full-state checkpoint 10-percent (step 18)

ckpt-10-percent/adapter_config.json

@@ -33,13 +33,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "gate_proj",
+    "o_proj",
+    "down_proj",
+    "v_proj",
     "up_proj",
     "k_proj",
-    "v_proj",
-    "q_proj",
-    "o_proj",
-    "down_proj"
+    "gate_proj",
+    "q_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

ckpt-10-percent/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:faba7bf27da211446cb3b05e6f02645be87ed91b0045d707881b90b6f5d73e61
+oid sha256:21af0c633d2da335d9d6aaaef29e293715e3dc29f05b3da105240e8a8c31cae1
 size 528550256

ckpt-10-percent/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d762b384fd4bead5fa88a0e2ab58599779ed7b6ee4e7c67aa6303e09c14ec88
+oid sha256:52c845fe274003df5673fa3d46febb9caa4300017ba87357e77253e579b83ec7
 size 268963141

ckpt-10-percent/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac181f8f177806b35c235b3f05e6a213cc2298fd58db12cff4427c8080df8355
+oid sha256:3dd479b6438efe3b53c98209a581eb1fbf7518bcf62036fd20199a24854bfa54
 size 14645

ckpt-10-percent/trainer_state.json

@@ -10,383 +10,383 @@
   "is_world_process_zero": true,
   "log_history": [
     {
-      "completion_length": 405.7375030517578,
+      "completion_length": 422.0800064086914,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1297.0,
-      "completions/max_terminated_length": 1297.0,
-      "completions/mean_length": 406.73748779296875,
-      "completions/mean_terminated_length": 406.73748779296875,
-      "completions/min_length": 11.0,
-      "completions/min_terminated_length": 11.0,
+      "completions/max_length": 1117.0,
+      "completions/max_terminated_length": 1117.0,
+      "completions/mean_length": 423.0799865722656,
+      "completions/mean_terminated_length": 423.0799865722656,
+      "completions/min_length": 163.0,
+      "completions/min_terminated_length": 163.0,
       "epoch": 0.0053475935828877,
-      "frac_reward_zero_std": 0.125,
-      "grad_norm": 0.24461160600185394,
+      "frac_reward_zero_std": 0.04999999701976776,
+      "grad_norm": 0.19562773406505585,
       "kl": 0.0,
       "learning_rate": 0.0,
       "loss": 0.0,
-      "num_tokens": 296975.0,
-      "reward": 0.5436059236526489,
-      "reward_std": 0.6562731862068176,
-      "rewards/multidomain_reward_func/mean": 0.5436059832572937,
-      "rewards/multidomain_reward_func/std": 1.3712321519851685,
+      "num_tokens": 303512.0,
+      "reward": 0.970195472240448,
+      "reward_std": 0.56549471616745,
+      "rewards/multidomain_reward_func/mean": 0.9701954126358032,
+      "rewards/multidomain_reward_func/std": 1.1971898078918457,
       "step": 1
     },
     {
-      "completion_length": 386.7625076293945,
+      "completion_length": 414.440007019043,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1488.0,
-      "completions/max_terminated_length": 1488.0,
-      "completions/mean_length": 387.7624816894531,
-      "completions/mean_terminated_length": 387.7624816894531,
-      "completions/min_length": 11.0,
-      "completions/min_terminated_length": 11.0,
+      "completions/max_length": 1388.0,
+      "completions/max_terminated_length": 1388.0,
+      "completions/mean_length": 415.44000244140625,
+      "completions/mean_terminated_length": 415.44000244140625,
+      "completions/min_length": 161.0,
+      "completions/min_terminated_length": 161.0,
       "epoch": 0.0106951871657754,
-      "frac_reward_zero_std": 0.04999999701976776,
-      "grad_norm": 0.39698562026023865,
+      "frac_reward_zero_std": 0.0,
+      "grad_norm": 0.16706927120685577,
       "kl": 0.0,
       "learning_rate": 1.6666666666666667e-06,
       "loss": 0.0,
-      "num_tokens": 573080.0,
-      "reward": 0.7371304035186768,
-      "reward_std": 0.7435028553009033,
-      "rewards/multidomain_reward_func/mean": 0.737130343914032,
-      "rewards/multidomain_reward_func/std": 1.2039875984191895,
+      "num_tokens": 590688.0,
+      "reward": 1.0831713676452637,
+      "reward_std": 0.4986020028591156,
+      "rewards/multidomain_reward_func/mean": 1.0831713676452637,
