{ "best_global_step": null, "best_metric": null, "best_model_checkpoint": null, "epoch": 1.0, "eval_steps": 200, "global_step": 681, "is_hyper_param_search": false, "is_local_process_zero": true, "is_world_process_zero": true, "log_history": [ { "epoch": 0.0014684287812041115, "fcm_dpo/beta": 0.10000000149011612, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": -0.02287006378173828, "fcm_dpo/q_t": 0.5005706548690796, "grad_norm": 83.52236938476562, "learning_rate": 0.0, "logits/chosen": -0.4974287748336792, "logits/rejected": -0.43299180269241333, "logps/chosen": -50.1435661315918, "logps/ref_chosen": -50.14883804321289, "logps/ref_rejected": -74.1280517578125, "logps/rejected": -74.09991455078125, "loss": 1.389, "margin_dpo/margin_mean": -0.02287048101425171, "margin_dpo/margin_std": 0.41920793056488037, "step": 1 }, { "epoch": 0.007342143906020558, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": -0.03249748796224594, "fcm_dpo/q_t": 0.5008123517036438, "grad_norm": 89.65718078613281, "learning_rate": 2.898550724637681e-08, "logits/chosen": -0.4900774359703064, "logits/rejected": -0.453380823135376, "logps/chosen": -56.06619644165039, "logps/ref_chosen": -56.05734634399414, "logps/ref_rejected": -78.69325256347656, "logps/rejected": -78.66961669921875, "loss": 1.3898, "margin_dpo/margin_mean": -0.03249724209308624, "margin_dpo/margin_std": 0.34048572182655334, "step": 5 }, { "epoch": 0.014684287812041116, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": -0.020858580246567726, "fcm_dpo/q_t": 0.5005213022232056, "grad_norm": 70.88235473632812, "learning_rate": 6.521739130434782e-08, "logits/chosen": -0.5124648809432983, "logits/rejected": -0.4746035933494568, "logps/chosen": -59.54156494140625, "logps/ref_chosen": -59.54457473754883, "logps/ref_rejected": -91.17041778564453, "logps/rejected": -91.14656066894531, "loss": 1.3888, "margin_dpo/margin_mean": -0.02085849642753601, "margin_dpo/margin_std": 0.38151517510414124, "step": 10 }, { "epoch": 0.022026431718061675, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": -0.008700313977897167, "fcm_dpo/q_t": 0.5002168416976929, "grad_norm": 64.12911987304688, "learning_rate": 1.0144927536231885e-07, "logits/chosen": -0.4861031174659729, "logits/rejected": -0.45661497116088867, "logps/chosen": -58.85633087158203, "logps/ref_chosen": -58.83195877075195, "logps/ref_rejected": -92.93949890136719, "logps/rejected": -92.95516967773438, "loss": 1.3876, "margin_dpo/margin_mean": -0.008700350299477577, "margin_dpo/margin_std": 0.3919045627117157, "step": 15 }, { "epoch": 0.02936857562408223, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": 0.060948461294174194, "fcm_dpo/q_t": 0.4984763562679291, "grad_norm": 73.97709655761719, "learning_rate": 1.3768115942028986e-07, "logits/chosen": -0.5145794153213501, "logits/rejected": -0.46688389778137207, "logps/chosen": -59.64154815673828, "logps/ref_chosen": -59.6396598815918, "logps/ref_rejected": -82.76026916503906, "logps/rejected": -82.82310485839844, "loss": 1.3805, "margin_dpo/margin_mean": 0.06094840168952942, "margin_dpo/margin_std": 0.3420618772506714, "step": 20 }, { "epoch": 0.03671071953010279, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": 0.19125637412071228, "fcm_dpo/q_t": 0.49522167444229126, "grad_norm": 74.20327758789062, "learning_rate": 1.7391304347826085e-07, "logits/chosen": -0.5046600103378296, "logits/rejected": -0.47779035568237305, "logps/chosen": -53.17488479614258, "logps/ref_chosen": -53.205284118652344, "logps/ref_rejected": -88.99608612060547, "logps/rejected": -89.15694427490234, "loss": 1.3678, "margin_dpo/margin_mean": 0.19125640392303467, "margin_dpo/margin_std": 0.4390522539615631, "step": 25 }, { "epoch": 0.04405286343612335, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": 0.465215265750885, "fcm_dpo/q_t": 0.4883840084075928, "grad_norm": 87.46487426757812, "learning_rate": 2.1014492753623187e-07, "logits/chosen": -0.524259626865387, "logits/rejected": -0.4842991232872009, "logps/chosen": -53.4693603515625, "logps/ref_chosen": -53.5526008605957, "logps/ref_rejected": -97.87371826171875, "logps/rejected": -98.25569152832031, "loss": 1.3412, "margin_dpo/margin_mean": 0.4652155041694641, "margin_dpo/margin_std": 0.5838907957077026, "step": 30 }, { "epoch": 0.0513950073421439, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": 0.7807711362838745, "fcm_dpo/q_t": 0.4805462956428528, "grad_norm": 83.3965072631836, "learning_rate": 2.463768115942029e-07, "logits/chosen": -0.5029053092002869, "logits/rejected": -0.4691234529018402, "logps/chosen": -56.19721221923828, "logps/ref_chosen": -56.3298454284668, "logps/ref_rejected": -91.76858520507812, "logps/rejected": -92.41671752929688, "loss": 1.3116, "margin_dpo/margin_mean": 0.7807710766792297, "margin_dpo/margin_std": 0.820033848285675, "step": 35 }, { "epoch": 0.05873715124816446, "fcm_dpo/beta": 0.10000000894069672, "fcm_dpo/delta": 0.0, "fcm_dpo/margin": 1.3780633211135864, "fcm_dpo/q_t": 0.46588778495788574, "grad_norm": 60.05709457397461, "learning_rate": 2.8260869565217386e-07, "logits/chosen": -0.5348217487335205, "logits/rejected": -0.4983762204647064, "logps/chosen": -54.2760009765625, "logps/ref_chosen": -54.38492965698242, "logps/ref_rejected": -82.94353485107422, "logps/rejected": -84.21268463134766, "loss": 1.2592, "margin_dpo/margin_mean": 1.3780632019042969, "margin_dpo/margin_std": 1.4964091777801514, "step": 40 }, { "epoch": 0.06607929515418502, "fcm_dpo/beta": 0.13015565276145935, "fcm_dpo/delta": 0.4826194643974304, "fcm_dpo/margin": 2.396104335784912, "fcm_dpo/q_t": 0.42787855863571167, "grad_norm": 84.72518920898438, "learning_rate": 3.188405797101449e-07, "logits/chosen": -0.5097236633300781, "logits/rejected": -0.4814884662628174, "logps/chosen": -54.64501953125, "logps/ref_chosen": -54.862335205078125, "logps/ref_rejected": -98.0264663696289, "logps/rejected": -100.20524597167969, "loss": 1.1198, "margin_dpo/margin_mean": 2.396104335784912, "margin_dpo/margin_std": 2.2702395915985107, "step": 45 }, { "epoch": 0.07342143906020558, "fcm_dpo/beta": 0.189666748046875, "fcm_dpo/delta": 0.21794584393501282, "fcm_dpo/margin": 3.3721413612365723, "fcm_dpo/q_t": 0.36233529448509216, "grad_norm": 93.39527130126953, "learning_rate": 3.5507246376811595e-07, "logits/chosen": -0.5606456398963928, "logits/rejected": -0.5093287229537964, "logps/chosen": -58.14433670043945, "logps/ref_chosen": -58.304595947265625, "logps/ref_rejected": -91.69480895996094, "logps/rejected": -94.90668487548828, "loss": 0.9364, "margin_dpo/margin_mean": 3.3721413612365723, "margin_dpo/margin_std": 3.3471221923828125, "step": 50 }, { "epoch": 0.08076358296622614, "fcm_dpo/beta": 0.1844942271709442, "fcm_dpo/delta": -0.19296136498451233, "fcm_dpo/margin": 5.56916618347168, "fcm_dpo/q_t": 0.30575454235076904, "grad_norm": 78.50489807128906, "learning_rate": 3.9130434782608694e-07, "logits/chosen": -0.5810452699661255, "logits/rejected": -0.5354865789413452, "logps/chosen": -56.305625915527344, "logps/ref_chosen": -56.06591796875, "logps/ref_rejected": -85.69367980957031, "logps/rejected": -91.50254821777344, "loss": 0.8026, "margin_dpo/margin_mean": 5.5691657066345215, "margin_dpo/margin_std": 5.670819282531738, "step": 55 }, { "epoch": 0.0881057268722467, "fcm_dpo/beta": 0.14939136803150177, "fcm_dpo/delta": -0.18733538687229156, "fcm_dpo/margin": 6.829007148742676, "fcm_dpo/q_t": 0.3050878942012787, "grad_norm": 80.88507080078125, "learning_rate": 4.2753623188405794e-07, "logits/chosen": -0.5981510877609253, "logits/rejected": -0.5482528805732727, "logps/chosen": -61.76525115966797, "logps/ref_chosen": -60.6871337890625, "logps/ref_rejected": -89.72715759277344, "logps/rejected": -97.63427734375, "loss": 0.8231, "margin_dpo/margin_mean": 6.829007148742676, "margin_dpo/margin_std": 7.180797576904297, "step": 60 }, { "epoch": 0.09544787077826726, "fcm_dpo/beta": 0.1279543936252594, "fcm_dpo/delta": -0.15479923784732819, "fcm_dpo/margin": 7.741368293762207, "fcm_dpo/q_t": 0.31432515382766724, "grad_norm": 60.72844696044922, "learning_rate": 4.63768115942029e-07, "logits/chosen": -0.6157333850860596, "logits/rejected": -0.5836382508277893, "logps/chosen": -63.23627853393555, "logps/ref_chosen": -61.75325393676758, "logps/ref_rejected": -93.30108642578125, "logps/rejected": -102.52547454833984, "loss": 0.8599, "margin_dpo/margin_mean": 7.741368293762207, "margin_dpo/margin_std": 8.94554615020752, "step": 65 }, { "epoch": 0.1027900146842878, "fcm_dpo/beta": 0.11097879707813263, "fcm_dpo/delta": -0.13162511587142944, "fcm_dpo/margin": 8.74174690246582, "fcm_dpo/q_t": 0.31369131803512573, "grad_norm": 56.712982177734375, "learning_rate": 5e-07, "logits/chosen": -0.6086438894271851, "logits/rejected": -0.5690315961837769, "logps/chosen": -61.81795120239258, "logps/ref_chosen": -59.548004150390625, "logps/ref_rejected": -84.01609802246094, "logps/rejected": -95.02778625488281, "loss": 0.8231, "margin_dpo/margin_mean": 8.741745948791504, "margin_dpo/margin_std": 9.22284984588623, "step": 70 }, { "epoch": 0.11013215859030837, "fcm_dpo/beta": 0.09055890142917633, "fcm_dpo/delta": -0.27352195978164673, "fcm_dpo/margin": 12.153611183166504, "fcm_dpo/q_t": 0.2983909249305725, "grad_norm": 43.50236129760742, "learning_rate": 4.999176576834721e-07, "logits/chosen": -0.6517193913459778, "logits/rejected": -0.6220844388008118, "logps/chosen": -63.3004264831543, "logps/ref_chosen": -59.86931228637695, "logps/ref_rejected": -98.05613708496094, "logps/rejected": -113.6408462524414, "loss": 0.8066, "margin_dpo/margin_mean": 12.153611183166504, "margin_dpo/margin_std": 13.398017883300781, "step": 75 }, { "epoch": 0.11747430249632893, "fcm_dpo/beta": 0.07161784917116165, "fcm_dpo/delta": -0.15674835443496704, "fcm_dpo/margin": 13.832998275756836, "fcm_dpo/q_t": 0.3114686608314514, "grad_norm": 41.86962890625, "learning_rate": 4.996706849759452e-07, "logits/chosen": -0.6776489615440369, "logits/rejected": -0.6389704942703247, "logps/chosen": -60.72328567504883, "logps/ref_chosen": -56.18925857543945, "logps/ref_rejected": -86.42393493652344, "logps/rejected": -104.79096984863281, "loss": 0.834, "margin_dpo/margin_mean": 13.832998275756836, "margin_dpo/margin_std": 14.778533935546875, "step": 80 }, { "epoch": 0.12481644640234948, "fcm_dpo/beta": 0.06048274785280228, "fcm_dpo/delta": -0.15613125264644623, "fcm_dpo/margin": 16.303586959838867, "fcm_dpo/q_t": 0.3134708106517792, "grad_norm": 42.009918212890625, "learning_rate": 4.992592445678582e-07, "logits/chosen": -0.6713468432426453, "logits/rejected": -0.6383522748947144, "logps/chosen": -66.5967788696289, "logps/ref_chosen": -60.018287658691406, "logps/ref_rejected": -98.01185607910156, "logps/rejected": -120.89393615722656, "loss": 