Add Luminia-8B-RP v0.3

Same v0.2 datasets. Update base model from 'NousResearch/Meta-Llama-3.1-8B-Instruct' to 'meta-llama/Meta-Llama-3.1-8B-Instruct', for possible misconfig.

Luminia-8B-RP-DPO/adapter_config.json

@@ -1,7 +1,7 @@
 {
   "alpha_pattern": {},
   "auto_mapping": null,
-  "base_model_name_or_path": "NousResearch/Meta-Llama-3.1-8B-Instruct",
+  "base_model_name_or_path": "meta-llama/Meta-Llama-3.1-8B-Instruct",
   "bias": "none",
   "fan_in_fan_out": false,
   "inference_mode": true,
@@ -20,13 +20,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
+    "q_proj",
+    "up_proj",
     "v_proj",
-    "k_proj",
+    "o_proj",
     "gate_proj",
-    "q_proj",
     "down_proj",
-    "up_proj",
-    "o_proj"
+    "k_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

Luminia-8B-RP-DPO/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b4989979b346c709c61624d9de12448bda97f42aad388f84d7eda002183f22a
+oid sha256:a678e1675c3ad8bb2684f7ad23bbe6dbfb5d1c29270f8e13f9b20c28eafeb8f0
 size 335604696

Luminia-8B-RP-DPO/all_results.json

@@ -2,8 +2,8 @@
     "epoch": 1.0,
     "num_input_tokens_seen": 75616,
     "total_flos": 3443013082939392.0,
-    "train_loss": 1.2913809109883136,
-    "train_runtime": 56.3341,
-    "train_samples_per_second": 1.473,
-    "train_steps_per_second": 1.473
+    "train_loss": 1.2913588443434383,
+    "train_runtime": 65.9669,
+    "train_samples_per_second": 1.258,
+    "train_steps_per_second": 1.258
 }

Luminia-8B-RP-DPO/train_results.json

@@ -2,8 +2,8 @@
     "epoch": 1.0,
     "num_input_tokens_seen": 75616,
     "total_flos": 3443013082939392.0,
-    "train_loss": 1.2913809109883136,
-    "train_runtime": 56.3341,
-    "train_samples_per_second": 1.473,
-    "train_steps_per_second": 1.473
+    "train_loss": 1.2913588443434383,
+    "train_runtime": 65.9669,
+    "train_samples_per_second": 1.258,
+    "train_steps_per_second": 1.258
 }

