Model save

README.md

@@ -0,0 +1,75 @@
+---
+license: apache-2.0
+library_name: peft
+tags:
+- trl
+- dpo
+- generated_from_trainer
+base_model: alignment-handbook/zephyr-7b-sft-full
+model-index:
+- name: nash_dpo_rank4_on_vanilla_iter_1
+  results: []
+---
+
+<!-- This model card has been generated automatically according to the information the Trainer had access to. You
+should probably proofread and complete it, then remove this comment. -->
+
+# nash_dpo_rank4_on_vanilla_iter_1
+
+This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.6753
+- Rewards/chosen: -0.1174
+- Rewards/rejected: -0.1651
+- Rewards/accuracies: 0.6360
+- Rewards/margins: 0.0477
+- Logps/rejected: -298.1530
+- Logps/chosen: -305.5673
+- Logits/rejected: -2.4653
+- Logits/chosen: -2.5565
+
+## Model description
+
+More information needed
+
+## Intended uses & limitations
+
+More information needed
+
+## Training and evaluation data
+
+More information needed
+
+## Training procedure
+
+### Training hyperparameters
+
+The following hyperparameters were used during training:
+- learning_rate: 5e-06
+- train_batch_size: 2
+- eval_batch_size: 2
+- seed: 42
+- distributed_type: multi-GPU
+- num_devices: 4
+- gradient_accumulation_steps: 16
+- total_train_batch_size: 128
+- total_eval_batch_size: 8
+- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
+- lr_scheduler_type: cosine
+- lr_scheduler_warmup_ratio: 0.1
+- num_epochs: 1
+
+### Training results
+
+| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 0.689         | 0.64  | 100  | 0.6753          | -0.1174        | -0.1651          | 0.6360             | 0.0477          | -298.1530      | -305.5673    | -2.4653         | -2.5565       |
+
+
+### Framework versions
+
+- PEFT 0.7.1
+- Transformers 4.36.2
+- Pytorch 2.3.0+cu121
+- Datasets 2.14.6
+- Tokenizers 0.15.2

all_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "train_loss": 0.6885996239307599,
+    "train_runtime": 7326.9096,
+    "train_samples": 20000,
+    "train_samples_per_second": 2.73,
+    "train_steps_per_second": 0.021
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "train_loss": 0.6885996239307599,
+    "train_runtime": 7326.9096,
+    "train_samples": 20000,
+    "train_samples_per_second": 2.73,
+    "train_steps_per_second": 0.021
+}