0.8553, "margin_dpo/margin_mean": 16.303586959838867, "margin_dpo/margin_std": 18.325000762939453, "step": 85 }, { "epoch": 0.13215859030837004, "fcm_dpo/beta": 0.055021800100803375, "fcm_dpo/delta": -0.09340845793485641, "fcm_dpo/margin": 16.962265014648438, "fcm_dpo/q_t": 0.3316206634044647, "grad_norm": 42.72884750366211, "learning_rate": 4.986836074908615e-07, "logits/chosen": -0.6945492625236511, "logits/rejected": -0.6738761067390442, "logps/chosen": -69.03782653808594, "logps/ref_chosen": -59.8709831237793, "logps/ref_rejected": -96.78519439697266, "logps/rejected": -122.91429138183594, "loss": 0.9273, "margin_dpo/margin_mean": 16.962265014648438, "margin_dpo/margin_std": 22.208675384521484, "step": 90 }, { "epoch": 0.1395007342143906, "fcm_dpo/beta": 0.049969423562288284, "fcm_dpo/delta": -0.06735582649707794, "fcm_dpo/margin": 18.21504783630371, "fcm_dpo/q_t": 0.3252207636833191, "grad_norm": 34.471012115478516, "learning_rate": 4.979441529392784e-07, "logits/chosen": -0.709299623966217, "logits/rejected": -0.6731777787208557, "logps/chosen": -64.95755004882812, "logps/ref_chosen": -55.94385528564453, "logps/ref_rejected": -83.6790542602539, "logps/rejected": -110.90779876708984, "loss": 0.8711, "margin_dpo/margin_mean": 18.215045928955078, "margin_dpo/margin_std": 20.571401596069336, "step": 95 }, { "epoch": 0.14684287812041116, "fcm_dpo/beta": 0.04660476744174957, "fcm_dpo/delta": -0.0532207265496254, "fcm_dpo/margin": 19.219341278076172, "fcm_dpo/q_t": 0.32488057017326355, "grad_norm": 40.7348518371582, "learning_rate": 4.970413680203148e-07, "logits/chosen": -0.670303225517273, "logits/rejected": -0.6368999481201172, "logps/chosen": -66.36493682861328, "logps/ref_chosen": -57.05888748168945, "logps/ref_rejected": -86.11727142333984, "logps/rejected": -114.64266204833984, "loss": 0.8736, "margin_dpo/margin_mean": 19.219341278076172, "margin_dpo/margin_std": 21.64118194580078, "step": 100 }, { "epoch": 0.15418502202643172, "fcm_dpo/beta": 0.04551302269101143, "fcm_dpo/delta": -0.15303857624530792, "fcm_dpo/margin": 21.88186264038086, "fcm_dpo/q_t": 0.315370112657547, "grad_norm": 30.820974349975586, "learning_rate": 4.959758474331832e-07, "logits/chosen": -0.7173506021499634, "logits/rejected": -0.6785070896148682, "logps/chosen": -70.07099914550781, "logps/ref_chosen": -59.20774459838867, "logps/ref_rejected": -86.49754333496094, "logps/rejected": -119.24266052246094, "loss": 0.8519, "margin_dpo/margin_mean": 21.88186264038086, "margin_dpo/margin_std": 24.000478744506836, "step": 105 }, { "epoch": 0.16152716593245228, "fcm_dpo/beta": 0.03847113624215126, "fcm_dpo/delta": -0.08476176112890244, "fcm_dpo/margin": 24.109981536865234, "fcm_dpo/q_t": 0.3201691508293152, "grad_norm": 30.027286529541016, "learning_rate": 4.947482930773511e-07, "logits/chosen": -0.6687729954719543, "logits/rejected": -0.6251232624053955, "logps/chosen": -72.03107452392578, "logps/ref_chosen": -60.437957763671875, "logps/ref_rejected": -90.83917999267578, "logps/rejected": -126.54228210449219, "loss": 0.8568, "margin_dpo/margin_mean": 24.109981536865234, "margin_dpo/margin_std": 25.9851016998291, "step": 110 }, { "epoch": 0.16886930983847284, "fcm_dpo/beta": 0.03624765947461128, "fcm_dpo/delta": -0.01841454766690731, "fcm_dpo/margin": 23.8579044342041, "fcm_dpo/q_t": 0.33425024151802063, "grad_norm": 42.781307220458984, "learning_rate": 4.933595135901732e-07, "logits/chosen": -0.6857893466949463, "logits/rejected": -0.6449792385101318, "logps/chosen": -76.1856918334961, "logps/ref_chosen": -61.7908821105957, "logps/ref_rejected": -85.36943054199219, "logps/rejected": -123.62214660644531, "loss": 0.903, "margin_dpo/margin_mean": 23.8579044342041, "margin_dpo/margin_std": 28.585891723632812, "step": 115 }, { "epoch": 0.1762114537444934, "fcm_dpo/beta": 0.0364445336163044, "fcm_dpo/delta": -0.015237887389957905, "fcm_dpo/margin": 23.689233779907227, "fcm_dpo/q_t": 0.33086472749710083, "grad_norm": 33.90849304199219, "learning_rate": 4.918104238142103e-07, "logits/chosen": -0.6950192451477051, "logits/rejected": -0.6637409925460815, "logps/chosen": -81.74562072753906, "logps/ref_chosen": -65.3261489868164, "logps/ref_rejected": -86.75518798828125, "logps/rejected": -126.8638916015625, "loss": 0.8772, "margin_dpo/margin_mean": 23.68923568725586, "margin_dpo/margin_std": 26.572423934936523, "step": 120 }, { "epoch": 0.18355359765051396, "fcm_dpo/beta": 0.03069928288459778, "fcm_dpo/delta": -0.22570106387138367, "fcm_dpo/margin": 34.196529388427734, "fcm_dpo/q_t": 0.3057743012905121, "grad_norm": 28.747299194335938, "learning_rate": 4.90102044194588e-07, "logits/chosen": -0.6552901268005371, "logits/rejected": -0.645017147064209, "logps/chosen": -76.18916320800781, "logps/ref_chosen": -58.323204040527344, "logps/ref_rejected": -101.2106704711914, "logps/rejected": -153.27313232421875, "loss": 0.8245, "margin_dpo/margin_mean": 34.196529388427734, "margin_dpo/margin_std": 37.522029876708984, "step": 125 }, { "epoch": 0.19089574155653452, "fcm_dpo/beta": 0.02867107465863228, "fcm_dpo/delta": -0.030061205849051476, "fcm_dpo/margin": 30.552875518798828, "fcm_dpo/q_t": 0.33143970370292664, "grad_norm": 29.514667510986328, "learning_rate": 4.882355001067891e-07, "logits/chosen": -0.6441887617111206, "logits/rejected": -0.6197634935379028, "logps/chosen": -75.02465057373047, "logps/ref_chosen": -56.38518524169922, "logps/ref_rejected": -86.15767669677734, "logps/rejected": -135.35000610351562, "loss": 0.9009, "margin_dpo/margin_mean": 30.552875518798828, "margin_dpo/margin_std": 35.57330322265625, "step": 130 }, { "epoch": 0.19823788546255505, "fcm_dpo/beta": 0.02630027011036873, "fcm_dpo/delta": -0.08074788004159927, "fcm_dpo/margin": 35.100746154785156, "fcm_dpo/q_t": 0.3213648200035095, "grad_norm": 30.23385238647461, "learning_rate": 4.862120211153265e-07, "logits/chosen": -0.6420992612838745, "logits/rejected": -0.6323195695877075, "logps/chosen": -74.02909851074219, "logps/ref_chosen": -54.59065628051758, "logps/ref_rejected": -95.26080322265625, "logps/rejected": -149.7999725341797, "loss": 0.8502, "margin_dpo/margin_mean": 35.100746154785156, "margin_dpo/margin_std": 38.36650848388672, "step": 135 }, { "epoch": 0.2055800293685756, "fcm_dpo/beta": 0.024506056681275368, "fcm_dpo/delta": -0.07339946180582047, "fcm_dpo/margin": 37.39527130126953, "fcm_dpo/q_t": 0.3280564844608307, "grad_norm": 29.89729881286621, "learning_rate": 4.840329401637809e-07, "logits/chosen": -0.6269393563270569, "logits/rejected": -0.6138561367988586, "logps/chosen": -81.1108169555664, "logps/ref_chosen": -56.04347610473633, "logps/ref_rejected": -93.27880859375, "logps/rejected": -155.74142456054688, "loss": 0.8815, "margin_dpo/margin_mean": 37.39527130126953, "margin_dpo/margin_std": 43.980499267578125, "step": 140 }, { "epoch": 0.21292217327459617, "fcm_dpo/beta": 0.023285897448658943, "fcm_dpo/delta": -0.010504422709345818, "fcm_dpo/margin": 36.870033264160156, "fcm_dpo/q_t": 0.3340276777744293, "grad_norm": 34.29755401611328, "learning_rate": 4.816996926967401e-07, "logits/chosen": -0.5899958610534668, "logits/rejected": -0.5646438598632812, "logps/chosen": -89.64959716796875, "logps/ref_chosen": -61.4414176940918, "logps/ref_rejected": -86.32813262939453, "logps/rejected": -151.40634155273438, "loss": 0.9115, "margin_dpo/margin_mean": 36.870033264160156, "margin_dpo/margin_std": 44.208961486816406, "step": 145 }, { "epoch": 0.22026431718061673, "fcm_dpo/beta": 0.02298033982515335, "fcm_dpo/delta": -0.04273203760385513, "fcm_dpo/margin": 38.58696746826172, "fcm_dpo/q_t": 0.33299776911735535, "grad_norm": 31.706588745117188, "learning_rate": 4.792138157142157e-07, "logits/chosen": -0.5931132435798645, "logits/rejected": -0.5684978365898132, "logps/chosen": -89.40864562988281, "logps/ref_chosen": -57.70451736450195, "logps/ref_rejected": -87.76991271972656, "logps/rejected": -158.06101989746094, "loss": 0.9086, "margin_dpo/margin_mean": 38.58696746826172, "margin_dpo/margin_std": 47.16357421875, "step": 150 }, { "epoch": 0.2276064610866373, "fcm_dpo/beta": 0.02123543620109558, "fcm_dpo/delta": -0.0736246332526207, "fcm_dpo/margin": 43.11958694458008, "fcm_dpo/q_t": 0.32444706559181213, "grad_norm": 28.768526077270508, "learning_rate": 4.7657694675916247e-07, "logits/chosen": -0.6274883151054382, "logits/rejected": -0.6081691980361938, "logps/chosen": -90.71160888671875, "logps/ref_chosen": -62.08925247192383, "logps/ref_rejected": -94.79930114746094, "logps/rejected": -166.54122924804688, "loss": 0.8772, "margin_dpo/margin_mean": 43.11958694458008, "margin_dpo/margin_std": 49.604820251464844, "step": 155 }, { "epoch": 0.23494860499265785, "fcm_dpo/beta": 0.020709924399852753, "fcm_dpo/delta": -0.028781410306692123, "fcm_dpo/margin": 42.2672004699707, "fcm_dpo/q_t": 0.3394644856452942, "grad_norm": 35.6558837890625, "learning_rate": 4.737908228387656e-07, "logits/chosen": -0.6209047436714172, "logits/rejected": -0.602047324180603, "logps/chosen": -102.9297866821289, "logps/ref_chosen": -67.15288543701172, "logps/ref_rejected": -96.92537689208984, "logps/rejected": -174.969482421875, "loss": 0.95, "margin_dpo/margin_mean": 42.26719665527344, "margin_dpo/margin_std": 56.8797607421875, "step": 160 }, { "epoch": 0.2422907488986784, "fcm_dpo/beta": 0.019984444603323936, "fcm_dpo/delta": 0.019107891246676445, "fcm_dpo/margin": 41.58051300048828, "fcm_dpo/q_t": 0.33927860856056213, "grad_norm": 44.44900894165039, "learning_rate": 4.708572792802069e-07, "logits/chosen": -0.5925561189651489, "logits/rejected": -0.5611370801925659, "logps/chosen": -88.99661254882812, "logps/ref_chosen": -57.40401077270508, "logps/ref_rejected": -80.31498718261719, "logps/rejected": -153.48812866210938, "loss": 0.9112, "margin_dpo/margin_mean": 41.58051681518555, "margin_dpo/margin_std": 49.86974334716797, "step": 165 }, { "epoch": 0.24963289280469897, "fcm_dpo/beta": 0.018490394577383995, "fcm_dpo/delta": -0.12243499606847763, "fcm_dpo/margin": 51.66387939453125, "fcm_dpo/q_t": 0.3191189765930176, "grad_norm": 27.767526626586914, "learning_rate": 4.6777824852166437e-07, "logits/chosen": -0.5405124425888062, "logits/rejected": -0.528901994228363, "logps/chosen": -85.2691650390625, "logps/ref_chosen": -52.029144287109375, "logps/ref_rejected": -85.73944091796875, "logps/rejected": -170.64334106445312, "loss": 0.8528, "margin_dpo/margin_mean": 51.66387939453125, "margin_dpo/margin_std": 58.07769775390625, "step": 170 }, { "epoch": 0.25697503671071953, "fcm_dpo/beta": 0.01827237568795681, "fcm_dpo/delta": -0.062149811536073685, "fcm_dpo/margin": 49.67750549316406, "fcm_dpo/q_t": 0.32835322618484497, "grad_norm": 37.51227569580078, "learning_rate": 