Luminia-8B-RP-DPO/trainer_log.jsonl

@@ -1,9 +1,9 @@
-{"current_steps": 10, "total_steps": 83, "loss": 1.4781, "accuracy": 0.8999999761581421, "learning_rate": 4.8230451807939135e-05, "epoch": 0.12048192771084337, "percentage": 12.05, "elapsed_time": "0:00:07", "remaining_time": "0:00:57", "throughput": 1118.38, "total_tokens": 8864}
-{"current_steps": 20, "total_steps": 83, "loss": 1.4732, "accuracy": 1.0, "learning_rate": 4.3172311296078595e-05, "epoch": 0.24096385542168675, "percentage": 24.1, "elapsed_time": "0:00:14", "remaining_time": "0:00:45", "throughput": 1219.49, "total_tokens": 17712}
-{"current_steps": 30, "total_steps": 83, "loss": 1.314, "accuracy": 1.0, "learning_rate": 3.55416283362546e-05, "epoch": 0.3614457831325301, "percentage": 36.14, "elapsed_time": "0:00:21", "remaining_time": "0:00:37", "throughput": 1301.78, "total_tokens": 27520}
-{"current_steps": 40, "total_steps": 83, "loss": 1.2266, "accuracy": 1.0, "learning_rate": 2.6418631827326857e-05, "epoch": 0.4819277108433735, "percentage": 48.19, "elapsed_time": "0:00:27", "remaining_time": "0:00:29", "throughput": 1322.02, "total_tokens": 36096}
-{"current_steps": 50, "total_steps": 83, "loss": 1.0201, "accuracy": 1.0, "learning_rate": 1.70948083275794e-05, "epoch": 0.6024096385542169, "percentage": 60.24, "elapsed_time": "0:00:32", "remaining_time": "0:00:21", "throughput": 1319.87, "total_tokens": 43456}
-{"current_steps": 60, "total_steps": 83, "loss": 1.4448, "accuracy": 1.0, "learning_rate": 8.890074238378074e-06, "epoch": 0.7228915662650602, "percentage": 72.29, "elapsed_time": "0:00:39", "remaining_time": "0:00:14", "throughput": 1318.76, "total_tokens": 51600}
-{"current_steps": 70, "total_steps": 83, "loss": 1.1615, "accuracy": 1.0, "learning_rate": 2.9659233496337786e-06, "epoch": 0.8433734939759037, "percentage": 84.34, "elapsed_time": "0:00:46", "remaining_time": "0:00:08", "throughput": 1347.76, "total_tokens": 62624}
-{"current_steps": 80, "total_steps": 83, "loss": 1.2626, "accuracy": 1.0, "learning_rate": 1.6100130092037703e-07, "epoch": 0.963855421686747, "percentage": 96.39, "elapsed_time": "0:00:53", "remaining_time": "0:00:02", "throughput": 1363.67, "total_tokens": 73184}
-{"current_steps": 83, "total_steps": 83, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:00:56", "remaining_time": "0:00:00", "throughput": 1342.33, "total_tokens": 75616}
+{"current_steps": 10, "total_steps": 83, "loss": 1.4779, "accuracy": 0.8999999761581421, "learning_rate": 4.8230451807939135e-05, "epoch": 0.12048192771084337, "percentage": 12.05, "elapsed_time": "0:00:08", "remaining_time": "0:01:02", "throughput": 1036.63, "total_tokens": 8864}
+{"current_steps": 20, "total_steps": 83, "loss": 1.4745, "accuracy": 1.0, "learning_rate": 4.3172311296078595e-05, "epoch": 0.24096385542168675, "percentage": 24.1, "elapsed_time": "0:00:16", "remaining_time": "0:00:52", "throughput": 1067.18, "total_tokens": 17712}
+{"current_steps": 30, "total_steps": 83, "loss": 1.3135, "accuracy": 1.0, "learning_rate": 3.55416283362546e-05, "epoch": 0.3614457831325301, "percentage": 36.14, "elapsed_time": "0:00:24", "remaining_time": "0:00:43", "throughput": 1117.95, "total_tokens": 27520}
+{"current_steps": 40, "total_steps": 83, "loss": 1.2285, "accuracy": 1.0, "learning_rate": 2.6418631827326857e-05, "epoch": 0.4819277108433735, "percentage": 48.19, "elapsed_time": "0:00:32", "remaining_time": "0:00:34", "throughput": 1122.66, "total_tokens": 36096}
+{"current_steps": 50, "total_steps": 83, "loss": 1.0171, "accuracy": 1.0, "learning_rate": 1.70948083275794e-05, "epoch": 0.6024096385542169, "percentage": 60.24, "elapsed_time": "0:00:38", "remaining_time": "0:00:25", "throughput": 1114.67, "total_tokens": 43456}
+{"current_steps": 60, "total_steps": 83, "loss": 1.4433, "accuracy": 1.0, "learning_rate": 8.890074238378074e-06, "epoch": 0.7228915662650602, "percentage": 72.29, "elapsed_time": "0:00:46", "remaining_time": "0:00:17", "throughput": 1112.96, "total_tokens": 51600}
+{"current_steps": 70, "total_steps": 83, "loss": 1.1624, "accuracy": 1.0, "learning_rate": 2.9659233496337786e-06, "epoch": 0.8433734939759037, "percentage": 84.34, "elapsed_time": "0:00:54", "remaining_time": "0:00:10", "throughput": 1141.19, "total_tokens": 62624}
+{"current_steps": 80, "total_steps": 83, "loss": 1.2645, "accuracy": 1.0, "learning_rate": 1.6100130092037703e-07, "epoch": 0.963855421686747, "percentage": 96.39, "elapsed_time": "0:01:03", "remaining_time": "0:00:02", "throughput": 1158.26, "total_tokens": 73184}
+{"current_steps": 83, "total_steps": 83, "epoch": 1.0, "percentage": 100.0, "elapsed_time": "0:01:05", "remaining_time": "0:00:00", "throughput": 1146.3, "total_tokens": 75616}

