Model save

README.md

@@ -0,0 +1,66 @@
+---
+library_name: transformers
+model_name: gemma-7b-sft-silvia_simpo-basic-5e-7-005-v144
+tags:
+- generated_from_trainer
+- trl
+- orpo
+licence: license
+---
+
+# Model Card for gemma-7b-sft-silvia_simpo-basic-5e-7-005-v144
+
+This model is a fine-tuned version of [None](https://huggingface.co/None).
+It has been trained using [TRL](https://github.com/huggingface/trl).
+
+## Quick start
+
+```python
+from transformers import pipeline
+
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="silviasapora/gemma-7b-sft-silvia_simpo-basic-5e-7-005-v144", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
+
+## Training procedure
+
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/silvias/huggingface/runs/detfchne) 
+
+
+This model was trained with ORPO, a method introduced in [ORPO: Monolithic Preference Optimization without Reference Model](https://huggingface.co/papers/2403.07691).
+
+### Framework versions
+
+- TRL: 0.15.2
+- Transformers: 4.49.0
+- Pytorch: 2.5.1
+- Datasets: 3.1.0
+- Tokenizers: 0.21.1
+
+## Citations
+
+Cite ORPO as:
+
+```bibtex
+@article{hong2024orpo,
+    title        = {{ORPO: Monolithic Preference Optimization without Reference Model}},
+    author       = {Jiwoo Hong and Noah Lee and James Thorne},
+    year         = 2024,
+    eprint       = {arXiv:2403.07691}
+}
+```
+
+Cite TRL as:
+    
+```bibtex
+@misc{vonwerra2022trl,
+	title        = {{TRL: Transformer Reinforcement Learning}},
+	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
+	year         = 2020,
+	journal      = {GitHub repository},
+	publisher    = {GitHub},
+	howpublished = {\url{https://github.com/huggingface/trl}}
+}
+```

all_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 0.9885148514851485,
+    "total_flos": 0.0,
+    "train_loss": 4.754565581297263,
+    "train_runtime": 2713.8691,
+    "train_samples": 5050,
+    "train_samples_per_second": 1.861,
+    "train_steps_per_second": 0.029
+}

train_results.json

@@ -0,0 +1,9 @@
+{
+    "epoch": 0.9885148514851485,
+    "total_flos": 0.0,
+    "train_loss": 4.754565581297263,
+    "train_runtime": 2713.8691,
+    "train_samples": 5050,
+    "train_samples_per_second": 1.861,
+    "train_steps_per_second": 0.029
+}