4.645557588393406e-07, "logits/chosen": -0.5422641038894653, "logits/rejected": -0.5189716815948486, "logps/chosen": -106.3938217163086, "logps/ref_chosen": -62.996971130371094, "logps/ref_rejected": -92.98394012451172, "logps/rejected": -186.0583038330078, "loss": 0.8859, "margin_dpo/margin_mean": 49.67750549316406, "margin_dpo/margin_std": 58.46363067626953, "step": 175 }, { "epoch": 0.2643171806167401, "fcm_dpo/beta": 0.01582510769367218, "fcm_dpo/delta": -0.1366010159254074, "fcm_dpo/margin": 61.5815315246582, "fcm_dpo/q_t": 0.31790629029273987, "grad_norm": 27.392812728881836, "learning_rate": 4.611919330113591e-07, "logits/chosen": -0.485221803188324, "logits/rejected": -0.48061603307724, "logps/chosen": -105.83013916015625, "logps/ref_chosen": -57.0670280456543, "logps/ref_rejected": -97.1115493774414, "logps/rejected": -207.4561767578125, "loss": 0.8569, "margin_dpo/margin_mean": 61.5815315246582, "margin_dpo/margin_std": 70.3671646118164, "step": 180 }, { "epoch": 0.27165932452276065, "fcm_dpo/beta": 0.015623442828655243, "fcm_dpo/delta": 0.13391907513141632, "fcm_dpo/margin": 46.54853057861328, "fcm_dpo/q_t": 0.3591853976249695, "grad_norm": 36.289039611816406, "learning_rate": 4.5768898691940836e-07, "logits/chosen": -0.5313425064086914, "logits/rejected": -0.5060660243034363, "logps/chosen": -97.86585998535156, "logps/ref_chosen": -54.840736389160156, "logps/ref_rejected": -75.51002502441406, "logps/rejected": -165.0836944580078, "loss": 0.9783, "margin_dpo/margin_mean": 46.54853057861328, "margin_dpo/margin_std": 60.472984313964844, "step": 185 }, { "epoch": 0.2790014684287812, "fcm_dpo/beta": 0.01625588908791542, "fcm_dpo/delta": -0.08447099477052689, "fcm_dpo/margin": 56.91190719604492, "fcm_dpo/q_t": 0.3247375786304474, "grad_norm": 40.74492645263672, "learning_rate": 4.5404922808905543e-07, "logits/chosen": -0.5222185254096985, "logits/rejected": -0.49877819418907166, "logps/chosen": -102.94236755371094, "logps/ref_chosen": -57.72148895263672, "logps/ref_rejected": -86.85997009277344, "logps/rejected": -188.99276733398438, "loss": 0.8786, "margin_dpo/margin_mean": 56.91190719604492, "margin_dpo/margin_std": 66.29203033447266, "step": 190 }, { "epoch": 0.28634361233480177, "fcm_dpo/beta": 0.014897446148097515, "fcm_dpo/delta": -0.034334540367126465, "fcm_dpo/margin": 58.907752990722656, "fcm_dpo/q_t": 0.32934439182281494, "grad_norm": 33.52628707885742, "learning_rate": 4.5027505416968985e-07, "logits/chosen": -0.4740094542503357, "logits/rejected": -0.4610044062137604, "logps/chosen": -110.0854263305664, "logps/ref_chosen": -58.26164627075195, "logps/ref_rejected": -89.46485900878906, "logps/rejected": -200.19639587402344, "loss": 0.8883, "margin_dpo/margin_mean": 58.907752990722656, "margin_dpo/margin_std": 67.69029235839844, "step": 195 }, { "epoch": 0.2936857562408223, "fcm_dpo/beta": 0.014423737302422523, "fcm_dpo/delta": -0.07692910730838776, "fcm_dpo/margin": 63.713844299316406, "fcm_dpo/q_t": 0.3252788186073303, "grad_norm": 31.517562866210938, "learning_rate": 4.4636895135509966e-07, "logits/chosen": -0.46841588616371155, "logits/rejected": -0.4528680443763733, "logps/chosen": -105.78775787353516, "logps/ref_chosen": -55.71953201293945, "logps/ref_rejected": -83.15235137939453, "logps/rejected": -196.93441772460938, "loss": 0.8697, "margin_dpo/margin_mean": 63.713844299316406, "margin_dpo/margin_std": 73.30004119873047, "step": 200 }, { "epoch": 0.2936857562408223, "eval_fcm_dpo/beta": 0.019768698140978813, "eval_fcm_dpo/delta": 0.06340710818767548, "eval_fcm_dpo/margin": 38.271907806396484, "eval_fcm_dpo/q_t": 0.37947365641593933, "eval_logits/chosen": -0.49050527811050415, "eval_logits/rejected": -0.47104698419570923, "eval_logps/chosen": -148.67208862304688, "eval_logps/ref_chosen": -79.05104064941406, "eval_logps/ref_rejected": -86.79793548583984, "eval_logps/rejected": -194.69088745117188, "eval_loss": 0.5858080387115479, "eval_margin_dpo/margin_mean": 38.271907806396484, "eval_margin_dpo/margin_std": 74.73805236816406, "eval_runtime": 39.5337, "eval_samples_per_second": 59.165, "eval_steps_per_second": 1.872, "step": 200 }, { "epoch": 0.3010279001468429, "fcm_dpo/beta": 0.018562760204076767, "fcm_dpo/delta": -0.2798296809196472, "fcm_dpo/margin": 58.922264099121094, "fcm_dpo/q_t": 0.30979180335998535, "grad_norm": 39.298316955566406, "learning_rate": 4.4233349274571974e-07, "logits/chosen": -0.4523549973964691, "logits/rejected": -0.43011409044265747, "logps/chosen": -121.86454772949219, "logps/ref_chosen": -65.13258361816406, "logps/ref_rejected": -92.10203552246094, "logps/rejected": -207.75625610351562, "loss": 0.8772, "margin_dpo/margin_mean": 58.9222526550293, "margin_dpo/margin_std": 72.63723754882812, "step": 205 }, { "epoch": 0.30837004405286345, "fcm_dpo/beta": 0.014536149799823761, "fcm_dpo/delta": -0.1799614131450653, "fcm_dpo/margin": 69.11328125, "fcm_dpo/q_t": 0.31058698892593384, "grad_norm": 31.66854476928711, "learning_rate": 4.381713366536311e-07, "logits/chosen": -0.4480575919151306, "logits/rejected": -0.43101295828819275, "logps/chosen": -109.48213195800781, "logps/ref_chosen": -54.52837371826172, "logps/ref_rejected": -87.06227111816406, "logps/rejected": -211.12930297851562, "loss": 0.8362, "margin_dpo/margin_mean": 69.11328125, "margin_dpo/margin_std": 75.44652557373047, "step": 210 }, { "epoch": 0.315712187958884, "fcm_dpo/beta": 0.013478484936058521, "fcm_dpo/delta": -0.06460683792829514, "fcm_dpo/margin": 67.31108093261719, "fcm_dpo/q_t": 0.3255953788757324, "grad_norm": 30.947105407714844, "learning_rate": 4.3388522485142885e-07, "logits/chosen": -0.47888296842575073, "logits/rejected": -0.4555312693119049, "logps/chosen": -116.98878479003906, "logps/ref_chosen": -59.905250549316406, "logps/ref_rejected": -90.25511932373047, "logps/rejected": -214.6497344970703, "loss": 0.8705, "margin_dpo/margin_mean": 67.31107330322266, "margin_dpo/margin_std": 74.4868392944336, "step": 215 }, { "epoch": 0.32305433186490456, "fcm_dpo/beta": 0.012834814377129078, "fcm_dpo/delta": 0.019026797264814377, "fcm_dpo/margin": 64.8343505859375, "fcm_dpo/q_t": 0.3348235487937927, "grad_norm": 32.80592346191406, "learning_rate": 4.2947798076611047e-07, "logits/chosen": -0.4836200177669525, "logits/rejected": -0.4622252583503723, "logps/chosen": -118.51663970947266, "logps/ref_chosen": -57.68498611450195, "logps/ref_rejected": -87.72792053222656, "logps/rejected": -213.3939208984375, "loss": 0.89, "margin_dpo/margin_mean": 64.8343505859375, "margin_dpo/margin_std": 71.27854919433594, "step": 220 }, { "epoch": 0.3303964757709251, "fcm_dpo/beta": 0.012188979424536228, "fcm_dpo/delta": -0.16104714572429657, "fcm_dpo/margin": 81.43888854980469, "fcm_dpo/q_t": 0.30945101380348206, "grad_norm": 35.1022834777832, "learning_rate": 4.249525076191759e-07, "logits/chosen": -0.40194278955459595, "logits/rejected": -0.3876434564590454, "logps/chosen": -134.11956787109375, "logps/ref_chosen": -54.47245407104492, "logps/ref_rejected": -93.26266479492188, "logps/rejected": -254.3486785888672, "loss": 0.8406, "margin_dpo/margin_mean": 81.43888854980469, "margin_dpo/margin_std": 86.77934265136719, "step": 225 }, { "epoch": 0.3377386196769457, "fcm_dpo/beta": 0.01108027994632721, "fcm_dpo/delta": -0.06726758182048798, "fcm_dpo/margin": 82.1200942993164, "fcm_dpo/q_t": 0.3215538561344147, "grad_norm": 34.77726364135742, "learning_rate": 4.203117865141635e-07, "logits/chosen": -0.3820115923881531, "logits/rejected": -0.35805052518844604, "logps/chosen": -139.94625854492188, "logps/ref_chosen": -58.7701301574707, "logps/ref_rejected": -87.85963439941406, "logps/rejected": -251.1558380126953, "loss": 0.865, "margin_dpo/margin_mean": 82.1200942993164, "margin_dpo/margin_std": 89.67250061035156, "step": 230 }, { "epoch": 0.34508076358296624, "fcm_dpo/beta": 0.010919634252786636, "fcm_dpo/delta": 0.039671458303928375, "fcm_dpo/margin": 74.35450744628906, "fcm_dpo/q_t": 0.34256863594055176, "grad_norm": 32.15520095825195, "learning_rate": 4.1555887447288255e-07, "logits/chosen": -0.4005228877067566, "logits/rejected": -0.3855542838573456, "logps/chosen": -133.8794403076172, "logps/ref_chosen": -59.0481071472168, "logps/ref_rejected": -91.19654846191406, "logps/rejected": -240.3823699951172, "loss": 0.9162, "margin_dpo/margin_mean": 74.35450744628906, "margin_dpo/margin_std": 88.09162902832031, "step": 235 }, { "epoch": 0.3524229074889868, "fcm_dpo/beta": 0.010825184173882008, "fcm_dpo/delta": -0.018980540335178375, "fcm_dpo/margin": 79.97639465332031, "fcm_dpo/q_t": 0.3274964690208435, "grad_norm": 40.340721130371094, "learning_rate": 4.106969024216348e-07, "logits/chosen": -0.3556009829044342, "logits/rejected": -0.34894007444381714, "logps/chosen": -138.43865966796875, "logps/ref_chosen": -55.238983154296875, "logps/ref_rejected": -91.08428955078125, "logps/rejected": -254.26034545898438, "loss": 0.8677, "margin_dpo/margin_mean": 79.97639465332031, "margin_dpo/margin_std": 84.8770523071289, "step": 240 }, { "epoch": 0.35976505139500736, "fcm_dpo/beta": 0.010110095143318176, "fcm_dpo/delta": -0.10127156972885132, "fcm_dpo/margin": 92.84043884277344, "fcm_dpo/q_t": 0.3163699507713318, "grad_norm": 42.48567581176758, "learning_rate": 4.057290731287531e-07, "logits/chosen": -0.35475802421569824, "logits/rejected": -0.3244848847389221, "logps/chosen": -158.22030639648438, "logps/ref_chosen": -65.08844757080078, "logps/ref_rejected": -86.05777740478516, "logps/rejected": -272.03009033203125, "loss": 0.8333, "margin_dpo/margin_mean": 92.8404312133789, "margin_dpo/margin_std": 95.6347885131836, "step": 245 }, { "epoch": 0.3671071953010279, "fcm_dpo/beta": 0.00962417759001255, "fcm_dpo/delta": -0.020193621516227722, "fcm_dpo/margin": 90.06238555908203, "fcm_dpo/q_t": 0.3266645073890686, "grad_norm": 54.82107162475586, "learning_rate": 4.006586590948141e-07, "logits/chosen": -0.37928658723831177, "logits/rejected": -0.35434064269065857, "logps/chosen": -154.8458709716797, "logps/ref_chosen": -59.08491897583008, "logps/ref_rejected": -87.36727142333984, "logps/rejected": -273.19061279296875, "loss": 0.8678, "margin_dpo/margin_mean": 90.06238555908203, "margin_dpo/margin_std": 93.82579040527344, "step": 250 }, { "epoch": 0.3744493392070485, "fcm_dpo/beta": 0.009607134386897087, "fcm_dpo/delta": 0.0022653634659945965, "fcm_dpo/margin": 88.13050842285156, "fcm_dpo/q_t": 0.3329293131828308, "grad_norm": 58.291961669921875, "learning_rate": 3.954890003969163e-07, "logits/chosen": -0.33103686571121216, "logits/rejected": -0.3057294487953186, "logps/chosen": -170.76805114746094, "logps/ref_chosen": -61.85979461669922, "logps/ref_rejected": -88.32804107666016, "logps/rejected": -285.3667907714844, "loss": 0.9102, "margin_dpo/margin_mean": 88.13050842285156, "margin_dpo/margin_std": 