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_simple_online_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_simple_online_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: 0.6782
+- Rewards/chosen: -0.0686
+- Rewards/rejected: -0.0971
+- Rewards/accuracies: 0.6100
+- Rewards/margins: 0.0284
+- Logps/rejected: -268.4390
+- Logps/chosen: -288.5429
+- Logits/rejected: -2.5385
+- Logits/chosen: -2.6271
+
+## 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 |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 0.6886        | 0.64  | 100  | 0.6782          | -0.0686        | -0.0971          | 0.6100             | 0.0284          | -268.4390      | -288.5429    | -2.5385         | -2.6271       |
+
+
+### 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": 0.6890946939969674,
+    "train_runtime": 6305.2604,
+    "train_samples": 20000,
+    "train_samples_per_second": 3.172,
+    "train_steps_per_second": 0.025
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "epoch": 1.0,
+    "train_loss": 0.6890946939969674,
+    "train_runtime": 6305.2604,
+    "train_samples": 20000,
+    "train_samples_per_second": 3.172,
+    "train_steps_per_second": 0.025
+}

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": -2.72961163520813,
+      "logits/rejected": -2.7527058124542236,
+      "logps/chosen": -133.97433471679688,
+      "logps/rejected": -138.8169403076172,
+      "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": -2.7357263565063477,
+      "logits/rejected": -2.7272207736968994,
+      "logps/chosen": -115.04521942138672,
+      "logps/rejected": -114.19779205322266,
+      "loss": 0.6932,
+      "rewards/accuracies": 0.3888888955116272,
+      "rewards/chosen": 0.0004528095596469939,
+      "rewards/margins": -0.000532312027644366,
+      "rewards/rejected": 0.00098512158729136,
+      "step": 10
+    },
+    {
+      "epoch": 0.13,
+      "learning_rate": 4.989935734988098e-06,
+      "logits/chosen": -2.721278667449951,
+      "logits/rejected": -2.7116167545318604,
+      "logps/chosen": -111.3840103149414,
+      "logps/rejected": -116.3367691040039,
+      "loss": 0.6927,
+      "rewards/accuracies": 0.5218750238418579,
+      "rewards/chosen": -0.014203068800270557,
+      "rewards/margins": 0.0006986708613112569,
+      "rewards/rejected": -0.014901740476489067,
+      "step": 20
+    },
+    {
+      "epoch": 0.19,
+      "learning_rate": 4.8776412907378845e-06,
+      "logits/chosen": -2.742332696914673,
+      "logits/rejected": -2.7299842834472656,
+      "logps/chosen": -123.17195129394531,
+      "logps/rejected": -122.3455581665039,
+      "loss": 0.6925,
+      "rewards/accuracies": 0.49687498807907104,
+      "rewards/chosen": -0.09704665839672089,
+      "rewards/margins": 0.00010085676331073046,
+      "rewards/rejected": -0.09714751690626144,
+      "step": 30
+    },
+    {
+      "epoch": 0.26,
+      "learning_rate": 4.646121984004666e-06,
+      "logits/chosen": -2.700500965118408,
+      "logits/rejected": -2.673189878463745,
+      "logps/chosen": -116.55732727050781,
+      "logps/rejected": -118.69517517089844,
+      "loss": 0.6924,
+      "rewards/accuracies": 0.49687498807907104,
+      "rewards/chosen": -0.08498911559581757,
+      "rewards/margins": 5.4714873840566725e-05,
+      "rewards/rejected": -0.08504383265972137,
+      "step": 40
+    },
+    {
+      "epoch": 0.32,
+      "learning_rate": 4.3069871595684795e-06,
+      "logits/chosen": -2.740999698638916,
+      "logits/rejected": -2.711369514465332,
+      "logps/chosen": -111.51325988769531,
+      "logps/rejected": -118.1547622680664,
+      "loss": 0.6904,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": -0.007794947363436222,
+      "rewards/margins": 0.005682565737515688,
+      "rewards/rejected": -0.013477511703968048,
+      "step": 50
+    },
+    {
+      "epoch": 0.38,
+      "learning_rate": 3.8772424536302565e-06,
+      "logits/chosen": -2.682302236557007,
+      "logits/rejected": -2.660250425338745,
+      "logps/chosen": -119.29142761230469,
+      "logps/rejected": -123.52491760253906,
+      "loss": 0.6896,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.05805445462465286,
+      "rewards/margins": 0.008577173575758934,
+      "rewards/rejected": -0.06663163006305695,
+      "step": 60
+    },
+    {
+      "epoch": 0.45,
+      "learning_rate": 3.3784370602033572e-06,
+      "logits/chosen": -2.6593658924102783,
+      "logits/rejected": -2.635847568511963,
+      "logps/chosen": -110.70475769042969,
+      "logps/rejected": -111.5876693725586,
+      "loss": 0.6896,
+      "rewards/accuracies": 0.578125,
+      "rewards/chosen": -0.05531386658549309,
+      "rewards/margins": 0.013766427524387836,
+      "rewards/rejected": -0.06908029317855835,
+      "step": 70
+    },
+    {
+      "epoch": 0.51,