Luminia-8B-RP-DPO/trainer_state.json

@@ -10,146 +10,146 @@
   "log_history": [
     {
       "epoch": 0.12048192771084337,
-      "grad_norm": 2.8706510066986084,
+      "grad_norm": 2.8577072620391846,
       "learning_rate": 4.8230451807939135e-05,
-      "logits/chosen": -0.11457079648971558,
-      "logits/rejected": -3.2889251708984375,
-      "logps/chosen": -1.456398367881775,
-      "logps/rejected": -3.6490638256073,
-      "loss": 1.4781,
+      "logits/chosen": -0.11954045295715332,
+      "logits/rejected": -3.3223273754119873,
+      "logps/chosen": -1.4568630456924438,
+      "logps/rejected": -3.726320266723633,
+      "loss": 1.4779,
       "num_input_tokens_seen": 8864,
-      "odds_ratio_loss": 14.7344388961792,
+      "odds_ratio_loss": 14.732812881469727,
       "rewards/accuracies": 0.8999999761581421,
-      "rewards/chosen": -0.14563985168933868,
-      "rewards/margins": 0.21926657855510712,
-      "rewards/rejected": -0.3649064302444458,
-      "sft_loss": 0.0046225739642977715,
+      "rewards/chosen": -0.1456862986087799,
+      "rewards/margins": 0.22694571316242218,
+      "rewards/rejected": -0.3726319968700409,
+      "sft_loss": 0.0046242037788033485,
       "step": 10
     },
     {
       "epoch": 0.24096385542168675,
-      "grad_norm": 2.050874710083008,
+      "grad_norm": 1.9434715509414673,
       "learning_rate": 4.3172311296078595e-05,
-      "logits/chosen": -0.26779693365097046,
-      "logits/rejected": -3.211538314819336,
-      "logps/chosen": -1.4712103605270386,
-      "logps/rejected": -5.510354042053223,
-      "loss": 1.4732,
+      "logits/chosen": -0.2733025550842285,
+      "logits/rejected": -3.2514376640319824,
+      "logps/chosen": -1.4726128578186035,
+      "logps/rejected": -5.625662803649902,
+      "loss": 1.4745,
       "num_input_tokens_seen": 17712,
-      "odds_ratio_loss": 14.650479316711426,
+      "odds_ratio_loss": 14.662857055664062,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.14712102711200714,
-      "rewards/margins": 0.4039144515991211,
-      "rewards/rejected": -0.551035463809967,
-      "sft_loss": 0.008165580220520496,
+      "rewards/chosen": -0.14726129174232483,
+      "rewards/margins": 0.4153049886226654,
+      "rewards/rejected": -0.562566339969635,
+      "sft_loss": 0.008168135769665241,
       "step": 20
     },
     {
       "epoch": 0.3614457831325301,
-      "grad_norm": 2.1595640182495117,
+      "grad_norm": 2.154024124145508,
       "learning_rate": 3.55416283362546e-05,
-      "logits/chosen": -0.2051524668931961,
-      "logits/rejected": -3.36714506149292,
-      "logps/chosen": -1.3134592771530151,
-      "logps/rejected": -6.383711814880371,
-      "loss": 1.314,
+      "logits/chosen": -0.20305636525154114,
+      "logits/rejected": -3.410961627960205,
+      "logps/chosen": -1.3130009174346924,
+      "logps/rejected": -6.4230146408081055,
+      "loss": 1.3135,
       "num_input_tokens_seen": 27520,
-      "odds_ratio_loss": 13.103729248046875,
+      "odds_ratio_loss": 13.098909378051758,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.1313459426164627,
-      "rewards/margins": 0.5070253610610962,