+      "rewards/multidomain_reward_func/std": 0.9529362916946411,
       "step": 2
     },
     {
-      "completion_length": 423.4725051879883,
+      "completion_length": 460.0900085449219,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1453.0,
-      "completions/max_terminated_length": 1453.0,
-      "completions/mean_length": 424.4725036621094,
-      "completions/mean_terminated_length": 424.4725036621094,
-      "completions/min_length": 15.0,
-      "completions/min_terminated_length": 15.0,
+      "completions/max_length": 1300.0,
+      "completions/max_terminated_length": 1300.0,
+      "completions/mean_length": 461.0899963378906,
+      "completions/mean_terminated_length": 461.0899963378906,
+      "completions/min_length": 147.0,
+      "completions/min_terminated_length": 147.0,
       "epoch": 0.016042780748663103,
-      "frac_reward_zero_std": 0.14999999105930328,
-      "grad_norm": 0.25578364729881287,
+      "frac_reward_zero_std": 0.04999999701976776,
+      "grad_norm": 0.1598707139492035,
       "kl": 0.0,
       "learning_rate": 3.3333333333333333e-06,
       "loss": 0.0001,
-      "num_tokens": 875249.0,
-      "reward": 0.8408688306808472,
-      "reward_std": 0.7539217472076416,
-      "rewards/multidomain_reward_func/mean": 0.8408687114715576,
-      "rewards/multidomain_reward_func/std": 1.2338955402374268,
+      "num_tokens": 907504.0,
+      "reward": 1.1258100271224976,
+      "reward_std": 0.6013069152832031,
+      "rewards/multidomain_reward_func/mean": 1.125809907913208,
+      "rewards/multidomain_reward_func/std": 1.093148946762085,
       "step": 3
     },
     {
-      "completion_length": 409.7625076293945,
+      "completion_length": 429.9500030517578,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 900.0,
-      "completions/max_terminated_length": 900.0,
-      "completions/mean_length": 410.7624816894531,
-      "completions/mean_terminated_length": 410.7624816894531,
-      "completions/min_length": 12.0,
-      "completions/min_terminated_length": 12.0,
+      "completions/max_length": 999.0,
+      "completions/max_terminated_length": 999.0,
+      "completions/mean_length": 430.9499816894531,
+      "completions/mean_terminated_length": 430.9499816894531,
+      "completions/min_length": 157.0,
+      "completions/min_terminated_length": 157.0,
       "epoch": 0.0213903743315508,
-      "frac_reward_zero_std": 0.17499999701976776,
-      "grad_norm": 0.2062363624572754,
+      "frac_reward_zero_std": 0.04999999701976776,
+      "grad_norm": 0.15568800270557404,
       "kl": 0.0,
       "learning_rate": 5e-06,
       "loss": 0.0001,
-      "num_tokens": 1160774.0,
-      "reward": 0.03626520559191704,
-      "reward_std": 1.2230231761932373,
-      "rewards/multidomain_reward_func/mean": 0.03626519814133644,
-      "rewards/multidomain_reward_func/std": 1.777443766593933,
+      "num_tokens": 1201104.0,
+      "reward": 1.0431935787200928,
+      "reward_std": 0.7159458994865417,
+      "rewards/multidomain_reward_func/mean": 1.0431934595108032,
+      "rewards/multidomain_reward_func/std": 1.1585919857025146,
       "step": 4
     },
     {
-      "completion_length": 414.07500610351565,
+      "completion_length": 430.0450012207031,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1260.0,
-      "completions/max_terminated_length": 1260.0,
-      "completions/mean_length": 415.0749816894531,
-      "completions/mean_terminated_length": 415.0749816894531,
-      "completions/min_length": 9.0,
-      "completions/min_terminated_length": 9.0,
+      "completions/max_length": 1000.0,
+      "completions/max_terminated_length": 1000.0,
+      "completions/mean_length": 431.04498291015625,
+      "completions/mean_terminated_length": 431.04498291015625,
+      "completions/min_length": 151.0,
+      "completions/min_terminated_length": 151.0,
       "epoch": 0.026737967914438502,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.09999999403953552,
+      "grad_norm": 0.14256730675697327,
       "kl": 0.0,
       "learning_rate": 6.666666666666667e-06,
-      "loss": 0.0,
-      "num_tokens": 1460064.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0001,
