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_qfUNL_entropy
   results: []
@@ -16,7 +22,7 @@ should probably proofread and complete it, then remove this comment. -->
 
 # qwen_qfUNL_entropy
 
-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: 0.6510
 - Rewards/chosen: -1.7989

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 2.999297541394882,
+    "eval_logits/chosen": -0.06038603559136391,
+    "eval_logits/rejected": 0.01924268528819084,
+    "eval_logps/chosen": -1.7989143133163452,
+    "eval_logps/rejected": -2.58301043510437,
+    "eval_loss": 0.6509566903114319,
+    "eval_rewards/accuracies": 0.6735904812812805,
+    "eval_rewards/chosen": -1.7989143133163452,
+    "eval_rewards/margins": 0.7840961217880249,
+    "eval_rewards/rejected": -2.58301043510437,
+    "eval_runtime": 40.198,
+    "eval_samples": 1345,
+    "eval_samples_per_second": 33.459,
+    "eval_steps_per_second": 8.383,
     "total_flos": 0.0,
     "train_loss": 0.6513626721822561,
     "train_runtime": 30253.6444,

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 +1,16 @@
 {
     "epoch": 2.999297541394882,
-    "eval_logits/chosen": NaN,
-    "eval_logits/rejected": NaN,
-    "eval_logps/chosen": NaN,
-    "eval_logps/rejected": NaN,
-    "eval_loss": NaN,
-    "eval_rewards/accuracies": 0.0,
-    "eval_rewards/chosen": NaN,
-    "eval_rewards/margins": NaN,
-    "eval_rewards/rejected": NaN,
-    "eval_runtime": 34.2826,
+    "eval_logits/chosen": -0.06038603559136391,
+    "eval_logits/rejected": 0.01924268528819084,
+    "eval_logps/chosen": -1.7989143133163452,
+    "eval_logps/rejected": -2.58301043510437,
+    "eval_loss": 0.6509566903114319,
+    "eval_rewards/accuracies": 0.6735904812812805,
+    "eval_rewards/chosen": -1.7989143133163452,
+    "eval_rewards/margins": 0.7840961217880249,
+    "eval_rewards/rejected": -2.58301043510437,
+    "eval_runtime": 40.198,
     "eval_samples": 1345,
-    "eval_samples_per_second": 39.233,
-    "eval_semantic_entropy": NaN,
-    "eval_steps_per_second": 9.83
+    "eval_samples_per_second": 33.459,
+    "eval_steps_per_second": 8.383
 }