End of training

README.md

@@ -1,9 +1,15 @@
 ---
 base_model: Minbyul/llama2-7b-wo-kqa_golden-sft
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: llama2-7b-dpo-full-sft-wo-kqa_golden
   results: []
@@ -14,17 +20,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # llama2-7b-dpo-full-sft-wo-kqa_golden
 
-This model is a fine-tuned version of [Minbyul/llama2-7b-wo-kqa_golden-sft](https://huggingface.co/Minbyul/llama2-7b-wo-kqa_golden-sft) on an unknown dataset.
+This model is a fine-tuned version of [Minbyul/llama2-7b-wo-kqa_golden-sft](https://huggingface.co/Minbyul/llama2-7b-wo-kqa_golden-sft) on the HuggingFaceH4/ultrafeedback_binarized dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.3024
-- Rewards/chosen: -0.0879
-- Rewards/rejected: -1.9222
+- Loss: 0.2778
+- Rewards/chosen: -0.1016
+- Rewards/rejected: -2.1516
 - Rewards/accuracies: 0.9500
-- Rewards/margins: 1.8343
-- Logps/rejected: -748.6945
-- Logps/chosen: -311.0383
-- Logits/rejected: -0.5637
-- Logits/chosen: -0.7827
+- Rewards/margins: 2.0501
+- Logps/rejected: -771.6371
+- Logps/chosen: -312.4064
+- Logits/rejected: -0.5673
+- Logits/chosen: -0.7867
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 1.0,
+    "eval_logits/chosen": -0.786741316318512,
+    "eval_logits/rejected": -0.567284107208252,
+    "eval_logps/chosen": -312.4064025878906,
+    "eval_logps/rejected": -771.6371459960938,
+    "eval_loss": 0.27778080105781555,
+    "eval_rewards/accuracies": 0.949999988079071,
+    "eval_rewards/chosen": -0.10158518701791763,
+    "eval_rewards/margins": 2.0500597953796387,
+    "eval_rewards/rejected": -2.1516449451446533,
+    "eval_runtime": 15.9542,
+    "eval_samples": 152,
+    "eval_samples_per_second": 9.527,
+    "eval_steps_per_second": 0.313,
     "train_loss": 0.3946254751575527,
     "train_runtime": 1910.8543,
     "train_samples": 8578,

config.json

@@ -23,6 +23,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.39.0.dev0",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 32000
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 1.0,
+    "eval_logits/chosen": -0.786741316318512,
+    "eval_logits/rejected": -0.567284107208252,
+    "eval_logps/chosen": -312.4064025878906,
+    "eval_logps/rejected": -771.6371459960938,
+    "eval_loss": 0.27778080105781555,
+    "eval_rewards/accuracies": 0.949999988079071,
+    "eval_rewards/chosen": -0.10158518701791763,
+    "eval_rewards/margins": 2.0500597953796387,
+    "eval_rewards/rejected": -2.1516449451446533,
+    "eval_runtime": 15.9542,
+    "eval_samples": 152,
+    "eval_samples_per_second": 9.527,
+    "eval_steps_per_second": 0.313
+}