End of training

README.md

@@ -3,20 +3,32 @@ 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: 5.8367
+- Rewards/chosen: -16.6951
+- Rewards/rejected: -17.6555
+- Rewards/accuracies: 0.5245
+- Rewards/margins: 0.9604
+- Logps/rejected: -1.7656
+- Logps/chosen: -1.6695
+- Logits/rejected: -0.3797
+- Logits/chosen: -0.4422
+- Semantic Entropy: 0.6774
 
 ## Model description
 

all_results.json

@@ -1,5 +1,19 @@
 {
     "epoch": 0.004531679270399638,
+    "eval_logits/chosen": -0.44219690561294556,
+    "eval_logits/rejected": -0.37968531250953674,
+    "eval_logps/chosen": -1.6695083379745483,
+    "eval_logps/rejected": -1.7655524015426636,
+    "eval_loss": 5.836697578430176,
+    "eval_rewards/accuracies": 0.5245413780212402,
+    "eval_rewards/chosen": -16.695083618164062,
+    "eval_rewards/margins": 0.9604394435882568,
+    "eval_rewards/rejected": -17.6555233001709,
+    "eval_runtime": 293.0199,
+    "eval_samples": 5595,
+    "eval_samples_per_second": 19.094,
+    "eval_semantic_entropy": 0.6773539781570435,
+    "eval_steps_per_second": 4.774,
     "total_flos": 0.0,
     "train_loss": 104.54919738769532,
     "train_runtime": 57.8361,

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

@@ -1,16 +1,17 @@
 {
     "epoch": 0.004531679270399638,
-    "eval_logits/chosen": -0.44351547956466675,
-    "eval_logits/rejected": -0.3813219964504242,
-    "eval_logps/chosen": -1.669515609741211,
-    "eval_logps/rejected": -1.7656141519546509,
-    "eval_loss": 5.836363315582275,
-    "eval_rewards/accuracies": 0.5254348516464233,
-    "eval_rewards/chosen": -16.69515609741211,
-    "eval_rewards/margins": 0.9609846472740173,
-    "eval_rewards/rejected": -17.65614128112793,
-    "eval_runtime": 292.82,
+    "eval_logits/chosen": -0.44219690561294556,
+    "eval_logits/rejected": -0.37968531250953674,
+    "eval_logps/chosen": -1.6695083379745483,
+    "eval_logps/rejected": -1.7655524015426636,
+    "eval_loss": 5.836697578430176,
+    "eval_rewards/accuracies": 0.5245413780212402,
+    "eval_rewards/chosen": -16.695083618164062,
+    "eval_rewards/margins": 0.9604394435882568,
+    "eval_rewards/rejected": -17.6555233001709,
+    "eval_runtime": 293.0199,
     "eval_samples": 5595,
-    "eval_samples_per_second": 19.107,
-    "eval_steps_per_second": 4.778
+    "eval_samples_per_second": 19.094,
+    "eval_semantic_entropy": 0.6773539781570435,
+    "eval_steps_per_second": 4.774
 }