trainer_state.json

@@ -0,0 +1,282 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9885148514851485,
+  "eval_steps": 500,
+  "global_step": 78,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.06336633663366337,
+      "grad_norm": 12.8125,
+      "learning_rate": 3.1249999999999997e-07,
+      "log_odds_chosen": 0.9911565780639648,
+      "log_odds_ratio": -0.41253289580345154,
+      "logps/chosen": -0.6660582423210144,
+      "logps/rejected": -1.228592872619629,
+      "loss": 4.3637,
+      "nll_loss": 0.9886282086372375,
+      "rewards/accuracies": 0.828125,
+      "rewards/chosen": -0.0333029143512249,
+      "rewards/margins": 0.028126735240221024,
+      "rewards/rejected": -0.061429642140865326,
+      "step": 5
+    },
+    {
+      "epoch": 0.12673267326732673,
+      "grad_norm": 17.875,
+      "learning_rate": 4.989935734988097e-07,
+      "log_odds_chosen": 1.1546833515167236,
+      "log_odds_ratio": -0.37236329913139343,
+      "logps/chosen": -0.5902147889137268,
+      "logps/rejected": -1.2310830354690552,
+      "loss": 4.5589,
+      "nll_loss": 0.8954709768295288,
+      "rewards/accuracies": 0.846875011920929,
+      "rewards/chosen": -0.02951074205338955,
+      "rewards/margins": 0.03204340860247612,
+      "rewards/rejected": -0.06155414506793022,
+      "step": 10
+    },
+    {
+      "epoch": 0.1900990099009901,
+      "grad_norm": 15.0,
+      "learning_rate": 4.877641290737883e-07,
+      "log_odds_chosen": 1.1432024240493774,
+      "log_odds_ratio": -0.3701472878456116,
+      "logps/chosen": -0.5654497146606445,
+      "logps/rejected": -1.1885969638824463,
+      "loss": 4.5936,
+      "nll_loss": 0.875917911529541,
+      "rewards/accuracies": 0.8843749761581421,
+      "rewards/chosen": -0.028272485360503197,
+      "rewards/margins": 0.03115735575556755,
+      "rewards/rejected": -0.0594298429787159,
+      "step": 15
+    },
+    {
+      "epoch": 0.25346534653465347,
+      "grad_norm": 15.1875,
+      "learning_rate": 4.646121984004665e-07,
+      "log_odds_chosen": 1.0618789196014404,
+      "log_odds_ratio": -0.38934561610221863,
+      "logps/chosen": -0.6174649000167847,
+      "logps/rejected": -1.2051403522491455,
+      "loss": 4.4157,
+      "nll_loss": 0.937028706073761,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -0.030873248353600502,
+      "rewards/margins": 0.029383767396211624,
+      "rewards/rejected": -0.06025701016187668,
+      "step": 20
+    },
+    {
+      "epoch": 0.31683168316831684,
+      "grad_norm": 16.5,
+      "learning_rate": 4.3069871595684787e-07,
+      "log_odds_chosen": 1.2300812005996704,
+      "log_odds_ratio": -0.3776150345802307,
+      "logps/chosen": -0.6107009649276733,
+      "logps/rejected": -1.3069515228271484,
+      "loss": 4.4461,
+      "nll_loss": 0.8909252882003784,
+      "rewards/accuracies": 0.856249988079071,
+      "rewards/chosen": -0.030535047873854637,
+      "rewards/margins": 0.03481253236532211,
+      "rewards/rejected": -0.0653475821018219,
+      "step": 25
+    },
+    {
+      "epoch": 0.3801980198019802,
+      "grad_norm": 16.25,
+      "learning_rate": 3.877242453630256e-07,
+      "log_odds_chosen": 1.3746024370193481,
+      "log_odds_ratio": -0.3285817801952362,
+      "logps/chosen": -0.5757168531417847,
+      "logps/rejected": -1.3395036458969116,
+      "loss": 4.232,
+      "nll_loss": 0.8684650659561157,
+      "rewards/accuracies": 0.890625,
+      "rewards/chosen": -0.02878584899008274,
+      "rewards/margins": 0.03818933293223381,
+      "rewards/rejected": -0.0669751763343811,
+      "step": 30
+    },
+    {
+      "epoch": 0.44356435643564357,
+      "grad_norm": 17.75,
+      "learning_rate": 3.378437060203357e-07,
+      "log_odds_chosen": 1.5565850734710693,
+      "log_odds_ratio": -0.28478947281837463,
+      "logps/chosen": -0.5244386196136475,
+      "logps/rejected": -1.3875908851623535,
+      "loss": 4.7516,
+      "nll_loss": 0.796803891658783,
+      "rewards/accuracies": 0.934374988079071,
+      "rewards/chosen": -0.026221930980682373,
+      "rewards/margins": 0.04315761476755142,
+      "rewards/rejected": -0.0693795457482338,
+      "step": 35
+    },
+    {
+      "epoch": 0.5069306930693069,
+      "grad_norm": 21.625,
+      "learning_rate": 2.8355831645441387e-07,
+      "log_odds_chosen": 1.7404935359954834,
+      "log_odds_ratio": -0.24158628284931183,
+      "logps/chosen": -0.4815496802330017,
+      "logps/rejected": -1.4325745105743408,
+      "loss": 5.0552,
+      "nll_loss": 0.7471240758895874,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.024077486246824265,