-      "rewards/rejected": -0.6383712291717529,
-      "sft_loss": 0.0035842768847942352,
+      "rewards/chosen": -0.13130010664463043,
+      "rewards/margins": 0.5110014081001282,
+      "rewards/rejected": -0.6423014998435974,
+      "sft_loss": 0.0035836666356772184,
       "step": 30
     },
     {
       "epoch": 0.4819277108433735,
-      "grad_norm": 2.4059877395629883,
+      "grad_norm": 2.372204303741455,
       "learning_rate": 2.6418631827326857e-05,
-      "logits/chosen": -0.26706236600875854,
-      "logits/rejected": -3.2312896251678467,
-      "logps/chosen": -1.2261059284210205,
-      "logps/rejected": -6.600752830505371,
-      "loss": 1.2266,
+      "logits/chosen": -0.2755209803581238,
+      "logits/rejected": -3.2902603149414062,
+      "logps/chosen": -1.2280548810958862,
+      "logps/rejected": -6.585225582122803,
+      "loss": 1.2285,
       "num_input_tokens_seen": 36096,
-      "odds_ratio_loss": 12.158276557922363,
+      "odds_ratio_loss": 12.17691707611084,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.12261058390140533,
-      "rewards/margins": 0.5374647378921509,
-      "rewards/rejected": -0.660075306892395,
-      "sft_loss": 0.010727915912866592,
+      "rewards/chosen": -0.12280547618865967,
+      "rewards/margins": 0.5357170104980469,
+      "rewards/rejected": -0.6585224866867065,
+      "sft_loss": 0.010825484991073608,
       "step": 40
     },
     {
       "epoch": 0.6024096385542169,
-      "grad_norm": 2.4229519367218018,
+      "grad_norm": 2.4067399501800537,
       "learning_rate": 1.70948083275794e-05,
-      "logits/chosen": -0.1631634533405304,
-      "logits/rejected": -3.3860981464385986,
-      "logps/chosen": -1.0198614597320557,
-      "logps/rejected": -6.75305700302124,
-      "loss": 1.0201,
+      "logits/chosen": -0.1695922613143921,
+      "logits/rejected": -3.419903516769409,
+      "logps/chosen": -1.0168521404266357,
+      "logps/rejected": -6.722726345062256,
+      "loss": 1.0171,
       "num_input_tokens_seen": 43456,
-      "odds_ratio_loss": 10.168444633483887,
+      "odds_ratio_loss": 10.138498306274414,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.10198613256216049,
-      "rewards/margins": 0.5733195543289185,
-      "rewards/rejected": -0.6753057241439819,
-      "sft_loss": 0.0032480526715517044,
+      "rewards/chosen": -0.10168520361185074,
+      "rewards/margins": 0.5705875158309937,
+      "rewards/rejected": -0.6722726821899414,
+      "sft_loss": 0.0032351273111999035,
       "step": 50
     },
     {
       "epoch": 0.7228915662650602,
-      "grad_norm": 2.3134255409240723,
+      "grad_norm": 2.4104135036468506,
       "learning_rate": 8.890074238378074e-06,
-      "logits/chosen": -0.19685813784599304,
-      "logits/rejected": -3.2213809490203857,
-      "logps/chosen": -1.4441629648208618,
-      "logps/rejected": -6.606205940246582,
-      "loss": 1.4448,
+      "logits/chosen": -0.2158750295639038,
+      "logits/rejected": -3.261337995529175,
+      "logps/chosen": -1.4426627159118652,
+      "logps/rejected": -6.552022457122803,
+      "loss": 1.4433,
       "num_input_tokens_seen": 51600,
-      "odds_ratio_loss": 14.304969787597656,