+      "learning_rate": 2.835583164544139e-06,
+      "logits/chosen": -2.6447205543518066,
+      "logits/rejected": -2.6153109073638916,
+      "logps/chosen": -116.82215881347656,
+      "logps/rejected": -122.38499450683594,
+      "loss": 0.6888,
+      "rewards/accuracies": 0.550000011920929,
+      "rewards/chosen": -0.044131264090538025,
+      "rewards/margins": 0.011304137296974659,
+      "rewards/rejected": -0.05543540045619011,
+      "step": 80
+    },
+    {
+      "epoch": 0.58,
+      "learning_rate": 2.2759017277414165e-06,
+      "logits/chosen": -2.6791300773620605,
+      "logits/rejected": -2.670039653778076,
+      "logps/chosen": -137.33778381347656,
+      "logps/rejected": -134.17361450195312,
+      "loss": 0.6871,
+      "rewards/accuracies": 0.559374988079071,
+      "rewards/chosen": -0.0856749638915062,
+      "rewards/margins": 0.015940625220537186,
+      "rewards/rejected": -0.10161559283733368,
+      "step": 90
+    },
+    {
+      "epoch": 0.64,
+      "learning_rate": 1.7274575140626318e-06,
+      "logits/chosen": -2.6628499031066895,
+      "logits/rejected": -2.6339688301086426,
+      "logps/chosen": -124.69636535644531,
+      "logps/rejected": -118.883544921875,
+      "loss": 0.6886,
+      "rewards/accuracies": 0.543749988079071,
+      "rewards/chosen": -0.03587063401937485,
+      "rewards/margins": 0.011753683909773827,
+      "rewards/rejected": -0.047624316066503525,
+      "step": 100
+    },
+    {
+      "epoch": 0.64,
+      "eval_logits/chosen": -2.6271259784698486,
+      "eval_logits/rejected": -2.5384714603424072,
+      "eval_logps/chosen": -288.5429382324219,
+      "eval_logps/rejected": -268.43902587890625,
+      "eval_loss": 0.678156852722168,
+      "eval_rewards/accuracies": 0.6100000143051147,
+      "eval_rewards/chosen": -0.06864660233259201,
+      "eval_rewards/margins": 0.02844993770122528,
+      "eval_rewards/rejected": -0.0970965251326561,
+      "eval_runtime": 383.7648,
+      "eval_samples_per_second": 5.212,
+      "eval_steps_per_second": 0.651,
+      "step": 100
+    },
+    {
+      "epoch": 0.7,
+      "learning_rate": 1.217751806485235e-06,
+      "logits/chosen": -2.6874401569366455,
+      "logits/rejected": -2.6627275943756104,
+      "logps/chosen": -120.39668273925781,
+      "logps/rejected": -122.85832214355469,
+      "loss": 0.6876,
+      "rewards/accuracies": 0.5531250238418579,
+      "rewards/chosen": -0.04335067793726921,
+      "rewards/margins": 0.012038113549351692,
+      "rewards/rejected": -0.055388789623975754,
+      "step": 110
+    },
+    {
+      "epoch": 0.77,
+      "learning_rate": 7.723433775328385e-07,
+      "logits/chosen": -2.6583964824676514,
+      "logits/rejected": -2.6295745372772217,
+      "logps/chosen": -124.8127670288086,
+      "logps/rejected": -123.85284423828125,
+      "loss": 0.6869,
+      "rewards/accuracies": 0.6156250238418579,
+      "rewards/chosen": -0.060910262167453766,
+      "rewards/margins": 0.01821967028081417,
+      "rewards/rejected": -0.07912993431091309,
+      "step": 120
+    },
+    {
+      "epoch": 0.83,
+      "learning_rate": 4.1356686569674344e-07,
+      "logits/chosen": -2.6148152351379395,
+      "logits/rejected": -2.5909037590026855,
+      "logps/chosen": -116.5543212890625,
+      "logps/rejected": -121.72233581542969,
+      "loss": 0.6867,
+      "rewards/accuracies": 0.53125,
+      "rewards/chosen": -0.08463772386312485,
+      "rewards/margins": 0.020932147279381752,
+      "rewards/rejected": -0.10556988418102264,
+      "step": 130
+    },
+    {
+      "epoch": 0.9,
+      "learning_rate": 1.59412823400657e-07,
+      "logits/chosen": -2.605799674987793,
+      "logits/rejected": -2.5549862384796143,
+      "logps/chosen": -116.57108306884766,
+      "logps/rejected": -122.12138366699219,
+      "loss": 0.6885,
+      "rewards/accuracies": 0.5718749761581421,
+      "rewards/chosen": -0.05592598766088486,
+      "rewards/margins": 0.015123754739761353,
+      "rewards/rejected": -0.07104974240064621,
+      "step": 140
+    },
+    {
+      "epoch": 0.96,
+      "learning_rate": 2.262559558016325e-08,
+      "logits/chosen": -2.643202066421509,
+      "logits/rejected": -2.605377674102783,
+      "logps/chosen": -114.8644027709961,
+      "logps/rejected": -112.22537994384766,
+      "loss": 0.6854,
+      "rewards/accuracies": 0.5843750238418579,
+      "rewards/chosen": -0.0501478835940361,
+      "rewards/margins": 0.02834610641002655,
+      "rewards/rejected": -0.07849399000406265,
+      "step": 150
+    },
+    {
+      "epoch": 1.0,
+      "step": 156,
+      "total_flos": 0.0,
+      "train_loss": 0.6890946939969674,
+      "train_runtime": 6305.2604,
+      "train_samples_per_second": 3.172,
+      "train_steps_per_second": 0.025
+    }
+  ],
+  "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
+}