/aipr_lab_sjkim_eval/ronpo_aaai_8h_20260712/models/r2_a0p5_lr2p5e7", "name": "r2_a0p5_lr2p5e7", "worst": 0.34022008377140195}}, "wandb_url": "https://wandb.ai/promotion-kim/mnpo/runs/2ppvatvi"}Selected on 500 prompt-disjoint validation prompts. On the sealed 620-prompt test, helpfulness--brevity prompt-worst was 0.292996 versus 0.315013 for SimPO; paired difference -0.022017, 95% CI [-0.055623, 0.011453]. This is a negative primary result and the checkpoint is released for transparent reproducibility, not as evidence of superiority.
Intended use: reproducibility and evaluation for the RONPO paper. This checkpoint is not intended as a production assistant.