103.13516998291016, "step": 255 }, { "epoch": 0.38179148311306904, "fcm_dpo/beta": 0.009414169006049633, "fcm_dpo/delta": -0.016010303050279617, "fcm_dpo/margin": 91.66337585449219, "fcm_dpo/q_t": 0.33001503348350525, "grad_norm": 39.26882553100586, "learning_rate": 3.9022350248844246e-07, "logits/chosen": -0.31099259853363037, "logits/rejected": -0.3064902424812317, "logps/chosen": -154.59693908691406, "logps/ref_chosen": -52.843467712402344, "logps/ref_rejected": -90.4744873046875, "logps/rejected": -283.89129638671875, "loss": 0.8885, "margin_dpo/margin_mean": 91.66337585449219, "margin_dpo/margin_std": 103.34080505371094, "step": 260 }, { "epoch": 0.3891336270190896, "fcm_dpo/beta": 0.009381098672747612, "fcm_dpo/delta": -0.015523952431976795, "fcm_dpo/margin": 92.02482604980469, "fcm_dpo/q_t": 0.3260762095451355, "grad_norm": 42.47758865356445, "learning_rate": 3.848656339557562e-07, "logits/chosen": -0.299502432346344, "logits/rejected": -0.27807632088661194, "logps/chosen": -151.35202026367188, "logps/ref_chosen": -59.35320281982422, "logps/ref_rejected": -90.73350524902344, "logps/rejected": -274.75714111328125, "loss": 0.8593, "margin_dpo/margin_mean": 92.02482604980469, "margin_dpo/margin_std": 95.39722442626953, "step": 265 }, { "epoch": 0.3964757709251101, "fcm_dpo/beta": 0.009062977507710457, "fcm_dpo/delta": -0.06297405064105988, "fcm_dpo/margin": 100.06938171386719, "fcm_dpo/q_t": 0.32098034024238586, "grad_norm": 51.16044616699219, "learning_rate": 3.794189242333106e-07, "logits/chosen": -0.23682311177253723, "logits/rejected": -0.21265558898448944, "logps/chosen": -184.3603057861328, "logps/ref_chosen": -66.30875396728516, "logps/ref_rejected": -95.4130630493164, "logps/rejected": -313.5339660644531, "loss": 0.8569, "margin_dpo/margin_mean": 100.06938171386719, "margin_dpo/margin_std": 107.111083984375, "step": 270 }, { "epoch": 0.40381791483113066, "fcm_dpo/beta": 0.008018719032406807, "fcm_dpo/delta": -0.12382125854492188, "fcm_dpo/margin": 119.62171936035156, "fcm_dpo/q_t": 0.3118273615837097, "grad_norm": 33.570701599121094, "learning_rate": 3.738869612786737e-07, "logits/chosen": -0.15879306197166443, "logits/rejected": -0.14661014080047607, "logps/chosen": -189.1214599609375, "logps/ref_chosen": -54.69990921020508, "logps/ref_rejected": -92.23838806152344, "logps/rejected": -346.2816467285156, "loss": 0.8306, "margin_dpo/margin_mean": 119.6217269897461, "margin_dpo/margin_std": 122.45941162109375, "step": 275 }, { "epoch": 0.4111600587371512, "fcm_dpo/beta": 0.007537333760410547, "fcm_dpo/delta": -0.07704190909862518, "fcm_dpo/margin": 122.15445709228516, "fcm_dpo/q_t": 0.3190959692001343, "grad_norm": 67.1026840209961, "learning_rate": 3.6827338920900253e-07, "logits/chosen": -0.15290072560310364, "logits/rejected": -0.14313052594661713, "logps/chosen": -187.32925415039062, "logps/ref_chosen": -54.64586639404297, "logps/ref_rejected": -88.19416809082031, "logps/rejected": -343.0320129394531, "loss": 0.8467, "margin_dpo/margin_mean": 122.15445709228516, "margin_dpo/margin_std": 127.57193756103516, "step": 280 }, { "epoch": 0.4185022026431718, "fcm_dpo/beta": 0.007103285752236843, "fcm_dpo/delta": -0.03963179141283035, "fcm_dpo/margin": 124.67244720458984, "fcm_dpo/q_t": 0.32029491662979126, "grad_norm": 36.61220169067383, "learning_rate": 3.625819059005228e-07, "logits/chosen": -0.09207823872566223, "logits/rejected": -0.07269109785556793, "logps/chosen": -234.484375, "logps/ref_chosen": -63.02496337890625, "logps/ref_rejected": -93.16323852539062, "logps/rejected": -389.29510498046875, "loss": 0.8457, "margin_dpo/margin_mean": 124.67244720458984, "margin_dpo/margin_std": 126.2774429321289, "step": 285 }, { "epoch": 0.42584434654919234, "fcm_dpo/beta": 0.006760433316230774, "fcm_dpo/delta": -0.0626429095864296, "fcm_dpo/margin": 134.1584930419922, "fcm_dpo/q_t": 0.3183870315551758, "grad_norm": 32.6433219909668, "learning_rate": 3.568162605525952e-07, "logits/chosen": -0.07117484509944916, "logits/rejected": -0.06140874698758125, "logps/chosen": -225.1572723388672, "logps/ref_chosen": -58.37105178833008, "logps/ref_rejected": -91.59428405761719, "logps/rejected": -392.53900146484375, "loss": 0.8386, "margin_dpo/margin_mean": 134.1584930419922, "margin_dpo/margin_std": 134.99864196777344, "step": 290 }, { "epoch": 0.4331864904552129, "fcm_dpo/beta": 0.006551184691488743, "fcm_dpo/delta": 0.024965018033981323, "fcm_dpo/margin": 126.14031982421875, "fcm_dpo/q_t": 0.3340519070625305, "grad_norm": 46.24834060668945, "learning_rate": 3.509802512179737e-07, "logits/chosen": -0.051733147352933884, "logits/rejected": -0.04382902756333351, "logps/chosen": -215.8157958984375, "logps/ref_chosen": -55.113426208496094, "logps/ref_rejected": -85.29283905029297, "logps/rejected": -372.1355285644531, "loss": 0.8958, "margin_dpo/margin_mean": 126.14030456542969, "margin_dpo/margin_std": 140.62646484375, "step": 295 }, { "epoch": 0.44052863436123346, "fcm_dpo/beta": 0.006881644017994404, "fcm_dpo/delta": -0.0010080442298203707, "fcm_dpo/margin": 123.4745101928711, "fcm_dpo/q_t": 0.3347456455230713, "grad_norm": 67.42388916015625, "learning_rate": 3.4507772230088147e-07, "logits/chosen": -0.06820065528154373, "logits/rejected": -0.0680488869547844, "logps/chosen": -241.1220703125, "logps/ref_chosen": -59.46582794189453, "logps/ref_rejected": -97.03690338134766, "logps/rejected": -402.16766357421875, "loss": 0.9105, "margin_dpo/margin_mean": 123.4745101928711, "margin_dpo/margin_std": 146.905029296875, "step": 300 }, { "epoch": 0.447870778267254, "fcm_dpo/beta": 0.006595266051590443, "fcm_dpo/delta": 0.0008342728251591325, "fcm_dpo/margin": 128.62730407714844, "fcm_dpo/q_t": 0.33005213737487793, "grad_norm": 42.539947509765625, "learning_rate": 3.391125620245535e-07, "logits/chosen": -0.09431194514036179, "logits/rejected": -0.08120522648096085, "logps/chosen": -225.65005493164062, "logps/ref_chosen": -62.78144454956055, "logps/ref_rejected": -91.95039367675781, "logps/rejected": -383.4462890625, "loss": 0.8706, "margin_dpo/margin_mean": 128.62728881835938, "margin_dpo/margin_std": 136.1685333251953, "step": 305 }, { "epoch": 0.4552129221732746, "fcm_dpo/beta": 0.006700504571199417, "fcm_dpo/delta": -0.0035556063521653414, "fcm_dpo/margin": 127.2578125, "fcm_dpo/q_t": 0.3245096802711487, "grad_norm": 40.4238395690918, "learning_rate": 3.3308869986991487e-07, "logits/chosen": -0.07853604853153229, "logits/rejected": -0.056885600090026855, "logps/chosen": -217.1204071044922, "logps/ref_chosen": -61.359039306640625, "logps/ref_rejected": -82.75496673583984, "logps/rejected": -365.7741394042969, "loss": 0.851, "margin_dpo/margin_mean": 127.2578125, "margin_dpo/margin_std": 124.93436431884766, "step": 310 }, { "epoch": 0.46255506607929514, "fcm_dpo/beta": 0.006514403969049454, "fcm_dpo/delta": 0.00254381587728858, "fcm_dpo/margin": 129.9599151611328, "fcm_dpo/q_t": 0.3323710560798645, "grad_norm": 71.91686248779297, "learning_rate": 3.270101039870797e-07, "logits/chosen": 0.050092458724975586, "logits/rejected": 0.06273595988750458, "logps/chosen": -250.1263885498047, "logps/ref_chosen": -51.77602005004883, "logps/ref_rejected": -84.58292388916016, "logps/rejected": -412.8932189941406, "loss": 0.8915, "margin_dpo/margin_mean": 129.9599151611328, "margin_dpo/margin_std": 146.8983612060547, "step": 315 }, { "epoch": 0.4698972099853157, "fcm_dpo/beta": 0.006153360940515995, "fcm_dpo/delta": -0.12349160015583038, "fcm_dpo/margin": 156.13824462890625, "fcm_dpo/q_t": 0.3038443326950073, "grad_norm": 37.25245666503906, "learning_rate": 3.208807785813777e-07, "logits/chosen": -0.020375465974211693, "logits/rejected": -0.011311917565762997, "logps/chosen": -244.4154815673828, "logps/ref_chosen": -56.777862548828125, "logps/ref_rejected": -99.26368713378906, "logps/rejected": -443.03955078125, "loss": 0.7846, "margin_dpo/margin_mean": 156.13824462890625, "margin_dpo/margin_std": 138.79611206054688, "step": 320 }, { "epoch": 0.47723935389133626, "fcm_dpo/beta": 0.00587537931278348, "fcm_dpo/delta": 0.004868075251579285, "fcm_dpo/margin": 143.66757202148438, "fcm_dpo/q_t": 0.3283257484436035, "grad_norm": 46.0535774230957, "learning_rate": 3.147047612756302e-07, "logits/chosen": 0.04561225324869156, "logits/rejected": 0.05908801406621933, "logps/chosen": -280.6909484863281, "logps/ref_chosen": -58.28468704223633, "logps/ref_rejected": -83.80326843261719, "logps/rejected": -449.8771057128906, "loss": 0.8735, "margin_dpo/margin_mean": 143.66757202148438, "margin_dpo/margin_std": 146.94345092773438, "step": 325 }, { "epoch": 0.4845814977973568, "fcm_dpo/beta": 0.0057464223355054855, "fcm_dpo/delta": -0.0528077557682991, "fcm_dpo/margin": 156.2254638671875, "fcm_dpo/q_t": 0.3170631527900696, "grad_norm": 50.687007904052734, "learning_rate": 3.084861204504122e-07, "logits/chosen": 0.018803680315613747, "logits/rejected": 0.03341676667332649, "logps/chosen": -278.55731201171875, "logps/ref_chosen": -62.75822067260742, "logps/ref_rejected": -94.04203033447266, "logps/rejected": -466.06658935546875, "loss": 0.8302, "margin_dpo/margin_mean": 156.2254638671875, "margin_dpo/margin_std": 149.75978088378906, "step": 330 }, { "epoch": 0.4919236417033774, "fcm_dpo/beta": 0.005659957882016897, "fcm_dpo/delta": 0.03993542492389679, "fcm_dpo/margin": 143.60537719726562, "fcm_dpo/q_t": 0.33370915055274963, "grad_norm": 52.71037673950195, "learning_rate": 3.022289525640531e-07, "logits/chosen": 0.07965292036533356, "logits/rejected": 0.0922585278749466, "logps/chosen": -285.34234619140625, "logps/ref_chosen": -58.59650421142578, "logps/ref_rejected": -88.69586944580078, "logps/rejected": -459.047119140625, "loss": 0.8834, "margin_dpo/margin_mean": 143.60537719726562, "margin_dpo/margin_std": 148.41580200195312, "step": 335 }, { "epoch": 0.49926578560939794, "fcm_dpo/beta": 0.005560899153351784, "fcm_dpo/delta": -0.058431338518857956, "fcm_dpo/margin": 162.19033813476562, "fcm_dpo/q_t": 0.3149014711380005, "grad_norm": 35.8148193359375, "learning_rate": 2.959373794541426e-07, "logits/chosen": 0.013386556878685951, "logits/rejected": 0.0291533712297678, "logps/chosen": -255.1841278076172, "logps/ref_chosen": -58.18162155151367, "logps/ref_rejected": -94.44358825683594, "logps/rejected": -453.63641357421875, "loss": 0.825, "margin_dpo/margin_mean": 162.1903533935547, "margin_dpo/margin_std": 153.74880981445312, "step": 340 }, { "epoch": 0.5066079295154186, "fcm_dpo/beta": 0.0053556510247290134, "fcm_dpo/delta": -0.04881957173347473, "fcm_dpo/margin": 166.8225860595703, "fcm_dpo/q_t": 0.31634849309921265, "grad_norm": 50.1009521484375, "learning_rate": 2.896155456223163e-07, "logits/chosen": -0.0006014652317389846, "logits/rejected": 