+      "rewards/margins": 0.047551244497299194,
+      "rewards/rejected": -0.07162873446941376,
+      "step": 40
+    },
+    {
+      "epoch": 0.5702970297029702,
+      "grad_norm": 20.125,
+      "learning_rate": 2.2759017277414164e-07,
+      "log_odds_chosen": 1.6410064697265625,
+      "log_odds_ratio": -0.26414045691490173,
+      "logps/chosen": -0.5090475678443909,
+      "logps/rejected": -1.4282280206680298,
+      "loss": 4.6016,
+      "nll_loss": 0.7733598947525024,
+      "rewards/accuracies": 0.9468749761581421,
+      "rewards/chosen": -0.025452375411987305,
+      "rewards/margins": 0.04595901817083359,
+      "rewards/rejected": -0.07141139358282089,
+      "step": 45
+    },
+    {
+      "epoch": 0.6336633663366337,
+      "grad_norm": 26.5,
+      "learning_rate": 1.7274575140626315e-07,
+      "log_odds_chosen": 1.7832257747650146,
+      "log_odds_ratio": -0.23228943347930908,
+      "logps/chosen": -0.46333685517311096,
+      "logps/rejected": -1.3944836854934692,
+      "loss": 5.7573,
+      "nll_loss": 0.709271252155304,
+      "rewards/accuracies": 0.953125,
+      "rewards/chosen": -0.023166844621300697,
+      "rewards/margins": 0.046557340770959854,
+      "rewards/rejected": -0.0697241872549057,
+      "step": 50
+    },
+    {
+      "epoch": 0.697029702970297,
+      "grad_norm": 24.125,
+      "learning_rate": 1.2177518064852348e-07,
+      "log_odds_chosen": 1.5008140802383423,
+      "log_odds_ratio": -0.30259355902671814,
+      "logps/chosen": -0.5552427172660828,
+      "logps/rejected": -1.3832123279571533,
+      "loss": 5.0102,
+      "nll_loss": 0.7843067646026611,
+      "rewards/accuracies": 0.9125000238418579,
+      "rewards/chosen": -0.027762139216065407,
+      "rewards/margins": 0.041398484259843826,
+      "rewards/rejected": -0.06916062533855438,
+      "step": 55
+    },
+    {
+      "epoch": 0.7603960396039604,
+      "grad_norm": 26.25,
+      "learning_rate": 7.723433775328384e-08,
+      "log_odds_chosen": 1.6810848712921143,
+      "log_odds_ratio": -0.2639027237892151,
+      "logps/chosen": -0.5468909740447998,
+      "logps/rejected": -1.5102508068084717,
+      "loss": 5.0576,
+      "nll_loss": 0.787597119808197,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.02734454534947872,
+      "rewards/margins": 0.04816799610853195,
+      "rewards/rejected": -0.07551254332065582,
+      "step": 60
+    },
+    {
+      "epoch": 0.8237623762376237,
+      "grad_norm": 26.25,
+      "learning_rate": 4.1356686569674335e-08,
+      "log_odds_chosen": 1.6594898700714111,
+      "log_odds_ratio": -0.2648767828941345,
+      "logps/chosen": -0.5453223586082458,
+      "logps/rejected": -1.4675935506820679,
+      "loss": 5.2911,
+      "nll_loss": 0.8060859441757202,
+      "rewards/accuracies": 0.9312499761581421,
+      "rewards/chosen": -0.0272661205381155,
+      "rewards/margins": 0.04611356183886528,
+      "rewards/rejected": -0.07337968051433563,
+      "step": 65
+    },
+    {
+      "epoch": 0.8871287128712871,
+      "grad_norm": 22.25,
+      "learning_rate": 1.5941282340065697e-08,
+      "log_odds_chosen": 1.8507673740386963,
+      "log_odds_ratio": -0.22900144755840302,
+      "logps/chosen": -0.500530481338501,
+      "logps/rejected": -1.5389080047607422,
+      "loss": 4.521,
+      "nll_loss": 0.7640606164932251,
+      "rewards/accuracies": 0.956250011920929,
+      "rewards/chosen": -0.02502652443945408,
+      "rewards/margins": 0.0519188828766346,
+      "rewards/rejected": -0.07694540917873383,
+      "step": 70
+    },
+    {
+      "epoch": 0.9504950495049505,
+      "grad_norm": 30.375,
+      "learning_rate": 2.2625595580163247e-09,
+      "log_odds_chosen": 1.575315237045288,
+      "log_odds_ratio": -0.2773153781890869,
+      "logps/chosen": -0.558388352394104,
+      "logps/rejected": -1.4570810794830322,
+      "loss": 4.7645,
+      "nll_loss": 0.8240699768066406,
+      "rewards/accuracies": 0.9375,
+      "rewards/chosen": -0.02791941724717617,
+      "rewards/margins": 0.04493463411927223,
+      "rewards/rejected": -0.07285405695438385,
+      "step": 75
+    },
+    {
+      "epoch": 0.9885148514851485,
+      "step": 78,
+      "total_flos": 0.0,
+      "train_loss": 4.754565581297263,
+      "train_runtime": 2713.8691,
+      "train_samples_per_second": 1.861,
+      "train_steps_per_second": 0.029
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 78,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 1,
+  "save_steps": 300,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 1,
+  "trial_name": null,
+  "trial_params": null
+}