+      "odds_ratio_loss": 14.290933609008789,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.14441628754138947,
-      "rewards/margins": 0.5162042379379272,
-      "rewards/rejected": -0.6606205701828003,
-      "sft_loss": 0.014286870136857033,
+      "rewards/chosen": -0.144266277551651,
+      "rewards/margins": 0.5109359622001648,
+      "rewards/rejected": -0.6552022695541382,
+      "sft_loss": 0.01422051526606083,
       "step": 60
     },
     {
       "epoch": 0.8433734939759037,
-      "grad_norm": 2.4275636672973633,
+      "grad_norm": 2.4223577976226807,
       "learning_rate": 2.9659233496337786e-06,
-      "logits/chosen": -0.13393770158290863,
-      "logits/rejected": -3.393664836883545,
-      "logps/chosen": -1.1611616611480713,
-      "logps/rejected": -6.605759620666504,
-      "loss": 1.1615,
+      "logits/chosen": -0.14738118648529053,
+      "logits/rejected": -3.434018611907959,
+      "logps/chosen": -1.1621037721633911,
+      "logps/rejected": -6.5490217208862305,
+      "loss": 1.1624,
       "num_input_tokens_seen": 62624,
-      "odds_ratio_loss": 11.589168548583984,
+      "odds_ratio_loss": 11.598767280578613,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.11611616611480713,
-      "rewards/margins": 0.5444598197937012,
-      "rewards/rejected": -0.6605759859085083,
-      "sft_loss": 0.002552599413320422,
+      "rewards/chosen": -0.11621036380529404,
+      "rewards/margins": 0.5386918783187866,
+      "rewards/rejected": -0.6549022197723389,
+      "sft_loss": 0.0025552159640938044,
       "step": 70
     },
     {
       "epoch": 0.963855421686747,
-      "grad_norm": 1.650820255279541,
+      "grad_norm": 1.6316858530044556,
       "learning_rate": 1.6100130092037703e-07,
-      "logits/chosen": -0.20625996589660645,
-      "logits/rejected": -3.3859100341796875,
-      "logps/chosen": -1.2622536420822144,
-      "logps/rejected": -6.649239540100098,
-      "loss": 1.2626,
+      "logits/chosen": -0.22155144810676575,
+      "logits/rejected": -3.4224212169647217,
+      "logps/chosen": -1.2641386985778809,
+      "logps/rejected": -6.619866371154785,
+      "loss": 1.2645,
       "num_input_tokens_seen": 73184,
-      "odds_ratio_loss": 12.59186840057373,
+      "odds_ratio_loss": 12.610678672790527,
       "rewards/accuracies": 1.0,
-      "rewards/chosen": -0.12622536718845367,
-      "rewards/margins": 0.5386986136436462,
-      "rewards/rejected": -0.664923906326294,
-      "sft_loss": 0.0034584165550768375,
+      "rewards/chosen": -0.12641386687755585,
+      "rewards/margins": 0.5355727672576904,
+      "rewards/rejected": -0.6619867086410522,
+      "sft_loss": 0.003463461296632886,
       "step": 80
     },
     {
@@ -157,10 +157,10 @@
       "num_input_tokens_seen": 75616,
       "step": 83,
       "total_flos": 3443013082939392.0,
-      "train_loss": 1.2913809109883136,
-      "train_runtime": 56.3341,
-      "train_samples_per_second": 1.473,
-      "train_steps_per_second": 1.473
+      "train_loss": 1.2913588443434383,
+      "train_runtime": 65.9669,
+      "train_samples_per_second": 1.258,
+      "train_steps_per_second": 1.258
     }
   ],
   "logging_steps": 10,