0.011017178185284138, "logps/chosen": -275.6701354980469, "logps/ref_chosen": -57.9904899597168, "logps/ref_rejected": -99.11092376708984, "logps/rejected": -483.6131896972656, "loss": 0.8223, "margin_dpo/margin_mean": 166.8225860595703, "margin_dpo/margin_std": 155.08956909179688, "step": 345 }, { "epoch": 0.5139500734214391, "fcm_dpo/beta": 0.005294440779834986, "fcm_dpo/delta": 0.02646171674132347, "fcm_dpo/margin": 155.80526733398438, "fcm_dpo/q_t": 0.3322368264198303, "grad_norm": 61.714805603027344, "learning_rate": 2.8326761550411346e-07, "logits/chosen": 0.13601693511009216, "logits/rejected": 0.14563313126564026, "logps/chosen": -326.2274169921875, "logps/ref_chosen": -58.29923629760742, "logps/ref_rejected": -89.25711822509766, "logps/rejected": -512.9906005859375, "loss": 0.8883, "margin_dpo/margin_mean": 155.80526733398438, "margin_dpo/margin_std": 168.00064086914062, "step": 350 }, { "epoch": 0.5212922173274597, "fcm_dpo/beta": 0.005374093074351549, "fcm_dpo/delta": -0.00456079700961709, "fcm_dpo/margin": 158.74806213378906, "fcm_dpo/q_t": 0.325572669506073, "grad_norm": 50.927425384521484, "learning_rate": 2.7689777072570284e-07, "logits/chosen": 0.049882251769304276, "logits/rejected": 0.07219198346138, "logps/chosen": -304.60711669921875, "logps/ref_chosen": -60.788482666015625, "logps/ref_rejected": -85.94129943847656, "logps/rejected": -488.50799560546875, "loss": 0.8631, "margin_dpo/margin_mean": 158.74806213378906, "margin_dpo/margin_std": 160.3525848388672, "step": 355 }, { "epoch": 0.5286343612334802, "fcm_dpo/beta": 0.005639675538986921, "fcm_dpo/delta": 0.06519472599029541, "fcm_dpo/margin": 139.63821411132812, "fcm_dpo/q_t": 0.33998966217041016, "grad_norm": 103.65924835205078, "learning_rate": 2.7051020734928443e-07, "logits/chosen": 0.07553316652774811, "logits/rejected": 0.0942920446395874, "logps/chosen": -303.3082580566406, "logps/ref_chosen": -57.6871337890625, "logps/ref_rejected": -80.62527465820312, "logps/rejected": -465.8846130371094, "loss": 0.9029, "margin_dpo/margin_mean": 139.63821411132812, "margin_dpo/margin_std": 152.2101593017578, "step": 360 }, { "epoch": 0.5359765051395007, "fcm_dpo/beta": 0.005498864687979221, "fcm_dpo/delta": -0.058943700045347214, "fcm_dpo/margin": 164.1685028076172, "fcm_dpo/q_t": 0.3161132335662842, "grad_norm": 103.95121002197266, "learning_rate": 2.641091331089811e-07, "logits/chosen": 0.13508400321006775, "logits/rejected": 0.1376177817583084, "logps/chosen": -293.01922607421875, "logps/ref_chosen": -51.490867614746094, "logps/ref_rejected": -91.02871704101562, "logps/rejected": -496.72552490234375, "loss": 0.8274, "margin_dpo/margin_mean": 164.1685028076172, "margin_dpo/margin_std": 157.7042236328125, "step": 365 }, { "epoch": 0.5433186490455213, "fcm_dpo/beta": 0.005576250609010458, "fcm_dpo/delta": 0.08467354625463486, "fcm_dpo/margin": 138.50732421875, "fcm_dpo/q_t": 0.3373372554779053, "grad_norm": 51.74324035644531, "learning_rate": 2.5769876463904263e-07, "logits/chosen": 0.10848675668239594, "logits/rejected": 0.11948434263467789, "logps/chosen": -298.34576416015625, "logps/ref_chosen": -58.113502502441406, "logps/ref_rejected": -89.43451690673828, "logps/rejected": -468.1741638183594, "loss": 0.8812, "margin_dpo/margin_mean": 138.50733947753906, "margin_dpo/margin_std": 134.93521118164062, "step": 370 }, { "epoch": 0.5506607929515418, "fcm_dpo/beta": 0.0056075444445014, "fcm_dpo/delta": -0.06286319345235825, "fcm_dpo/margin": 161.43272399902344, "fcm_dpo/q_t": 0.3194437623023987, "grad_norm": 49.77079772949219, "learning_rate": 2.512833246961859e-07, "logits/chosen": 0.1206691637635231, "logits/rejected": 0.14743848145008087, "logps/chosen": -326.29534912109375, "logps/ref_chosen": -65.23600769042969, "logps/ref_rejected": -89.24995422363281, "logps/rejected": -511.74200439453125, "loss": 0.8508, "margin_dpo/margin_mean": 161.43270874023438, "margin_dpo/margin_std": 164.4888458251953, "step": 375 }, { "epoch": 0.5580029368575624, "fcm_dpo/beta": 0.005182487424463034, "fcm_dpo/delta": -0.11963502317667007, "fcm_dpo/margin": 185.00277709960938, "fcm_dpo/q_t": 0.3068377375602722, "grad_norm": 47.95132064819336, "learning_rate": 2.4486703937790243e-07, "logits/chosen": 0.1993691623210907, "logits/rejected": 0.2027546912431717, "logps/chosen": -336.24005126953125, "logps/ref_chosen": -53.33893966674805, "logps/ref_rejected": -102.15375518798828, "logps/rejected": -570.0576171875, "loss": 0.8158, "margin_dpo/margin_mean": 185.00277709960938, "margin_dpo/margin_std": 178.25515747070312, "step": 380 }, { "epoch": 0.5653450807635829, "fcm_dpo/beta": 0.0049522449262440205, "fcm_dpo/delta": 0.019767077639698982, "fcm_dpo/margin": 167.88037109375, "fcm_dpo/q_t": 0.3291183114051819, "grad_norm": 34.479740142822266, "learning_rate": 2.3845413533856514e-07, "logits/chosen": 0.18332220613956451, "logits/rejected": 0.19640536606311798, "logps/chosen": -353.31121826171875, "logps/ref_chosen": -58.36262130737305, "logps/ref_rejected": -89.44685363769531, "logps/rejected": -552.2758178710938, "loss": 0.8731, "margin_dpo/margin_mean": 167.88037109375, "margin_dpo/margin_std": 172.85272216796875, "step": 385 }, { "epoch": 0.5726872246696035, "fcm_dpo/beta": 0.004950010217726231, "fcm_dpo/delta": -0.0016832177061587572, "fcm_dpo/margin": 171.85813903808594, "fcm_dpo/q_t": 0.32719507813453674, "grad_norm": 46.603187561035156, "learning_rate": 2.320488370051681e-07, "logits/chosen": 0.12718243896961212, "logits/rejected": 0.1411055475473404, "logps/chosen": -333.3777160644531, "logps/ref_chosen": -56.380653381347656, "logps/ref_rejected": -90.47447204589844, "logps/rejected": -539.3297119140625, "loss": 0.8665, "margin_dpo/margin_mean": 171.85813903808594, "margin_dpo/margin_std": 178.92532348632812, "step": 390 }, { "epoch": 0.580029368575624, "fcm_dpo/beta": 0.005209635477513075, "fcm_dpo/delta": 0.042330484837293625, "fcm_dpo/margin": 155.28634643554688, "fcm_dpo/q_t": 0.33868879079818726, "grad_norm": 44.36903381347656, "learning_rate": 2.2565536379453404e-07, "logits/chosen": 0.03185752406716347, "logits/rejected": 0.039340805262327194, "logps/chosen": -274.38555908203125, "logps/ref_chosen": -55.95304489135742, "logps/ref_rejected": -87.13162994384766, "logps/rejected": -460.8504943847656, "loss": 0.9001, "margin_dpo/margin_mean": 155.28634643554688, "margin_dpo/margin_std": 174.56260681152344, "step": 395 }, { "epoch": 0.5873715124816447, "fcm_dpo/beta": 0.005051156505942345, "fcm_dpo/delta": -0.08489851653575897, "fcm_dpo/margin": 183.75982666015625, "fcm_dpo/q_t": 0.31105726957321167, "grad_norm": 40.749473571777344, "learning_rate": 2.192779273338215e-07, "logits/chosen": 0.04275448992848396, "logits/rejected": 0.0653674453496933, "logps/chosen": -283.33929443359375, "logps/ref_chosen": -64.59160614013672, "logps/ref_rejected": -96.700927734375, "logps/rejected": -499.20843505859375, "loss": 0.8159, "margin_dpo/margin_mean": 183.7598419189453, "margin_dpo/margin_std": 173.43460083007812, "step": 400 }, { "epoch": 0.5873715124816447, "eval_fcm_dpo/beta": 0.0069984132423996925, "eval_fcm_dpo/delta": 0.07261277735233307, "eval_fcm_dpo/margin": 111.74861907958984, "eval_fcm_dpo/q_t": 0.36032289266586304, "eval_logits/chosen": 0.07495035231113434, "eval_logits/rejected": 0.09724592417478561, "eval_logps/chosen": -360.3431091308594, "eval_logps/ref_chosen": -79.05104064941406, "eval_logps/ref_rejected": -86.79793548583984, "eval_logps/rejected": -479.838623046875, "eval_loss": 0.5361973643302917, "eval_margin_dpo/margin_mean": 111.74861907958984, "eval_margin_dpo/margin_std": 182.2795867919922, "eval_runtime": 39.5126, "eval_samples_per_second": 59.196, "eval_steps_per_second": 1.873, "step": 400 }, { "epoch": 0.5947136563876652, "fcm_dpo/beta": 0.006445285864174366, "fcm_dpo/delta": -0.3454572856426239, "fcm_dpo/margin": 178.49400329589844, "fcm_dpo/q_t": 0.29232606291770935, "grad_norm": 75.37237548828125, "learning_rate": 2.129207286861638e-07, "logits/chosen": 0.15050789713859558, "logits/rejected": 0.16446948051452637, "logps/chosen": -286.4220886230469, "logps/ref_chosen": -53.61777877807617, "logps/ref_rejected": -81.28938293457031, "logps/rejected": -492.58770751953125, "loss": 0.8372, "margin_dpo/margin_mean": 178.4940185546875, "margin_dpo/margin_std": 191.11578369140625, "step": 405 }, { "epoch": 0.6020558002936858, "fcm_dpo/beta": 0.005619804374873638, "fcm_dpo/delta": -0.025178026407957077, "fcm_dpo/margin": 155.10745239257812, "fcm_dpo/q_t": 0.32710129022598267, "grad_norm": 44.97209167480469, "learning_rate": 2.065879555832674e-07, "logits/chosen": 0.16110701858997345, "logits/rejected": 0.18340006470680237, "logps/chosen": -324.66778564453125, "logps/ref_chosen": -58.9287223815918, "logps/ref_rejected": -85.55818176269531, "logps/rejected": -506.4046936035156, "loss": 0.8828, "margin_dpo/margin_mean": 155.10745239257812, "margin_dpo/margin_std": 172.00360107421875, "step": 410 }, { "epoch": 0.6093979441997063, "fcm_dpo/beta": 0.00546681135892868, "fcm_dpo/delta": -0.03474488854408264, "fcm_dpo/margin": 160.98959350585938, "fcm_dpo/q_t": 0.3214709758758545, "grad_norm": 46.68992614746094, "learning_rate": 2.002837796667909e-07, "logits/chosen": 0.17981475591659546, "logits/rejected": 0.1913680136203766, "logps/chosen": -350.0954895019531, "logps/ref_chosen": -58.45662307739258, "logps/ref_rejected": -93.67063903808594, "logps/rejected": -546.299072265625, "loss": 0.848, "margin_dpo/margin_mean": 160.98959350585938, "margin_dpo/margin_std": 161.0213623046875, "step": 415 }, { "epoch": 0.6167400881057269, "fcm_dpo/beta": 0.005015296395868063, "fcm_dpo/delta": -0.08632484823465347, "fcm_dpo/margin": 184.5193634033203, "fcm_dpo/q_t": 0.3122255802154541, "grad_norm": 47.35343933105469, "learning_rate": 1.9401235374032425e-07, "logits/chosen": 0.1722276657819748, "logits/rejected": 0.20106978714466095, "logps/chosen": -375.71832275390625, "logps/ref_chosen": -64.2349853515625, "logps/ref_rejected": -84.86299133300781, "logps/rejected": -580.8656616210938, "loss": 0.8229, "margin_dpo/margin_mean": 184.5193634033203, "margin_dpo/margin_std": 177.80752563476562, "step": 420 }, { "epoch": 0.6240822320117474, "fcm_dpo/beta": 0.005182936787605286, "fcm_dpo/delta": 0.06720009446144104, "fcm_dpo/margin": 151.88052368164062, "fcm_dpo/q_t": 0.33798202872276306, "grad_norm": 35.061790466308594, "learning_rate": 1.8777780903377732e-07, "logits/chosen": 0.07602418959140778, "logits/rejected": 0.090725377202034, "logps/chosen": -300.1365661621094, "logps/ref_chosen": -56.054161071777344, "logps/ref_rejected": -85.17119598388672, "logps/rejected": -481.1341247558594, "loss": 0.9033, "margin_dpo/margin_mean": 151.88052368164062, "margin_dpo/margin_std": 166.2532958984375, "step": 425 }, { "epoch": 