+      "num_tokens": 1506782.0,
+      "reward": 1.0953773260116577,
+      "reward_std": 0.6102118492126465,
+      "rewards/multidomain_reward_func/mean": 1.0953773260116577,
+      "rewards/multidomain_reward_func/std": 1.120749831199646,
       "step": 5
     },
     {
-      "completion_length": 419.4750061035156,
-      "completions/clipped_ratio": 0.0024999999441206455,
-      "completions/max_length": 2048.0,
-      "completions/max_terminated_length": 899.0,
-      "completions/mean_length": 420.4725036621094,
-      "completions/mean_terminated_length": 416.39349365234375,
-      "completions/min_length": 37.0,
-      "completions/min_terminated_length": 37.0,
+      "completion_length": 415.45750732421874,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 904.0,
+      "completions/max_terminated_length": 904.0,
+      "completions/mean_length": 416.4574890136719,
+      "completions/mean_terminated_length": 416.4574890136719,
+      "completions/min_length": 151.0,
+      "completions/min_terminated_length": 151.0,
       "epoch": 0.03208556149732621,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.25,
+      "grad_norm": 0.13746652007102966,
       "kl": 0.0,
       "learning_rate": 8.333333333333334e-06,
-      "loss": 0.0,
-      "num_tokens": 1759043.0,
-      "reward": -0.07499999552965164,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": -0.07500000298023224,
-      "rewards/multidomain_reward_func/std": 0.4689613878726959,
+      "loss": 0.0001,
+      "num_tokens": 1804155.0,
+      "reward": 1.2259290218353271,
+      "reward_std": 0.566255509853363,
+      "rewards/multidomain_reward_func/mean": 1.2259290218353271,
+      "rewards/multidomain_reward_func/std": 1.1134203672409058,
       "step": 6
     },
     {
-      "completion_length": 385.43250732421876,
+      "completion_length": 378.8675033569336,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 954.0,
-      "completions/max_terminated_length": 954.0,
-      "completions/mean_length": 386.4324951171875,
-      "completions/mean_terminated_length": 386.4324951171875,
-      "completions/min_length": 118.0,
-      "completions/min_terminated_length": 118.0,
+      "completions/max_length": 928.0,
+      "completions/max_terminated_length": 928.0,
+      "completions/mean_length": 379.86749267578125,
+      "completions/mean_terminated_length": 379.86749267578125,
+      "completions/min_length": 109.0,
+      "completions/min_terminated_length": 109.0,
       "epoch": 0.0374331550802139,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.22499999403953552,
+      "grad_norm": 0.149386927485466,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 2022876.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0002,
+      "num_tokens": 2065362.0,
+      "reward": 1.0153580904006958,
+      "reward_std": 0.49818137288093567,
+      "rewards/multidomain_reward_func/mean": 1.0153579711914062,
+      "rewards/multidomain_reward_func/std": 1.130789875984192,
       "step": 7
     },
     {
-      "completion_length": 395.05250549316406,
+      "completion_length": 395.4450088500977,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 804.0,
-      "completions/max_terminated_length": 804.0,
-      "completions/mean_length": 396.052490234375,
-      "completions/mean_terminated_length": 396.052490234375,
-      "completions/min_length": 124.0,
-      "completions/min_terminated_length": 124.0,
+      "completions/max_length": 911.0,
+      "completions/max_terminated_length": 911.0,
+      "completions/mean_length": 396.4449768066406,
+      "completions/mean_terminated_length": 396.4449768066406,
+      "completions/min_length": 117.0,
+      "completions/min_terminated_length": 117.0,
       "epoch": 0.0427807486631016,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.22499999403953552,
+      "grad_norm": 0.13969184458255768,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 2299967.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0002,
+      "num_tokens": 2342610.0,
+      "reward": 0.9266104102134705,
