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_2
+  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_2
+
+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: 0.6580
+- Rewards/chosen: -0.1638
+- Rewards/rejected: -0.2696
+- Rewards/accuracies: 0.5920
+- Rewards/margins: 0.1058
+- Logps/rejected: -289.0013
+- Logps/chosen: -304.2577
+- Logits/rejected: -2.3485
+- Logits/chosen: -2.4282
+
+## 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 |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 0.5679        | 0.64  | 100  | 0.6580          | -0.1638        | -0.2696          | 0.5920             | 0.1058          | -289.0013      | -304.2577    | -2.3485         | -2.4282       |
+
+
+### 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": 0.5971177717049917,
+    "train_runtime": 6956.9181,
+    "train_samples": 20000,
+    "train_samples_per_second": 2.875,
+    "train_steps_per_second": 0.022
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "train_loss": 0.5971177717049917,
+    "train_runtime": 6956.9181,
+    "train_samples": 20000,
+    "train_samples_per_second": 2.875,
+    "train_steps_per_second": 0.022
+}

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": -2.735111713409424,
+      "logits/rejected": -2.6982784271240234,
+      "logps/chosen": -269.989501953125,
+      "logps/rejected": -160.53433227539062,
+      "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": -2.7510337829589844,
+      "logits/rejected": -2.70198392868042,
+      "logps/chosen": -225.9598388671875,
+      "logps/rejected": -146.84454345703125,
+      "loss": 0.6918,
+      "rewards/accuracies": 0.5729166865348816,
+      "rewards/chosen": 0.0024252247530966997,
+      "rewards/margins": 0.0031480363104492426,
+      "rewards/rejected": -0.0007228113245218992,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.989935734988098e-06,
+      "logits/chosen": -2.7219443321228027,
+      "logits/rejected": -2.676025152206421,
+      "logps/chosen": -227.4296112060547,
+      "logps/rejected": -151.27586364746094,
+      "loss": 0.6778,
+      "rewards/accuracies": 0.746874988079071,
+      "rewards/chosen": 0.0456942617893219,
+      "rewards/margins": 0.03698232024908066,
+      "rewards/rejected": 0.00871194526553154,
+      "step": 20
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": -2.6784729957580566,
+      "logits/rejected": -2.6449942588806152,
+      "logps/chosen": -236.11929321289062,
+      "logps/rejected": -164.32357788085938,
+      "loss": 0.649,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": 0.05519301816821098,
+      "rewards/margins": 0.10018298774957657,
+      "rewards/rejected": -0.04498996213078499,
+      "step": 30
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.646121984004666e-06,
+      "logits/chosen": -2.586902141571045,
+      "logits/rejected": -2.5570755004882812,
+      "logps/chosen": -226.29769897460938,
+      "logps/rejected": -173.32815551757812,
+      "loss": 0.6268,
+      "rewards/accuracies": 0.746874988079071,
+      "rewards/chosen": -0.07233832776546478,
+      "rewards/margins": 0.1450045257806778,
+      "rewards/rejected": -0.21734285354614258,
+      "step": 40
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.3069871595684795e-06,
+      "logits/chosen": -2.5778377056121826,
+      "logits/rejected": -2.5205721855163574,
+      "logps/chosen": -245.772705078125,
+      "logps/rejected": -193.9845733642578,
+      "loss": 0.6005,
+      "rewards/accuracies": 0.699999988079071,
+      "rewards/chosen": -0.13813601434230804,
+      "rewards/margins": 0.2017526626586914,
+      "rewards/rejected": -0.33988869190216064,
+      "step": 50
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.8772424536302565e-06,
+      "logits/chosen": -2.4997265338897705,
+      "logits/rejected": -2.4495482444763184,
+      "logps/chosen": -247.75387573242188,
+      "logps/rejected": -203.0021514892578,
+      "loss": 0.5905,
+      "rewards/accuracies": 0.765625,
+      "rewards/chosen": -0.18028748035430908,
+      "rewards/margins": 0.2833089232444763,
+      "rewards/rejected": -0.4635964035987854,
+      "step": 60
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 3.3784370602033572e-06,
+      "logits/chosen": -2.4846343994140625,
+      "logits/rejected": -2.396561861038208,
+      "logps/chosen": -257.1364440917969,
+      "logps/rejected": -203.55770874023438,
+      "loss": 0.5887,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.31378373503685,
+      "rewards/margins": 0.26320117712020874,
+      "rewards/rejected": -0.5769849419593811,
+      "step": 70
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.835583164544139e-06,