0.631424375917768, "fcm_dpo/beta": 0.005413433536887169, "fcm_dpo/delta": 0.0383293442428112, "fcm_dpo/margin": 150.35128784179688, "fcm_dpo/q_t": 0.3347117006778717, "grad_norm": 38.60581970214844, "learning_rate": 1.8158425248197928e-07, "logits/chosen": 0.00837210938334465, "logits/rejected": 0.03842420130968094, "logps/chosen": -316.62884521484375, "logps/ref_chosen": -69.24568176269531, "logps/ref_rejected": -91.8664321899414, "logps/rejected": -489.60089111328125, "loss": 0.8974, "margin_dpo/margin_mean": 150.35128784179688, "margin_dpo/margin_std": 164.39129638671875, "step": 430 }, { "epoch": 0.6387665198237885, "fcm_dpo/beta": 0.0052519310265779495, "fcm_dpo/delta": -0.04517149180173874, "fcm_dpo/margin": 169.1701202392578, "fcm_dpo/q_t": 0.3183755874633789, "grad_norm": 28.78717613220215, "learning_rate": 1.7543576401928218e-07, "logits/chosen": 0.07158392667770386, "logits/rejected": 0.0934370905160904, "logps/chosen": -321.0721130371094, "logps/ref_chosen": -60.03449630737305, "logps/ref_rejected": -90.6872329711914, "logps/rejected": -520.89501953125, "loss": 0.8427, "margin_dpo/margin_mean": 169.1701202392578, "margin_dpo/margin_std": 165.15988159179688, "step": 435 }, { "epoch": 0.6461086637298091, "fcm_dpo/beta": 0.00523256603628397, "fcm_dpo/delta": 0.028819337487220764, "fcm_dpo/margin": 157.2620086669922, "fcm_dpo/q_t": 0.33098357915878296, "grad_norm": 44.71712875366211, "learning_rate": 1.6933639389195134e-07, "logits/chosen": 0.05249354988336563, "logits/rejected": 0.08613702654838562, "logps/chosen": -312.62689208984375, "logps/ref_chosen": -65.50349426269531, "logps/ref_rejected": -85.66627502441406, "logps/rejected": -490.05169677734375, "loss": 0.867, "margin_dpo/margin_mean": 157.2620086669922, "margin_dpo/margin_std": 156.13658142089844, "step": 440 }, { "epoch": 0.6534508076358296, "fcm_dpo/beta": 0.00520352041348815, "fcm_dpo/delta": -0.07986284047365189, "fcm_dpo/margin": 177.13780212402344, "fcm_dpo/q_t": 0.3184102475643158, "grad_norm": 36.37860107421875, "learning_rate": 1.6329015999011182e-07, "logits/chosen": 0.12497935444116592, "logits/rejected": 0.14361031353473663, "logps/chosen": -362.954345703125, "logps/ref_chosen": -60.72443389892578, "logps/ref_rejected": -89.9255142211914, "logps/rejected": -569.2933349609375, "loss": 0.8507, "margin_dpo/margin_mean": 177.13780212402344, "margin_dpo/margin_std": 186.6912384033203, "step": 445 }, { "epoch": 0.6607929515418502, "fcm_dpo/beta": 0.004721947945654392, "fcm_dpo/delta": -0.10506816953420639, "fcm_dpo/margin": 200.17857360839844, "fcm_dpo/q_t": 0.3099059462547302, "grad_norm": 43.16559982299805, "learning_rate": 1.573010452010098e-07, "logits/chosen": 0.18583309650421143, "logits/rejected": 0.19731836020946503, "logps/chosen": -397.82037353515625, "logps/ref_chosen": -59.96248245239258, "logps/ref_rejected": -98.71509552001953, "logps/rejected": -636.7515869140625, "loss": 0.8135, "margin_dpo/margin_mean": 200.17857360839844, "margin_dpo/margin_std": 195.84307861328125, "step": 450 }, { "epoch": 0.6681350954478708, "fcm_dpo/beta": 0.004480073228478432, "fcm_dpo/delta": -0.03945715352892876, "fcm_dpo/margin": 197.69662475585938, "fcm_dpo/q_t": 0.31971079111099243, "grad_norm": 35.01383972167969, "learning_rate": 1.5137299478533064e-07, "logits/chosen": 0.23781771957874298, "logits/rejected": 0.259443998336792, "logps/chosen": -383.10321044921875, "logps/ref_chosen": -54.48131561279297, "logps/ref_rejected": -90.6321029663086, "logps/rejected": -616.95068359375, "loss": 0.8442, "margin_dpo/margin_mean": 197.69662475585938, "margin_dpo/margin_std": 198.9961395263672, "step": 455 }, { "epoch": 0.6754772393538914, "fcm_dpo/beta": 0.004186900332570076, "fcm_dpo/delta": -0.045056529343128204, "fcm_dpo/margin": 212.65853881835938, "fcm_dpo/q_t": 0.31360098719596863, "grad_norm": 38.242950439453125, "learning_rate": 1.4550991377830423e-07, "logits/chosen": 0.1622854471206665, "logits/rejected": 0.17482970654964447, "logps/chosen": -349.6832580566406, "logps/ref_chosen": -52.97611618041992, "logps/ref_rejected": -95.65971374511719, "logps/rejected": -605.0254516601562, "loss": 0.8071, "margin_dpo/margin_mean": 212.65853881835938, "margin_dpo/margin_std": 186.55628967285156, "step": 460 }, { "epoch": 0.6828193832599119, "fcm_dpo/beta": 0.00432776240631938, "fcm_dpo/delta": 0.006079354323446751, "fcm_dpo/margin": 194.74021911621094, "fcm_dpo/q_t": 0.3281356990337372, "grad_norm": 34.140296936035156, "learning_rate": 1.3971566441730714e-07, "logits/chosen": 0.18998193740844727, "logits/rejected": 0.19938889145851135, "logps/chosen": -407.28875732421875, "logps/ref_chosen": -58.2827033996582, "logps/ref_rejected": -94.15567779541016, "logps/rejected": -637.9019775390625, "loss": 0.8641, "margin_dpo/margin_mean": 194.74020385742188, "margin_dpo/margin_std": 195.8119659423828, "step": 465 }, { "epoch": 0.6901615271659325, "fcm_dpo/beta": 0.004061209969222546, "fcm_dpo/delta": -0.04996412992477417, "fcm_dpo/margin": 220.33602905273438, "fcm_dpo/q_t": 0.3175013065338135, "grad_norm": 49.6895866394043, "learning_rate": 1.339940635976592e-07, "logits/chosen": 0.18324866890907288, "logits/rejected": 0.2053983509540558, "logps/chosen": -454.7452087402344, "logps/ref_chosen": -62.69774627685547, "logps/ref_rejected": -96.33873748779297, "logps/rejected": -708.7222900390625, "loss": 0.8415, "margin_dpo/margin_mean": 220.33602905273438, "margin_dpo/margin_std": 218.0626220703125, "step": 470 }, { "epoch": 0.697503671071953, "fcm_dpo/beta": 0.004032416269183159, "fcm_dpo/delta": 0.020370058715343475, "fcm_dpo/margin": 205.8577423095703, "fcm_dpo/q_t": 0.33193299174308777, "grad_norm": 61.28792190551758, "learning_rate": 1.2834888035828596e-07, "logits/chosen": 0.18428394198417664, "logits/rejected": 0.20524394512176514, "logps/chosen": -447.81591796875, "logps/ref_chosen": -61.12194061279297, "logps/ref_rejected": -92.42192077636719, "logps/rejected": -684.9736328125, "loss": 0.89, "margin_dpo/margin_mean": 205.8577423095703, "margin_dpo/margin_std": 225.5378875732422, "step": 475 }, { "epoch": 0.7048458149779736, "fcm_dpo/beta": 0.004136770032346249, "fcm_dpo/delta": 0.05575358867645264, "fcm_dpo/margin": 192.84674072265625, "fcm_dpo/q_t": 0.3363509476184845, "grad_norm": 75.26780700683594, "learning_rate": 1.227838333989088e-07, "logits/chosen": 0.1994629055261612, "logits/rejected": 0.222102090716362, "logps/chosen": -393.8836364746094, "logps/ref_chosen": -53.550628662109375, "logps/ref_rejected": -82.65167999267578, "logps/rejected": -615.8314208984375, "loss": 0.8956, "margin_dpo/margin_mean": 192.84674072265625, "margin_dpo/margin_std": 205.8238525390625, "step": 480 }, { "epoch": 0.7121879588839941, "fcm_dpo/beta": 0.004281387198716402, "fcm_dpo/delta": -0.022714762017130852, "fcm_dpo/margin": 203.22549438476562, "fcm_dpo/q_t": 0.32248061895370483, "grad_norm": 33.045082092285156, "learning_rate": 1.1730258863039347e-07, "logits/chosen": 0.14324723184108734, "logits/rejected": 0.16249491274356842, "logps/chosen": -369.179931640625, "logps/ref_chosen": -60.76704788208008, "logps/ref_rejected": -92.70649719238281, "logps/rejected": -604.3448486328125, "loss": 0.849, "margin_dpo/margin_mean": 203.2254638671875, "margin_dpo/margin_std": 202.1529541015625, "step": 485 }, { "epoch": 0.7195301027900147, "fcm_dpo/beta": 0.004233093000948429, "fcm_dpo/delta": 0.0409715473651886, "fcm_dpo/margin": 191.93649291992188, "fcm_dpo/q_t": 0.33092981576919556, "grad_norm": 38.83848571777344, "learning_rate": 1.1190875675987355e-07, "logits/chosen": 0.09927807748317719, "logits/rejected": 0.11963877826929092, "logps/chosen": -345.2209777832031, "logps/ref_chosen": -53.78142547607422, "logps/ref_rejected": -91.67438507080078, "logps/rejected": -575.0504150390625, "loss": 0.8672, "margin_dpo/margin_mean": 191.93650817871094, "margin_dpo/margin_std": 187.42601013183594, "step": 490 }, { "epoch": 0.7268722466960352, "fcm_dpo/beta": 0.004637162666767836, "fcm_dpo/delta": 0.03998362645506859, "fcm_dpo/margin": 174.73489379882812, "fcm_dpo/q_t": 0.33525848388671875, "grad_norm": 36.38589096069336, "learning_rate": 1.0660589091223854e-07, "logits/chosen": 0.17240563035011292, "logits/rejected": 0.19218161702156067, "logps/chosen": -392.77386474609375, "logps/ref_chosen": -58.9004020690918, "logps/ref_rejected": -82.15424346923828, "logps/rejected": -590.7625732421875, "loss": 0.905, "margin_dpo/margin_mean": 174.73489379882812, "margin_dpo/margin_std": 193.69952392578125, "step": 495 }, { "epoch": 0.7342143906020558, "fcm_dpo/beta": 0.004700102843344212, "fcm_dpo/delta": 0.027824098244309425, "fcm_dpo/margin": 175.03207397460938, "fcm_dpo/q_t": 0.33339887857437134, "grad_norm": 64.69699096679688, "learning_rate": 1.0139748428955333e-07, "logits/chosen": 0.20382651686668396, "logits/rejected": 0.22221975028514862, "logps/chosen": -471.3478088378906, "logps/ref_chosen": -62.13483810424805, "logps/ref_rejected": -87.91773223876953, "logps/rejected": -672.1627197265625, "loss": 0.8955, "margin_dpo/margin_mean": 175.03207397460938, "margin_dpo/margin_std": 190.1815948486328, "step": 500 }, { "epoch": 0.7415565345080763, "fcm_dpo/beta": 0.004485036246478558, "fcm_dpo/delta": -0.030254876241087914, "fcm_dpo/margin": 195.27134704589844, "fcm_dpo/q_t": 0.3203888535499573, "grad_norm": 84.53753662109375, "learning_rate": 9.628696786995188e-08, "logits/chosen": 0.2238425463438034, "logits/rejected": 0.2508949339389801, "logps/chosen": -467.5668029785156, "logps/ref_chosen": -62.631813049316406, "logps/ref_rejected": -87.59168243408203, "logps/rejected": -687.7979736328125, "loss": 0.8415, "margin_dpo/margin_mean": 195.2713623046875, "margin_dpo/margin_std": 185.7364959716797, "step": 505 }, { "epoch": 0.748898678414097, "fcm_dpo/beta": 0.004516937769949436, "fcm_dpo/delta": -0.005912508815526962, "fcm_dpo/margin": 189.2612762451172, "fcm_dpo/q_t": 0.32770413160324097, "grad_norm": 47.08674240112305, "learning_rate": 9.127770814751932e-08, "logits/chosen": 0.21628710627555847, "logits/rejected": 0.23013325035572052, "logps/chosen": -452.9044494628906, "logps/ref_chosen": -60.552574157714844, "logps/ref_rejected": -91.0874252319336, "logps/rejected": -672.7006225585938, "loss": 0.8647, "margin_dpo/margin_mean": 189.2612762451172, "margin_dpo/margin_std": 197.85511779785156, "step": 510 }, { "epoch": 0.7562408223201175, "fcm_dpo/beta": 0.004476507194340229, "fcm_dpo/delta": 0.04546705633401871, "fcm_dpo/margin": 180.66757202148438, "fcm_dpo/q_t": 0.3367224335670471, "grad_norm": 41.41278076171875, "learning_rate": 8.637300491465272e-08, "logits/chosen": 0.23490139842033386, "logits/rejected": 0.25515562295913696, "logps/chosen": -423.93670654296875, "logps/ref_chosen": -60.9382438659668, "logps/ref_rejected": -83.36767578125, "logps/rejected": -627.0337524414062, "loss": 