+      "reward_std": 0.6934035420417786,
+      "rewards/multidomain_reward_func/mean": 0.9266103506088257,
+      "rewards/multidomain_reward_func/std": 1.2882972955703735,
       "step": 8
     },
     {
-      "completion_length": 415.39000701904297,
+      "completion_length": 414.67000579833984,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1002.0,
-      "completions/max_terminated_length": 1002.0,
-      "completions/mean_length": 416.3899841308594,
-      "completions/mean_terminated_length": 416.3899841308594,
-      "completions/min_length": 16.0,
-      "completions/min_terminated_length": 16.0,
+      "completions/max_length": 1099.0,
+      "completions/max_terminated_length": 1099.0,
+      "completions/mean_length": 415.66998291015625,
+      "completions/mean_terminated_length": 415.66998291015625,
+      "completions/min_length": 168.0,
+      "completions/min_terminated_length": 168.0,
       "epoch": 0.0481283422459893,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.14999999105930328,
+      "grad_norm": 0.1436053365468979,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 2589473.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0002,
+      "num_tokens": 2631828.0,
+      "reward": 1.1906582117080688,
+      "reward_std": 0.6041886806488037,
+      "rewards/multidomain_reward_func/mean": 1.1906582117080688,
+      "rewards/multidomain_reward_func/std": 1.0831336975097656,
       "step": 9
     },
     {
-      "completion_length": 442.06250915527346,
+      "completion_length": 440.33250885009767,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1095.0,
-      "completions/max_terminated_length": 1095.0,
-      "completions/mean_length": 443.0625,
-      "completions/mean_terminated_length": 443.0625,
-      "completions/min_length": 28.0,
-      "completions/min_terminated_length": 28.0,
+      "completions/max_length": 961.0,
+      "completions/max_terminated_length": 961.0,
+      "completions/mean_length": 441.3324890136719,
+      "completions/mean_terminated_length": 441.3324890136719,
+      "completions/min_length": 125.0,
+      "completions/min_terminated_length": 125.0,
       "epoch": 0.053475935828877004,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.125,
+      "grad_norm": 0.14258432388305664,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 2898338.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0001,
+      "num_tokens": 2940001.0,
+      "reward": 0.8959593772888184,
+      "reward_std": 0.506439745426178,
+      "rewards/multidomain_reward_func/mean": 0.8959593176841736,
+      "rewards/multidomain_reward_func/std": 1.2946927547454834,
       "step": 10
     },
     {
-      "completion_length": 398.72000579833986,
+      "completion_length": 410.9975051879883,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 904.0,
-      "completions/max_terminated_length": 904.0,
-      "completions/mean_length": 399.7200012207031,
-      "completions/mean_terminated_length": 399.7200012207031,
+      "completions/max_length": 920.0,
+      "completions/max_terminated_length": 920.0,
+      "completions/mean_length": 411.99749755859375,
+      "completions/mean_terminated_length": 411.99749755859375,
       "completions/min_length": 110.0,
       "completions/min_terminated_length": 110.0,
       "epoch": 0.058823529411764705,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.17499999701976776,
+      "grad_norm": 0.15353518724441528,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 3179696.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": 0.0002,
+      "num_tokens": 3226270.0,
+      "reward": 1.0268709659576416,
+      "reward_std": 0.5810268521308899,
+      "rewards/multidomain_reward_func/mean": 1.0268710851669312,
+      "rewards/multidomain_reward_func/std": 1.0052762031555176,
       "step": 11
     },
     {
-      "completion_length": 401.6100082397461,
+      "completion_length": 411.8925079345703,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 906.0,
-      "completions/max_terminated_length": 906.0,
-      "completions/mean_length": 402.6099853515625,