trainer_state.json

@@ -0,0 +1,270 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9984,
+  "eval_steps": 100,
+  "global_step": 156,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.01,
+      "learning_rate": 3.125e-07,
+      "logits/chosen": -2.7403299808502197,
+      "logits/rejected": -2.7255897521972656,
+      "logps/chosen": -195.21282958984375,
+      "logps/rejected": -184.09413146972656,
+      "loss": 0.6931,
+      "rewards/accuracies": 0.0,
+      "rewards/chosen": 0.0,
+      "rewards/margins": 0.0,
+      "rewards/rejected": 0.0,
+      "step": 1
+    },
+    {
+      "epoch": 0.06,
+      "learning_rate": 3.125e-06,
+      "logits/chosen": -2.6841495037078857,
+      "logits/rejected": -2.682373523712158,
+      "logps/chosen": -163.6064453125,
+      "logps/rejected": -165.0059356689453,
+      "loss": 0.6929,
+      "rewards/accuracies": 0.4756944477558136,
+      "rewards/chosen": -0.008276204578578472,
+      "rewards/margins": 0.001318673719651997,
+      "rewards/rejected": -0.009594877250492573,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.989935734988098e-06,
+      "logits/chosen": -2.692333698272705,
+      "logits/rejected": -2.665407180786133,
+      "logps/chosen": -176.43678283691406,
+      "logps/rejected": -172.32717895507812,
+      "loss": 0.6935,
+      "rewards/accuracies": 0.4906249940395355,
+      "rewards/chosen": -0.06923351436853409,
+      "rewards/margins": 0.003520409809425473,
+      "rewards/rejected": -0.0727539211511612,
+      "step": 20
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": -2.5914735794067383,
+      "logits/rejected": -2.585202932357788,
+      "logps/chosen": -168.06890869140625,
+      "logps/rejected": -169.85023498535156,
+      "loss": 0.6917,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.034951962530612946,
+      "rewards/margins": 0.007884470745921135,
+      "rewards/rejected": -0.04283643513917923,
+      "step": 30
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.646121984004666e-06,
+      "logits/chosen": -2.671989917755127,
+      "logits/rejected": -2.6600332260131836,
+      "logps/chosen": -178.48281860351562,
+      "logps/rejected": -179.00379943847656,
+      "loss": 0.6901,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": -0.01323291938751936,
+      "rewards/margins": 0.006769413594156504,
+      "rewards/rejected": -0.020002331584692,
+      "step": 40
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.3069871595684795e-06,
+      "logits/chosen": -2.6603026390075684,
+      "logits/rejected": -2.6338632106781006,
+      "logps/chosen": -188.98532104492188,
+      "logps/rejected": -187.17684936523438,
+      "loss": 0.6888,
+      "rewards/accuracies": 0.5406249761581421,
+      "rewards/chosen": -0.09863803535699844,
+      "rewards/margins": 0.015061549842357635,
+      "rewards/rejected": -0.11369959264993668,
+      "step": 50
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.8772424536302565e-06,
+      "logits/chosen": -2.6301541328430176,
+      "logits/rejected": -2.6398401260375977,
+      "logps/chosen": -184.1541290283203,
+      "logps/rejected": -188.9746856689453,
+      "loss": 0.6898,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": -0.16303817927837372,
+      "rewards/margins": 0.013844095170497894,
+      "rewards/rejected": -0.1768822818994522,
+      "step": 60
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 3.3784370602033572e-06,
+      "logits/chosen": -2.587550401687622,
+      "logits/rejected": -2.550933361053467,
+      "logps/chosen": -184.53257751464844,
+      "logps/rejected": -183.46517944335938,
+      "loss": 0.6902,
+      "rewards/accuracies": 0.4937500059604645,
+      "rewards/chosen": -0.008703355677425861,
+      "rewards/margins": 0.0026464841794222593,
+      "rewards/rejected": -0.011349838227033615,
+      "step": 70
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.835583164544139e-06,
+      "logits/chosen": -2.56795072555542,
+      "logits/rejected": -2.5637052059173584,
+      "logps/chosen": -176.33071899414062,
+      "logps/rejected": -184.88938903808594,
+      "loss": 0.6906,
+      "rewards/accuracies": 0.578125,
+      "rewards/chosen": 0.0037323920987546444,
+      "rewards/margins": 0.009280242025852203,
+      "rewards/rejected": -0.005547848995774984,
+      "step": 80
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.2759017277414165e-06,
+      "logits/chosen": -2.6137020587921143,
+      "logits/rejected": -2.5704140663146973,
+      "logps/chosen": -187.82107543945312,
+      "logps/rejected": -184.28587341308594,
+      "loss": 0.6883,
+      "rewards/accuracies": 0.5375000238418579,
+      "rewards/chosen": -0.08189593255519867,
+      "rewards/margins": 0.0156090734526515,
+      "rewards/rejected": -0.09750500321388245,
+      "step": 90
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": -2.6152119636535645,
+      "logits/rejected": -2.640665292739868,
+      "logps/chosen": -176.8973388671875,
+      "logps/rejected": -184.32557678222656,
+      "loss": 0.689,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": -0.08713572472333908,
+      "rewards/margins": 0.014782750979065895,
+      "rewards/rejected": -0.10191845893859863,
+      "step": 100
+    },
+    {
+      "epoch": 0.64,
+      "eval_logits/chosen": -2.556511878967285,
+      "eval_logits/rejected": -2.465257406234741,
+      "eval_logps/chosen": -305.5672607421875,
+      "eval_logps/rejected": -298.1529541015625,
+      "eval_loss": 0.6753088235855103,
+      "eval_rewards/accuracies": 0.6359999775886536,
+      "eval_rewards/chosen": -0.11736557632684708,
+      "eval_rewards/margins": 0.04772435873746872,
+      "eval_rewards/rejected": -0.1650899201631546,
+      "eval_runtime": 384.6543,
+      "eval_samples_per_second": 5.199,
+      "eval_steps_per_second": 0.65,
+      "step": 100
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.217751806485235e-06,
+      "logits/chosen": -2.62018084526062,
+      "logits/rejected": -2.610757350921631,
+      "logps/chosen": -171.35316467285156,
+      "logps/rejected": -176.16299438476562,
+      "loss": 0.6845,
+      "rewards/accuracies": 0.534375011920929,
+      "rewards/chosen": -0.05986649543046951,
+      "rewards/margins": 0.020178433507680893,
+      "rewards/rejected": -0.0800449326634407,
+      "step": 110
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 7.723433775328385e-07,
+      "logits/chosen": -2.548755168914795,
+      "logits/rejected": -2.529273271560669,
+      "logps/chosen": -173.8794403076172,
+      "logps/rejected": -172.54251098632812,
+      "loss": 0.6854,
+      "rewards/accuracies": 0.559374988079071,
+      "rewards/chosen": -0.06846034526824951,
+      "rewards/margins": 0.019297173246741295,
+      "rewards/rejected": -0.08775752037763596,
+      "step": 120
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 4.1356686569674344e-07,
+      "logits/chosen": -2.6056151390075684,
+      "logits/rejected": -2.5934414863586426,
+      "logps/chosen": -181.0892791748047,
+      "logps/rejected": -185.13357543945312,
+      "loss": 0.686,
+      "rewards/accuracies": 0.5562499761581421,
+      "rewards/chosen": -0.055159904062747955,
+      "rewards/margins": 0.01887853816151619,
+      "rewards/rejected": -0.07403843104839325,
+      "step": 130
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.59412823400657e-07,
+      "logits/chosen": -2.592637538909912,
+      "logits/rejected": -2.583139657974243,
+      "logps/chosen": -176.57772827148438,
+      "logps/rejected": -182.9418182373047,
+      "loss": 0.6847,
+      "rewards/accuracies": 0.581250011920929,
+      "rewards/chosen": -0.04002287983894348,
+      "rewards/margins": 0.02553923986852169,
+      "rewards/rejected": -0.06556212902069092,
+      "step": 140
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.262559558016325e-08,
+      "logits/chosen": -2.5700881481170654,
+      "logits/rejected": -2.561204433441162,
+      "logps/chosen": -178.7554473876953,
+      "logps/rejected": -186.81182861328125,
+      "loss": 0.6849,
+      "rewards/accuracies": 0.503125011920929,
+      "rewards/chosen": -0.05323126167058945,
+      "rewards/margins": 0.007863727398216724,
+      "rewards/rejected": -0.0610949881374836,
+      "step": 150
+    },
+    {
+      "epoch": 1.0,
+      "step": 156,
+      "total_flos": 0.0,
+      "train_loss": 0.6885996239307599,
+      "train_runtime": 7326.9096,
+      "train_samples_per_second": 2.73,
+      "train_steps_per_second": 0.021
+    }
+  ],
+  "logging_steps": 10,
+  "max_steps": 156,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 100,
+  "total_flos": 0.0,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}