Luminia-8B-RP-DPO/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a75126f66edacb9330b4a2c3f30116925026b7803559bfaede779f9523508b96
+oid sha256:0fe52b7fade73dd088cdcd69b8c8544dc5dd90b1cf4621e2065f7741b1c16551
 size 5368

Luminia-8B-RP/adapter_config.json

@@ -1,7 +1,7 @@
 {
   "alpha_pattern": {},
   "auto_mapping": null,
-  "base_model_name_or_path": "NousResearch/Meta-Llama-3.1-8B-Instruct",
+  "base_model_name_or_path": "meta-llama/Meta-Llama-3.1-8B-Instruct",
   "bias": "none",
   "fan_in_fan_out": false,
   "inference_mode": true,
@@ -20,13 +20,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "down_proj",
-    "up_proj",
-    "o_proj",
     "gate_proj",
-    "v_proj",
+    "o_proj",
     "k_proj",
-    "q_proj"
+    "down_proj",
+    "q_proj",
+    "up_proj",
+    "v_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

Luminia-8B-RP/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:069a90de33ea664773a0e1a2ddf3bc448258a3939af91d9700688e41bc266c5f
+oid sha256:f4d4d536113e7ed7303f3655a0294f5b72393f72ce8d11a17ac4b868ac83e1cf
 size 335604696

Luminia-8B-RP/all_results.json

@@ -2,8 +2,8 @@
     "epoch": 1.0,
     "num_input_tokens_seen": 170598400,
     "total_flos": 7.767833833163981e+18,
-    "train_loss": 0.9246458945440359,
-    "train_runtime": 106837.984,
-    "train_samples_per_second": 0.195,
-    "train_steps_per_second": 0.195
+    "train_loss": 0.2516314516834566,
+    "train_runtime": 26805.2364,
+    "train_samples_per_second": 0.777,
+    "train_steps_per_second": 0.777
 }

Luminia-8B-RP/train_results.json

@@ -2,8 +2,8 @@
     "epoch": 1.0,
     "num_input_tokens_seen": 170598400,
     "total_flos": 7.767833833163981e+18,
-    "train_loss": 0.9246458945440359,
-    "train_runtime": 106837.984,
-    "train_samples_per_second": 0.195,
-    "train_steps_per_second": 0.195
+    "train_loss": 0.2516314516834566,
+    "train_runtime": 26805.2364,
+    "train_samples_per_second": 0.777,
+    "train_steps_per_second": 0.777
 }

Luminia-8B-RP/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:647788f6ff8795c8a174c7c1ee498404fbc9e215ce08ad78ab52dbb99c69cf16
-size 5368
+oid sha256:3c4e993a9ca7f914dcb5321f549e6a90cf5f5ce8708dfcfa54902b8f3c7ea7e6
+size 5432

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "NousResearch/Meta-Llama-3.1-8B-Instruct",
+  "_name_or_path": "meta-llama/Meta-Llama-3.1-8B-Instruct",
   "architectures": [
     "LlamaForCausalLM"
   ],

model-00001-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:350e200b9eb52b75184c0575ba28d637cdb53c85bcd631435e6602d31bec8674
+oid sha256:e3c08b21643e94b5ce74a199dd69bec58c068d31a3cab5f08ea62093f0dac2b7
 size 4976698672

model-00002-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:647859e10cd90f805a9da0b9371814573038c5d560fff74f947a751e0e3db0d7
+oid sha256:1e098f89af9d5a9068f6a336957cc648bbd3b6ae96d2f63f52526dea45f5eadd
 size 4999802720

model-00003-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d8b5d5615448401ea58c4a227b141f7d3fa38a6824bb6ba726c8818809e4fc83
+oid sha256:b128dbc253b1cab6210be96bae6070ebe7039f1f50d22e97a75efb59693800e4
 size 4915916176

model-00004-of-00004.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc072beba5702e05eceb1cde3ab0c985452d0dabc29d866965eedd13b337de8a
+oid sha256:527189a1ceebada68fa13b52c6bf6a1c97ac92c9a0de0b76185de397f4399516
 size 1168138808