End of training

README.md

@@ -3,20 +3,31 @@ library_name: transformers
 license: other
 base_model: trl-lib/qwen1.5-0.5b-sft
 tags:
+- alignment-handbook
 - trl
 - simpo
 - generated_from_trainer
 model-index:
-- name: simpo-exps_qwen05b
+- name: trl-lib/qwen1.5-0.5b-sft
   results: []
 ---
 
 <!-- This model card has been generated automatically according to the information the Trainer had access to. You
 should probably proofread and complete it, then remove this comment. -->
 
-# simpo-exps_qwen05b
+# trl-lib/qwen1.5-0.5b-sft
 
 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.
+It achieves the following results on the evaluation set:
+- Loss: 5.8350
+- Rewards/chosen: -16.6895
+- Rewards/rejected: -17.6490
+- Rewards/accuracies: 0.5245
+- Rewards/margins: 0.9596
+- Logps/rejected: -1.7649
+- Logps/chosen: -1.6689
+- Logits/rejected: -0.3783
+- Logits/chosen: -0.4409
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 0.64,
+    "eval_logits/chosen": -0.4409240782260895,
+    "eval_logits/rejected": -0.3783179819583893,
+    "eval_logps/chosen": -1.6689454317092896,
+    "eval_logps/rejected": -1.7649023532867432,
+    "eval_loss": 5.835046768188477,
+    "eval_rewards/accuracies": 0.5245413780212402,
+    "eval_rewards/chosen": -16.689455032348633,
+    "eval_rewards/margins": 0.9595676064491272,
+    "eval_rewards/rejected": -17.649023056030273,
+    "eval_runtime": 294.9751,
+    "eval_samples": 5595,
+    "eval_samples_per_second": 18.968,
+    "eval_steps_per_second": 4.743,
     "total_flos": 0.0,
     "train_loss": 98.40630645751953,
     "train_runtime": 107.1933,

config.json

@@ -24,7 +24,7 @@
   "tie_word_embeddings": true,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.48.3",
-  "use_cache": false,
+  "use_cache": true,
   "use_sliding_window": false,
   "vocab_size": 151648
 }

eval_results.json

@@ -1,17 +1,16 @@
 {
-    "epoch": 0.009063358540799276,
-    "eval_logits/chosen": -0.4422663152217865,
-    "eval_logits/rejected": -0.3794980049133301,
-    "eval_logps/chosen": -1.6693823337554932,
-    "eval_logps/rejected": -1.7654027938842773,
-    "eval_loss": 5.836271286010742,
-    "eval_rewards/accuracies": 0.5248987674713135,
-    "eval_rewards/chosen": -16.69382095336914,
-    "eval_rewards/margins": 0.960205614566803,
-    "eval_rewards/rejected": -17.654029846191406,
-    "eval_runtime": 294.0455,
+    "epoch": 0.64,
+    "eval_logits/chosen": -0.4409240782260895,
+    "eval_logits/rejected": -0.3783179819583893,
+    "eval_logps/chosen": -1.6689454317092896,
+    "eval_logps/rejected": -1.7649023532867432,
+    "eval_loss": 5.835046768188477,
+    "eval_rewards/accuracies": 0.5245413780212402,
+    "eval_rewards/chosen": -16.689455032348633,
+    "eval_rewards/margins": 0.9595676064491272,
+    "eval_rewards/rejected": -17.649023056030273,
+    "eval_runtime": 294.9751,
     "eval_samples": 5595,
-    "eval_samples_per_second": 19.028,
-    "eval_semantic_entropy": 0.677388608455658,
-    "eval_steps_per_second": 4.758
+    "eval_samples_per_second": 18.968,
+    "eval_steps_per_second": 4.743
 }