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_cCPO_entropy
   results: []
@@ -16,9 +22,9 @@ should probably proofread and complete it, then remove this comment. -->
 
 # qwen_cCPO_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: 0.5210
+- Loss: 0.5209
 - Rewards/chosen: -9.5408
 - Rewards/rejected: -10.6523
 - Rewards/accuracies: 0.7166

all_results.json

@@ -1,5 +1,19 @@
 {
     "epoch": 2.999297541394882,
+    "eval_logits/chosen": 0.3541771471500397,
+    "eval_logits/rejected": 0.4142882525920868,
+    "eval_logps/chosen": -9.540809631347656,
+    "eval_logps/rejected": -10.652300834655762,
+    "eval_loss": 0.5209499597549438,
+    "eval_rewards/accuracies": 0.716617226600647,
+    "eval_rewards/chosen": -9.540809631347656,
+    "eval_rewards/margins": 1.111490249633789,
+    "eval_rewards/rejected": -10.652300834655762,
+    "eval_runtime": 34.2971,
+    "eval_samples": 1345,
+    "eval_samples_per_second": 39.216,
+    "eval_semantic_entropy": 0.0024765790440142155,
+    "eval_steps_per_second": 9.826,
     "total_flos": 0.0,
     "train_loss": 0.5415812020936581,
     "train_runtime": 28601.8206,

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

@@ -0,0 +1,17 @@
+{
+    "epoch": 2.999297541394882,
+    "eval_logits/chosen": 0.3541771471500397,
+    "eval_logits/rejected": 0.4142882525920868,
+    "eval_logps/chosen": -9.540809631347656,
+    "eval_logps/rejected": -10.652300834655762,
+    "eval_loss": 0.5209499597549438,
+    "eval_rewards/accuracies": 0.716617226600647,
+    "eval_rewards/chosen": -9.540809631347656,
+    "eval_rewards/margins": 1.111490249633789,
+    "eval_rewards/rejected": -10.652300834655762,
+    "eval_runtime": 34.2971,
+    "eval_samples": 1345,
+    "eval_samples_per_second": 39.216,
+    "eval_semantic_entropy": 0.0024765790440142155,
+    "eval_steps_per_second": 9.826
+}