+      "logits/chosen": -2.45512318611145,
+      "logits/rejected": -2.375077486038208,
+      "logps/chosen": -242.77249145507812,
+      "logps/rejected": -204.02432250976562,
+      "loss": 0.5853,
+      "rewards/accuracies": 0.734375,
+      "rewards/chosen": -0.2053767442703247,
+      "rewards/margins": 0.27405497431755066,
+      "rewards/rejected": -0.47943171858787537,
+      "step": 80
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.2759017277414165e-06,
+      "logits/chosen": -2.455535888671875,
+      "logits/rejected": -2.434126138687134,
+      "logps/chosen": -274.60833740234375,
+      "logps/rejected": -236.69607543945312,
+      "loss": 0.5714,
+      "rewards/accuracies": 0.7437499761581421,
+      "rewards/chosen": -0.31336939334869385,
+      "rewards/margins": 0.33501023054122925,
+      "rewards/rejected": -0.6483795046806335,
+      "step": 90
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": -2.4610631465911865,
+      "logits/rejected": -2.3715062141418457,
+      "logps/chosen": -265.38323974609375,
+      "logps/rejected": -213.9130859375,
+      "loss": 0.5679,
+      "rewards/accuracies": 0.753125011920929,
+      "rewards/chosen": -0.2271123230457306,
+      "rewards/margins": 0.34284231066703796,
+      "rewards/rejected": -0.5699546337127686,
+      "step": 100
+    },
+    {
+      "epoch": 0.64,
+      "eval_logits/chosen": -2.42817759513855,
+      "eval_logits/rejected": -2.348471164703369,
+      "eval_logps/chosen": -304.2576599121094,
+      "eval_logps/rejected": -289.00128173828125,
+      "eval_loss": 0.6580253839492798,
+      "eval_rewards/accuracies": 0.5920000076293945,
+      "eval_rewards/chosen": -0.16384616494178772,
+      "eval_rewards/margins": 0.10580243170261383,
+      "eval_rewards/rejected": -0.26964858174324036,
+      "eval_runtime": 384.0219,
+      "eval_samples_per_second": 5.208,
+      "eval_steps_per_second": 0.651,
+      "step": 100
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.217751806485235e-06,
+      "logits/chosen": -2.4483482837677,
+      "logits/rejected": -2.3885979652404785,
+      "logps/chosen": -273.8882751464844,
+      "logps/rejected": -224.49832153320312,
+      "loss": 0.557,
+      "rewards/accuracies": 0.762499988079071,
+      "rewards/chosen": -0.25214630365371704,
+      "rewards/margins": 0.3829244375228882,
+      "rewards/rejected": -0.6350707411766052,
+      "step": 110
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 7.723433775328385e-07,
+      "logits/chosen": -2.4284257888793945,
+      "logits/rejected": -2.383587598800659,
+      "logps/chosen": -279.2298889160156,
+      "logps/rejected": -230.93368530273438,
+      "loss": 0.5667,
+      "rewards/accuracies": 0.75,
+      "rewards/chosen": -0.284829318523407,
+      "rewards/margins": 0.3690839409828186,
+      "rewards/rejected": -0.6539133191108704,
+      "step": 120
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 4.1356686569674344e-07,
+      "logits/chosen": -2.4198520183563232,
+      "logits/rejected": -2.323582172393799,
+      "logps/chosen": -245.6083221435547,
+      "logps/rejected": -208.6503448486328,
+      "loss": 0.5768,
+      "rewards/accuracies": 0.703125,
+      "rewards/chosen": -0.2928313910961151,
+      "rewards/margins": 0.2934364080429077,
+      "rewards/rejected": -0.5862677693367004,
+      "step": 130
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.59412823400657e-07,
+      "logits/chosen": -2.3871216773986816,
+      "logits/rejected": -2.3270230293273926,
+      "logps/chosen": -268.95526123046875,
+      "logps/rejected": -234.47000122070312,
+      "loss": 0.5646,
+      "rewards/accuracies": 0.731249988079071,
+      "rewards/chosen": -0.26539891958236694,
+      "rewards/margins": 0.384247362613678,
+      "rewards/rejected": -0.6496462821960449,
+      "step": 140
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.262559558016325e-08,
+      "logits/chosen": -2.404141664505005,
+      "logits/rejected": -2.3580636978149414,
+      "logps/chosen": -244.45816040039062,
+      "logps/rejected": -202.0813751220703,
+      "loss": 0.5598,
+      "rewards/accuracies": 0.784375011920929,
+      "rewards/chosen": -0.26347464323043823,
+      "rewards/margins": 0.37577688694000244,
+      "rewards/rejected": -0.6392515301704407,
+      "step": 150
+    },
+    {
+      "epoch": 1.0,
+      "step": 156,
+      "total_flos": 0.0,
+      "train_loss": 0.5971177717049917,
+      "train_runtime": 6956.9181,
+      "train_samples_per_second": 2.875,
+      "train_steps_per_second": 0.022
+    }
+  ],
+  "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
+}