End of training

README.md

@@ -3,9 +3,15 @@ library_name: transformers
 license: other
 base_model: trl-lib/qwen1.5-0.5b-sft
 tags:
+- alignment-handbook
 - trl
 - simpo
 - generated_from_trainer
+- trl
+- simpo
+- generated_from_trainer
+datasets:
+- yakazimir/ultrafeedback_binarized
 model-index:
 - name: qwen_cpo_entropy_0_3
   results: []
@@ -16,7 +22,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 # qwen_cpo_entropy_0_3
 
-This model is a fine-tuned version of [trl-lib/qwen1.5-0.5b-sft](https://huggingface.co/trl-lib/qwen1.5-0.5b-sft) on an unknown dataset.
+This model is a fine-tuned version of [trl-lib/qwen1.5-0.5b-sft](https://huggingface.co/trl-lib/qwen1.5-0.5b-sft) on the yakazimir/ultrafeedback_binarized dataset.
 It achieves the following results on the evaluation set:
 - Loss: 1.0416
 - Sft Loss: 1.4031

all_results.json

@@ -1,5 +1,19 @@
 {
     "epoch": 2.999297541394882,
+    "eval_logits/chosen": 0.1268635243177414,
+    "eval_logits/rejected": 0.2186587154865265,
+    "eval_logps/chosen": -1.3990017175674438,
+    "eval_logps/rejected": -1.8440245389938354,
+    "eval_loss": 1.0415802001953125,
+    "eval_rewards/accuracies": 0.6157270073890686,
+    "eval_rewards/chosen": -1.3990017175674438,
+    "eval_rewards/margins": 0.44502273201942444,
+    "eval_rewards/rejected": -1.8440245389938354,
+    "eval_runtime": 49.8615,
+    "eval_samples": 1345,
+    "eval_samples_per_second": 26.975,
+    "eval_sft_loss": 1.4031308889389038,
+    "eval_steps_per_second": 6.759,
     "total_flos": 0.0,
     "train_loss": 1.0402103091545567,
     "train_runtime": 33797.3752,

config.json

@@ -22,7 +22,7 @@
   "tie_word_embeddings": true,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.44.2",
-  "use_cache": false,
+  "use_cache": true,
   "use_sliding_window": false,
   "vocab_size": 151648
 }

eval_results.json

@@ -1,17 +1,17 @@
 {
     "epoch": 2.999297541394882,
-    "eval_logits/chosen": -0.47352659702301025,
-    "eval_logits/rejected": -0.42484986782073975,
-    "eval_logps/chosen": -1.535322666168213,
-    "eval_logps/rejected": -2.2016897201538086,
-    "eval_loss": 1.043636679649353,
-    "eval_rewards/accuracies": 0.6476261019706726,
-    "eval_rewards/chosen": -1.535322666168213,
-    "eval_rewards/margins": 0.6663669943809509,
-    "eval_rewards/rejected": -2.2016897201538086,
-    "eval_runtime": 42.6866,
+    "eval_logits/chosen": 0.1268635243177414,
+    "eval_logits/rejected": 0.2186587154865265,
+    "eval_logps/chosen": -1.3990017175674438,
+    "eval_logps/rejected": -1.8440245389938354,
+    "eval_loss": 1.0415802001953125,
+    "eval_rewards/accuracies": 0.6157270073890686,
+    "eval_rewards/chosen": -1.3990017175674438,
+    "eval_rewards/margins": 0.44502273201942444,
+    "eval_rewards/rejected": -1.8440245389938354,
+    "eval_runtime": 49.8615,
     "eval_samples": 1345,
-    "eval_samples_per_second": 31.509,
-    "eval_sft_loss": 1.4855674505233765,
-    "eval_steps_per_second": 7.895
+    "eval_samples_per_second": 26.975,
+    "eval_sft_loss": 1.4031308889389038,
+    "eval_steps_per_second": 6.759
 }