-      "completions/mean_terminated_length": 402.6099853515625,
-      "completions/min_length": 104.0,
-      "completions/min_terminated_length": 104.0,
+      "completions/max_length": 1092.0,
+      "completions/max_terminated_length": 1092.0,
+      "completions/mean_length": 412.8924865722656,
+      "completions/mean_terminated_length": 412.8924865722656,
+      "completions/min_length": 97.0,
+      "completions/min_terminated_length": 97.0,
       "epoch": 0.06417112299465241,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.02499999850988388,
+      "grad_norm": 0.1631866842508316,
       "kl": 0.0,
       "learning_rate": 1e-05,
       "loss": 0.0,
-      "num_tokens": 3461840.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "num_tokens": 3512527.0,
+      "reward": 1.2455800771713257,
+      "reward_std": 0.5915975570678711,
+      "rewards/multidomain_reward_func/mean": 1.2455800771713257,
+      "rewards/multidomain_reward_func/std": 1.0622782707214355,
       "step": 12
     },
     {
-      "completion_length": 378.29750671386716,
+      "completion_length": 404.74000396728513,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 800.0,
-      "completions/max_terminated_length": 800.0,
-      "completions/mean_length": 379.2974853515625,
-      "completions/mean_terminated_length": 379.2974853515625,
-      "completions/min_length": 148.0,
-      "completions/min_terminated_length": 148.0,
+      "completions/max_length": 1191.0,
+      "completions/max_terminated_length": 1191.0,
+      "completions/mean_length": 405.739990234375,
+      "completions/mean_terminated_length": 405.739990234375,
+      "completions/min_length": 136.0,
+      "completions/min_terminated_length": 136.0,
       "epoch": 0.06951871657754011,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.0,
+      "grad_norm": 0.17720060050487518,
       "kl": 0.0,
       "learning_rate": 1e-05,
       "loss": 0.0,
-      "num_tokens": 3730789.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "num_tokens": 3792053.0,
+      "reward": 1.055298924446106,
+      "reward_std": 0.6600240468978882,
+      "rewards/multidomain_reward_func/mean": 1.0552990436553955,
+      "rewards/multidomain_reward_func/std": 1.296553134918213,
       "step": 13
     },
     {
-      "completion_length": 443.53000793457034,
-      "completions/clipped_ratio": 0.004999999888241291,
-      "completions/max_length": 2048.0,
-      "completions/max_terminated_length": 1181.0,
-      "completions/mean_length": 444.5249938964844,
-      "completions/mean_terminated_length": 436.46734619140625,
-      "completions/min_length": 156.0,
-      "completions/min_terminated_length": 156.0,
+      "completion_length": 446.06500854492185,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 1243.0,
+      "completions/max_terminated_length": 1243.0,
+      "completions/mean_length": 447.06500244140625,
+      "completions/mean_terminated_length": 447.06500244140625,
+      "completions/min_length": 110.0,
+      "completions/min_terminated_length": 110.0,
       "epoch": 0.0748663101604278,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.04999999701976776,
+      "grad_norm": 0.2615317106246948,
       "kl": 0.0,
       "learning_rate": 1e-05,
       "loss": 0.0,
-      "num_tokens": 4043349.0,
-      "reward": -0.07499999552965164,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": -0.07500000298023224,
-      "rewards/multidomain_reward_func/std": 0.4689613878726959,
+      "num_tokens": 4105629.0,
+      "reward": 0.9167600870132446,
+      "reward_std": 0.8242074251174927,
+      "rewards/multidomain_reward_func/mean": 0.9167600870132446,
+      "rewards/multidomain_reward_func/std": 1.2414448261260986,
       "step": 14
     },
     {
-      "completion_length": 438.34000396728516,
-      "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1026.0,
-      "completions/max_terminated_length": 1026.0,
-      "completions/mean_length": 439.3399963378906,
-      "completions/mean_terminated_length": 439.3399963378906,
-      "completions/min_length": 104.0,
-      "completions/min_terminated_length": 104.0,