0.9098, "margin_dpo/margin_mean": 180.66757202148438, "margin_dpo/margin_std": 204.07958984375, "step": 515 }, { "epoch": 0.7635829662261381, "fcm_dpo/beta": 0.0045339628122746944, "fcm_dpo/delta": -0.015271050855517387, "fcm_dpo/margin": 190.21194458007812, "fcm_dpo/q_t": 0.32560357451438904, "grad_norm": 33.79608917236328, "learning_rate": 8.15760890883607e-08, "logits/chosen": 0.1384505331516266, "logits/rejected": 0.14931073784828186, "logps/chosen": -399.65496826171875, "logps/ref_chosen": -65.47642517089844, "logps/ref_rejected": -98.10872650146484, "logps/rejected": -622.4992065429688, "loss": 0.8604, "margin_dpo/margin_mean": 190.21194458007812, "margin_dpo/margin_std": 195.6290740966797, "step": 520 }, { "epoch": 0.7709251101321586, "fcm_dpo/beta": 0.0047523765824735165, "fcm_dpo/delta": 0.011808094568550587, "fcm_dpo/margin": 176.2508087158203, "fcm_dpo/q_t": 0.3255954086780548, "grad_norm": 39.144676208496094, "learning_rate": 7.689012058193384e-08, "logits/chosen": 0.17502716183662415, "logits/rejected": 0.1978307068347931, "logps/chosen": -376.6075134277344, "logps/ref_chosen": -59.072021484375, "logps/ref_rejected": -82.8821792602539, "logps/rejected": -576.66845703125, "loss": 0.8642, "margin_dpo/margin_mean": 176.2508087158203, "margin_dpo/margin_std": 174.07464599609375, "step": 525 }, { "epoch": 0.7782672540381792, "fcm_dpo/beta": 0.0046645584516227245, "fcm_dpo/delta": -0.0016397640574723482, "fcm_dpo/margin": 182.26202392578125, "fcm_dpo/q_t": 0.3263426423072815, "grad_norm": 32.63557434082031, "learning_rate": 7.231818622338822e-08, "logits/chosen": 0.18306271731853485, "logits/rejected": 0.20520134270191193, "logps/chosen": -398.6494445800781, "logps/ref_chosen": -61.11234664916992, "logps/ref_rejected": -87.01112365722656, "logps/rejected": -606.8101806640625, "loss": 0.8608, "margin_dpo/margin_mean": 182.26202392578125, "margin_dpo/margin_std": 181.18548583984375, "step": 530 }, { "epoch": 0.7856093979441997, "fcm_dpo/beta": 0.004603945650160313, "fcm_dpo/delta": -0.03259692341089249, "fcm_dpo/margin": 190.7316436767578, "fcm_dpo/q_t": 0.32039114832878113, "grad_norm": 34.20729064941406, "learning_rate": 6.786329772205246e-08, "logits/chosen": 0.19900277256965637, "logits/rejected": 0.2242613583803177, "logps/chosen": -409.9130859375, "logps/ref_chosen": -60.96736526489258, "logps/ref_rejected": -81.727294921875, "logps/rejected": -621.4046630859375, "loss": 0.8464, "margin_dpo/margin_mean": 190.73162841796875, "margin_dpo/margin_std": 186.18397521972656, "step": 535 }, { "epoch": 0.7929515418502202, "fcm_dpo/beta": 0.0043141767382621765, "fcm_dpo/delta": -0.06867679208517075, "fcm_dpo/margin": 211.166015625, "fcm_dpo/q_t": 0.3138847351074219, "grad_norm": 36.321292877197266, "learning_rate": 6.352838968463919e-08, "logits/chosen": 0.1988883763551712, "logits/rejected": 0.2228163480758667, "logps/chosen": -403.5604248046875, "logps/ref_chosen": -58.64385986328125, "logps/ref_rejected": -90.40965270996094, "logps/rejected": -646.4922485351562, "loss": 0.8217, "margin_dpo/margin_mean": 211.166015625, "margin_dpo/margin_std": 199.3925323486328, "step": 540 }, { "epoch": 0.8002936857562408, "fcm_dpo/beta": 0.004401062615215778, "fcm_dpo/delta": 0.036251001060009, "fcm_dpo/margin": 185.14137268066406, "fcm_dpo/q_t": 0.3344213664531708, "grad_norm": 47.214256286621094, "learning_rate": 5.9316317682106294e-08, "logits/chosen": 0.1989498883485794, "logits/rejected": 0.21334727108478546, "logps/chosen": -444.91400146484375, "logps/ref_chosen": -64.73474884033203, "logps/ref_rejected": -100.44208526611328, "logps/rejected": -665.7626953125, "loss": 0.8888, "margin_dpo/margin_mean": 185.14137268066406, "margin_dpo/margin_std": 199.3173828125, "step": 545 }, { "epoch": 0.8076358296622613, "fcm_dpo/beta": 0.0044622840359807014, "fcm_dpo/delta": 0.047435857355594635, "fcm_dpo/margin": 180.65365600585938, "fcm_dpo/q_t": 0.3330041766166687, "grad_norm": 44.434165954589844, "learning_rate": 5.5229856368582376e-08, "logits/chosen": 0.20213136076927185, "logits/rejected": 0.22051355242729187, "logps/chosen": -433.3397521972656, "logps/ref_chosen": -59.13951873779297, "logps/ref_rejected": -87.15635681152344, "logps/rejected": -642.0101928710938, "loss": 0.8799, "margin_dpo/margin_mean": 180.65365600585938, "margin_dpo/margin_std": 186.55726623535156, "step": 550 }, { "epoch": 0.8149779735682819, "fcm_dpo/beta": 0.004191819578409195, "fcm_dpo/delta": -0.12720482051372528, "fcm_dpo/margin": 229.26779174804688, "fcm_dpo/q_t": 0.3060959577560425, "grad_norm": 43.85456848144531, "learning_rate": 5.127169765359515e-08, "logits/chosen": 0.1819799244403839, "logits/rejected": 0.20495197176933289, "logps/chosen": -431.9147033691406, "logps/ref_chosen": -62.1995849609375, "logps/ref_rejected": -102.51883697509766, "logps/rejected": -701.501708984375, "loss": 0.8051, "margin_dpo/margin_mean": 229.26779174804688, "margin_dpo/margin_std": 213.53018188476562, "step": 555 }, { "epoch": 0.8223201174743024, "fcm_dpo/beta": 0.004056096542626619, "fcm_dpo/delta": 0.024886898696422577, "fcm_dpo/margin": 203.6185760498047, "fcm_dpo/q_t": 0.33151647448539734, "grad_norm": 43.7994270324707, "learning_rate": 4.7444448928806615e-08, "logits/chosen": 0.23396944999694824, "logits/rejected": 0.26040390133857727, "logps/chosen": -446.46905517578125, "logps/ref_chosen": -61.541908264160156, "logps/ref_rejected": -97.37491607666016, "logps/rejected": -685.920654296875, "loss": 0.8683, "margin_dpo/margin_mean": 203.6185760498047, "margin_dpo/margin_std": 203.64950561523438, "step": 560 }, { "epoch": 0.8296622613803231, "fcm_dpo/beta": 0.0042213378474116325, "fcm_dpo/delta": 0.0051836492493748665, "fcm_dpo/margin": 199.88539123535156, "fcm_dpo/q_t": 0.3270900845527649, "grad_norm": 34.782867431640625, "learning_rate": 4.375063135042445e-08, "logits/chosen": 0.271693617105484, "logits/rejected": 0.29448002576828003, "logps/chosen": -460.4130859375, "logps/ref_chosen": -62.85475540161133, "logps/ref_rejected": -93.8392105102539, "logps/rejected": -691.2828979492188, "loss": 0.8593, "margin_dpo/margin_mean": 199.88539123535156, "margin_dpo/margin_std": 199.06076049804688, "step": 565 }, { "epoch": 0.8370044052863436, "fcm_dpo/beta": 0.004123511724174023, "fcm_dpo/delta": -0.02085905149579048, "fcm_dpo/margin": 210.5379180908203, "fcm_dpo/q_t": 0.32254526019096375, "grad_norm": 37.619651794433594, "learning_rate": 4.019267817841834e-08, "logits/chosen": 0.2957739233970642, "logits/rejected": 0.3121660649776459, "logps/chosen": -445.2286682128906, "logps/ref_chosen": -57.98622512817383, "logps/ref_rejected": -87.91555786132812, "logps/rejected": -685.6959228515625, "loss": 0.857, "margin_dpo/margin_mean": 210.5379180908203, "margin_dpo/margin_std": 210.2278289794922, "step": 570 }, { "epoch": 0.8443465491923642, "fcm_dpo/beta": 0.004019825719296932, "fcm_dpo/delta": -0.013523074798285961, "fcm_dpo/margin": 214.1552276611328, "fcm_dpo/q_t": 0.326122909784317, "grad_norm": 44.181026458740234, "learning_rate": 3.677293317363864e-08, "logits/chosen": 0.2930538058280945, "logits/rejected": 0.3044101595878601, "logps/chosen": -452.9761657714844, "logps/ref_chosen": -55.194114685058594, "logps/ref_rejected": -89.68229675292969, "logps/rejected": -701.61962890625, "loss": 0.8759, "margin_dpo/margin_mean": 214.1552276611328, "margin_dpo/margin_std": 227.3217315673828, "step": 575 }, { "epoch": 0.8516886930983847, "fcm_dpo/beta": 0.004043132532387972, "fcm_dpo/delta": 0.005774930119514465, "fcm_dpo/margin": 208.61080932617188, "fcm_dpo/q_t": 0.32523873448371887, "grad_norm": 33.26530456542969, "learning_rate": 3.349364905389032e-08, "logits/chosen": 0.2503589987754822, "logits/rejected": 0.2608030438423157, "logps/chosen": -421.0636291503906, "logps/ref_chosen": -54.605796813964844, "logps/ref_rejected": -88.9614486694336, "logps/rejected": -664.0301513671875, "loss": 0.8625, "margin_dpo/margin_mean": 208.61080932617188, "margin_dpo/margin_std": 206.9934539794922, "step": 580 }, { "epoch": 0.8590308370044053, "fcm_dpo/beta": 0.004104516003280878, "fcm_dpo/delta": -0.005921521689742804, "fcm_dpo/margin": 208.0009307861328, "fcm_dpo/q_t": 0.3261064291000366, "grad_norm": 40.046321868896484, "learning_rate": 3.035698600998121e-08, "logits/chosen": 0.28162720799446106, "logits/rejected": 0.2981635630130768, "logps/chosen": -438.46337890625, "logps/ref_chosen": -59.03770065307617, "logps/ref_rejected": -94.4029541015625, "logps/rejected": -681.82958984375, "loss": 0.8647, "margin_dpo/margin_mean": 208.0009307861328, "margin_dpo/margin_std": 209.61935424804688, "step": 585 }, { "epoch": 0.8663729809104258, "fcm_dpo/beta": 0.00406238529831171, "fcm_dpo/delta": -0.020377200096845627, "fcm_dpo/margin": 213.8004913330078, "fcm_dpo/q_t": 0.32257598638534546, "grad_norm": 41.44359588623047, "learning_rate": 2.736501028272095e-08, "logits/chosen": 0.2646399140357971, "logits/rejected": 0.268844336271286, "logps/chosen": -428.0189514160156, "logps/ref_chosen": -53.5163688659668, "logps/ref_rejected": -99.90290832519531, "logps/rejected": -688.2059326171875, "loss": 0.8486, "margin_dpo/margin_mean": 213.8004913330078, "margin_dpo/margin_std": 211.3090057373047, "step": 590 }, { "epoch": 0.8737151248164464, "fcm_dpo/beta": 0.004042100626975298, "fcm_dpo/delta": 0.045332007110118866, "fcm_dpo/margin": 199.73629760742188, "fcm_dpo/q_t": 0.3346654772758484, "grad_norm": 32.38268280029297, "learning_rate": 2.451969280180849e-08, "logits/chosen": 0.30068081617355347, "logits/rejected": 0.32306617498397827, "logps/chosen": -435.3770446777344, "logps/ref_chosen": -51.44538497924805, "logps/ref_rejected": -77.43083190917969, "logps/rejected": -661.0988159179688, "loss": 0.9023, "margin_dpo/margin_mean": 199.73629760742188, "margin_dpo/margin_std": 218.01443481445312, "step": 595 }, { "epoch": 0.8810572687224669, "fcm_dpo/beta": 0.0041966065764427185, "fcm_dpo/delta": 0.029329314827919006, "fcm_dpo/margin": 195.78518676757812, "fcm_dpo/q_t": 0.3304395377635956, "grad_norm": 49.09275436401367, "learning_rate": 2.1822907887504932e-08, "logits/chosen": 0.2844870686531067, "logits/rejected": 0.29689234495162964, "logps/chosen": -457.60186767578125, "logps/ref_chosen": -57.161705017089844, "logps/ref_rejected": -87.73274230957031, "logps/rejected": -683.9581298828125, "loss": 0.8894, "margin_dpo/margin_mean": 195.78518676757812, "margin_dpo/margin_std": 207.5529327392578, "step": 600 }, { "epoch": 0.8810572687224669, "eval_fcm_dpo/beta": 0.005688491743057966, "eval_fcm_dpo/delta": 0.04913259670138359, "eval_fcm_dpo/margin": 139.8485870361328, "eval_fcm_dpo/q_t": 0.3537566661834717, "eval_logits/chosen": 0.25595730543136597, "eval_logits/rejected": 0.27860310673713684, "eval_logps/chosen": -511.802978515625, "eval_logps/ref_chosen": -79.05104064941406, "eval_logps/ref_rejected": -86.79793548583984, "eval_logps/rejected": -659.3984375, "eval_loss": 0.5130776166915894, "eval_margin_dpo/margin_mean": 139.8485870361328, "eval_margin_dpo/margin_std": 209.58946228027344, "eval_runtime": 39.5413, "eval_samples_per_second": 59.153, "eval_steps_per_second": 1.871, "step": 600 }, { "epoch": 0.8883994126284875, "fcm_dpo/beta": 0.005007568746805191, "fcm_dpo/delta": -0.2992556691169739, "fcm_dpo/margin": 222.298583984375, "fcm_dpo/q_t": 0.2881125211715698, "grad_norm": 34.59627914428711, "learning_rate": 1.9276432015946446e-08, "logits/chosen": 0.24366600811481476, "logits/rejected": 0.25845664739608765, "logps/chosen": -441.85162353515625, "logps/ref_chosen": -58.169830322265625, "logps/ref_rejected": -95.36891174316406, "logps/rejected": -701.3492431640625, "loss": 0.7792, "margin_dpo/margin_mean": 222.298583984375, "margin_dpo/margin_std": 220.7966766357422, "step": 605 }, { "epoch": 0.895741556534508, "fcm_dpo/beta": 0.004342931322753429, "fcm_dpo/delta": -0.05808112025260925, "fcm_dpo/margin": 207.65054321289062, "fcm_dpo/q_t": 0.31335949897766113, "grad_norm": 41.44016647338867, "learning_rate": 1.6881942648911074e-08, "logits/chosen": 0.2341168224811554, "logits/rejected": 0.2553400695323944, "logps/chosen": -439.32952880859375, "logps/ref_chosen": -58.97087860107422, "logps/ref_rejected": -89.0286865234375, "logps/rejected": -677.037841796875, "loss": 0.8149, "margin_dpo/margin_mean": 207.65054321289062, "margin_dpo/margin_std": 189.12887573242188, "step": 610 }, { "epoch": 0.9030837004405287, "fcm_dpo/beta": 0.00410483730956912, "fcm_dpo/delta": -0.03579559922218323, "fcm_dpo/margin": 214.7685089111328, "fcm_dpo/q_t": 0.3158699870109558, "grad_norm": 38.849464416503906, "learning_rate": 1.4641017128809801e-08, "logits/chosen": 0.2525313198566437, "logits/rejected": 0.26471230387687683, "logps/chosen": -440.29595947265625, "logps/ref_chosen": -58.081878662109375, "logps/ref_rejected": -95.92155456542969, "logps/rejected": -692.904052734375, "loss": 0.8145, "margin_dpo/margin_mean": 214.7685089111328, "margin_dpo/margin_std": 188.67098999023438, "step": 615 }, { "epoch": 0.9104258443465492, "fcm_dpo/beta": 0.004342620261013508, "fcm_dpo/delta": 0.06593510508537292, "fcm_dpo/margin": 181.14906311035156, "fcm_dpo/q_t": 0.34116441011428833, "grad_norm": 46.56772232055664, "learning_rate": 1.2555131639630567e-08, "logits/chosen": 0.24702966213226318, "logits/rejected": 0.2741623520851135, "logps/chosen": -464.8133850097656, "logps/ref_chosen": -62.203094482421875, "logps/ref_rejected": -80.53683471679688, "logps/rejected": -664.2962036132812, "loss": 0.9129, "margin_dpo/margin_mean": 181.14906311035156, "margin_dpo/margin_std": 204.39202880859375, "step": 620 }, { "epoch": 0.9177679882525698, "fcm_dpo/beta": 0.004249856807291508, "fcm_dpo/delta": -0.03912333771586418, "fcm_dpo/margin": 208.0172576904297, "fcm_dpo/q_t": 0.32065385580062866, "grad_norm": 40.270851135253906, "learning_rate": 1.0625660234518913e-08, "logits/chosen": 0.2736581265926361, "logits/rejected": 0.2979634702205658, "logps/chosen": -449.50860595703125, "logps/ref_chosen": -61.727455139160156, "logps/ref_rejected": -88.4387435913086, "logps/rejected": -684.2371826171875, "loss": 0.8444, "margin_dpo/margin_mean": 208.0172882080078, "margin_dpo/margin_std": 208.2506103515625, "step": 625 }, { "epoch": 0.9251101321585903, "fcm_dpo/beta": 0.004256346728652716, "fcm_dpo/delta": 0.025741109624505043, "fcm_dpo/margin": 193.95529174804688, "fcm_dpo/q_t": 0.33398014307022095, "grad_norm": 46.073326110839844, "learning_rate": 8.85387393063622e-09, "logits/chosen": 0.1979989856481552, "logits/rejected": 0.21419593691825867, "logps/chosen": -437.273193359375, "logps/ref_chosen": -61.30865478515625, "logps/ref_rejected": -96.54997253417969, "logps/rejected": -666.4697875976562, "loss": 0.8807, "margin_dpo/margin_mean": 193.95529174804688, "margin_dpo/margin_std": 204.22341918945312, "step": 630 }, { "epoch": 0.9324522760646109, "fcm_dpo/beta": 0.004541632253676653, "fcm_dpo/delta": 0.08948247134685516, "fcm_dpo/margin": 168.99508666992188, "fcm_dpo/q_t": 0.3458203673362732, "grad_norm": 41.188926696777344, "learning_rate": 7.240939871891699e-09, "logits/chosen": 0.26888564229011536, "logits/rejected": 0.2919827103614807, "logps/chosen": -468.99005126953125, "logps/ref_chosen": -63.7315673828125, "logps/ref_rejected": -89.66435241699219, "logps/rejected": -663.9180297851562, "loss": 0.944, "margin_dpo/margin_mean": 168.99508666992188, "margin_dpo/margin_std": 206.3029022216797, "step": 635 }, { "epoch": 0.9397944199706314, "fcm_dpo/beta": 0.004565252922475338, "fcm_dpo/delta": -0.08675488084554672, "fcm_dpo/margin": 203.2274932861328, "fcm_dpo/q_t": 0.3128082752227783, "grad_norm": 34.060367584228516, "learning_rate": 5.7879205600998296e-09, "logits/chosen": 0.22282154858112335, "logits/rejected": 0.24946781992912292, "logps/chosen": -445.6302185058594, "logps/ref_chosen": -59.17915725708008, "logps/ref_rejected": -88.51210021972656, "logps/rejected": -678.190673828125, "loss": 0.823, "margin_dpo/margin_mean": 203.2274932861328, "margin_dpo/margin_std": 194.40232849121094, "step": 640 }, { "epoch": 0.947136563876652, "fcm_dpo/beta": 0.004437420517206192, "fcm_dpo/delta": 0.06028025597333908, "fcm_dpo/margin": 178.86392211914062, "fcm_dpo/q_t": 0.3435667157173157, "grad_norm": 50.218833923339844, "learning_rate": 4.495773155069299e-09, "logits/chosen": 0.2428874969482422, "logits/rejected": 0.2540218234062195, "logps/chosen": -454.20220947265625, "logps/ref_chosen": -59.50596237182617, "logps/ref_rejected": -93.92404174804688, "logps/rejected": -667.4842529296875, "loss": 0.9275, "margin_dpo/margin_mean": 178.86392211914062, "margin_dpo/margin_std": 214.05029296875, "step": 645 }, { "epoch": 0.9544787077826725, "fcm_dpo/beta": 0.004491353873163462, "fcm_dpo/delta": -0.053510040044784546, "fcm_dpo/margin": 199.87738037109375, "fcm_dpo/q_t": 0.3246760070323944, "grad_norm": 47.475616455078125, "learning_rate": 3.3653488440851253e-09, "logits/chosen": 0.27860307693481445, "logits/rejected": 0.2957097887992859, "logps/chosen": -443.88482666015625, "logps/ref_chosen": -57.774566650390625, "logps/ref_rejected": -89.61600494384766, "logps/rejected": -675.6035766601562, "loss": 0.8711, "margin_dpo/margin_mean": 199.87738037109375, "margin_dpo/margin_std": 219.21707153320312, "step": 650 }, { "epoch": 0.9618208516886931, "fcm_dpo/beta": 0.004239661619067192, "fcm_dpo/delta": -0.012966930866241455, "fcm_dpo/margin": 203.01885986328125, "fcm_dpo/q_t": 0.3295535147190094, "grad_norm": 43.02524185180664, "learning_rate": 2.397392281198729e-09, "logits/chosen": 0.2716834545135498, "logits/rejected": 0.277962863445282, "logps/chosen": -437.599609375, "logps/ref_chosen": -55.68403244018555, "logps/ref_rejected": -102.4081802368164, "logps/rejected": -687.3425903320312, "loss": 0.8756, "margin_dpo/margin_mean": 203.01885986328125, "margin_dpo/margin_std": 221.51998901367188, "step": 655 }, { "epoch": 0.9691629955947136, "fcm_dpo/beta": 0.003964302130043507, "fcm_dpo/delta": -0.10601212829351425, "fcm_dpo/margin": 237.7954559326172, "fcm_dpo/q_t": 0.30317944288253784, "grad_norm": 35.75173568725586, "learning_rate": 1.592541096695571e-09, "logits/chosen": 0.2771255373954773, "logits/rejected": 0.2976723611354828, "logps/chosen": -440.38336181640625, "logps/ref_chosen": -59.19981002807617, "logps/ref_rejected": -94.19200134277344, "logps/rejected": -713.1710815429688, "loss": 0.7768, "margin_dpo/margin_mean": 237.7954559326172, "margin_dpo/margin_std": 198.18447875976562, "step": 660 }, { "epoch": 0.9765051395007343, "fcm_dpo/beta": 0.003923547919839621, "fcm_dpo/delta": 0.008545741438865662, "fcm_dpo/margin": 214.4600830078125, "fcm_dpo/q_t": 0.3268297612667084, "grad_norm": 39.92269515991211, "learning_rate": 9.513254770636137e-10, "logits/chosen": 0.2574383616447449, "logits/rejected": 0.2810707688331604, "logps/chosen": -439.8953552246094, "logps/ref_chosen": -61.2533073425293, "logps/ref_rejected": -95.86351013183594, "logps/rejected": -688.9656982421875, "loss": 0.8615, "margin_dpo/margin_mean": 214.46011352539062, "margin_dpo/margin_std": 217.1322021484375, "step": 665 }, { "epoch": 0.9838472834067548, "fcm_dpo/beta": 0.003961815033107996, "fcm_dpo/delta": -0.015965379774570465, "fcm_dpo/margin": 218.12954711914062, "fcm_dpo/q_t": 0.31941553950309753, "grad_norm": 31.397859573364258, "learning_rate": 4.741678157389739e-10, "logits/chosen": 0.2319277822971344, "logits/rejected": 0.25498372316360474, "logps/chosen": -442.1675720214844, "logps/ref_chosen": -62.95263671875, "logps/ref_rejected": -92.4662094116211, "logps/rejected": -689.8107299804688, "loss": 0.835, "margin_dpo/margin_mean": 218.12954711914062, "margin_dpo/margin_std": 201.41766357421875, "step": 670 }, { "epoch": 0.9911894273127754, "fcm_dpo/beta": 0.00390138803049922, "fcm_dpo/delta": -0.02992972731590271, "fcm_dpo/margin": 224.89852905273438, "fcm_dpo/q_t": 0.31749120354652405, "grad_norm": 32.01641845703125, "learning_rate": 1.6138243485910863e-10, "logits/chosen": 0.2925337255001068, "logits/rejected": 0.3101763129234314, "logps/chosen": -432.8349609375, "logps/ref_chosen": -48.5856819152832, "logps/ref_rejected": -81.27871704101562, "logps/rejected": -690.4266357421875, "loss": 0.8262, "margin_dpo/margin_mean": 224.8985595703125, "margin_dpo/margin_std": 204.8478546142578, "step": 675 }, { "epoch": 0.9985315712187959, "fcm_dpo/beta": 0.0037871976383030415, "fcm_dpo/delta": 0.029484611004590988, "fcm_dpo/margin": 217.26602172851562, "fcm_dpo/q_t": 0.32743436098098755, "grad_norm": 36.42646408081055, "learning_rate": 1.31753782067201e-11, "logits/chosen": 0.2187252789735794, "logits/rejected": 0.2422519475221634, "logps/chosen": -443.4412536621094, "logps/ref_chosen": -60.25421905517578, "logps/ref_rejected": -87.23457336425781, "logps/rejected": -687.6876220703125, "loss": 0.8521, "margin_dpo/margin_mean": 217.2659912109375, "margin_dpo/margin_std": 204.62794494628906, "step": 680 }, { "epoch": 1.0, "step": 681, "total_flos": 0.0, "train_loss": 0.8947697063024523, "train_runtime": 1846.2605, "train_samples_per_second": 23.614, "train_steps_per_second": 0.369 } ], "logging_steps": 5, "max_steps": 681, "num_input_tokens_seen": 0, "num_train_epochs": 1, "save_steps": 50, "stateful_callbacks": { "TrainerControl": { "args": { "should_epoch_stop": false, "should_evaluate": false, "should_log": false, "should_save": false, "should_training_stop": false }, "attributes": {} } }, "total_flos": 0.0, "train_batch_size": 8, "trial_name": null, "trial_params": null }