+      "completion_length": 469.11250762939454,
+      "completions/clipped_ratio": 0.0024999999441206455,
+      "completions/max_length": 2048.0,
+      "completions/max_terminated_length": 1411.0,
+      "completions/mean_length": 470.1099853515625,
+      "completions/mean_terminated_length": 466.1553955078125,
+      "completions/min_length": 118.0,
+      "completions/min_terminated_length": 118.0,
       "epoch": 0.08021390374331551,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.07499999552965164,
+      "grad_norm": 0.1610775738954544,
       "kl": 0.0,
       "learning_rate": 1e-05,
       "loss": 0.0,
-      "num_tokens": 4347675.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "num_tokens": 4422263.0,
+      "reward": 0.9777088165283203,
+      "reward_std": 0.8839526772499084,
+      "rewards/multidomain_reward_func/mean": 0.9777088761329651,
+      "rewards/multidomain_reward_func/std": 1.301257610321045,
       "step": 15
     },
     {
-      "completion_length": 399.96250610351564,
+      "completion_length": 439.1025085449219,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 1510.0,
-      "completions/max_terminated_length": 1510.0,
-      "completions/mean_length": 400.9624938964844,
-      "completions/mean_terminated_length": 400.9624938964844,
-      "completions/min_length": 124.0,
-      "completions/min_terminated_length": 124.0,
+      "completions/max_length": 1598.0,
+      "completions/max_terminated_length": 1598.0,
+      "completions/mean_length": 440.10247802734375,
+      "completions/mean_terminated_length": 440.10247802734375,
+      "completions/min_length": 126.0,
+      "completions/min_terminated_length": 126.0,
       "epoch": 0.0855614973262032,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.0,
+      "grad_norm": 0.17706210911273956,
       "kl": 0.0,
       "learning_rate": 1e-05,
-      "loss": 0.0,
-      "num_tokens": 4635080.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "loss": -0.0,
+      "num_tokens": 4725324.0,
+      "reward": 1.0704784393310547,
+      "reward_std": 0.7980135083198547,
+      "rewards/multidomain_reward_func/mean": 1.0704783201217651,
+      "rewards/multidomain_reward_func/std": 1.1782567501068115,
       "step": 16
     },
     {
-      "completion_length": 409.47500762939455,
+      "completion_length": 408.14000701904297,
       "completions/clipped_ratio": 0.0,
-      "completions/max_length": 942.0,
-      "completions/max_terminated_length": 942.0,
-      "completions/mean_length": 410.4749755859375,
-      "completions/mean_terminated_length": 410.4749755859375,
-      "completions/min_length": 15.0,
-      "completions/min_terminated_length": 15.0,
+      "completions/max_length": 1055.0,
+      "completions/max_terminated_length": 1055.0,
+      "completions/mean_length": 409.1399841308594,
+      "completions/mean_terminated_length": 409.1399841308594,
+      "completions/min_length": 113.0,
+      "completions/min_terminated_length": 113.0,
       "epoch": 0.09090909090909091,
-      "frac_reward_zero_std": 1.0,
-      "grad_norm": 0.0,
+      "frac_reward_zero_std": 0.02499999850988388,
+      "grad_norm": 0.17319922149181366,
       "kl": 0.0,
       "learning_rate": 1e-05,
       "loss": 0.0,
-      "num_tokens": 4919250.0,
-      "reward": 0.0,
-      "reward_std": 0.0,
-      "rewards/multidomain_reward_func/mean": 0.0,
-      "rewards/multidomain_reward_func/std": 0.0,
+      "num_tokens": 5008960.0,
+      "reward": 0.7932982444763184,
+      "reward_std": 0.8552511930465698,
+      "rewards/multidomain_reward_func/mean": 0.7932982444763184,
+      "rewards/multidomain_reward_func/std": 1.2791845798492432,
       "step": 17
     }
   ],
   "logging_steps": 1,
   "max_steps": 187,
-  "num_input_tokens_seen": 5230914,
+  "num_input_tokens_seen": 5323631,
   "num_train_epochs": 1,
   "save_steps": 250,
   "stateful_callbacks": {

ckpt-10-percent/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dcce334ac159b3123921c38139431dda6fa8294e603a7775109a50b297b6712f
+oid sha256:632bb5b102154a6ce7dc9eb22a9348f4184ae0be1ccdcbfaebb8eca6d1ca7512
 size 7505