End of training

README.md

@@ -2,10 +2,16 @@
 license: other
 base_model: Columbia-NLP/gemma-2b-zephyr-sft
 tags:
+- alignment-handbook
+- trl
+- dpo
+- generated_from_trainer
 - trl
 - dpo
 - alignment-handbook
 - generated_from_trainer
+datasets:
+- argilla/dpo-mix-7k
 model-index:
 - name: zephyr-7b-gemma-dpo
   results: []
@@ -16,17 +22,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # zephyr-7b-gemma-dpo
 
-This model is a fine-tuned version of [Columbia-NLP/gemma-2b-zephyr-sft](https://huggingface.co/Columbia-NLP/gemma-2b-zephyr-sft) on an unknown dataset.
+This model is a fine-tuned version of [Columbia-NLP/gemma-2b-zephyr-sft](https://huggingface.co/Columbia-NLP/gemma-2b-zephyr-sft) on the argilla/dpo-mix-7k dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5972
-- Rewards/chosen: 0.3534
-- Rewards/rejected: 0.1096
-- Rewards/accuracies: 0.6771
-- Rewards/margins: 0.2437
-- Logps/rejected: -373.4094
-- Logps/chosen: -372.0067
-- Logits/rejected: -12.6654
-- Logits/chosen: -13.2926
+- Loss: 0.5964
+- Rewards/chosen: 0.3541
+- Rewards/rejected: 0.1118
+- Rewards/accuracies: 0.6562
+- Rewards/margins: 0.2424
+- Logps/rejected: -373.3665
+- Logps/chosen: -371.9911
+- Logits/rejected: -12.6656
+- Logits/chosen: -13.2953
 
 ## Model description
 

all_results.json

@@ -1,18 +1,18 @@
 {
     "epoch": 1.971563981042654,
-    "eval_logits/chosen": 96.0078125,
-    "eval_logits/rejected": 90.1099853515625,
-    "eval_logps/chosen": -434.6161193847656,
-    "eval_logps/rejected": -466.3072814941406,
-    "eval_loss": 0.4578173756599426,
-    "eval_rewards/accuracies": 0.7604166865348816,
-    "eval_rewards/chosen": -3.5473945140838623,
-    "eval_rewards/margins": 1.6644223928451538,
-    "eval_rewards/rejected": -5.211816787719727,
-    "eval_runtime": 124.9704,
+    "eval_logits/chosen": -13.29525375366211,
+    "eval_logits/rejected": -12.665611267089844,
+    "eval_logps/chosen": -371.9910583496094,
+    "eval_logps/rejected": -373.3665466308594,
+    "eval_loss": 0.596422553062439,
+    "eval_rewards/accuracies": 0.65625,
+    "eval_rewards/chosen": 0.3541353642940521,
+    "eval_rewards/margins": 0.24235813319683075,
+    "eval_rewards/rejected": 0.11177725344896317,
+    "eval_runtime": 14.5207,
     "eval_samples": 750,
-    "eval_samples_per_second": 6.001,
-    "eval_steps_per_second": 0.192,
+    "eval_samples_per_second": 51.65,
+    "eval_steps_per_second": 1.653,
     "total_flos": 0.0,
     "train_loss": 0.627926590350958,
     "train_runtime": 756.4701,

config.json

@@ -24,6 +24,6 @@
   "rope_theta": 10000.0,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.40.1",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 256000
 }

eval_results.json

@@ -1,16 +1,16 @@
 {
     "epoch": 1.971563981042654,
-    "eval_logits/chosen": 96.0078125,
-    "eval_logits/rejected": 90.1099853515625,
-    "eval_logps/chosen": -434.6161193847656,
-    "eval_logps/rejected": -466.3072814941406,
-    "eval_loss": 0.4578173756599426,
-    "eval_rewards/accuracies": 0.7604166865348816,
-    "eval_rewards/chosen": -3.5473945140838623,
-    "eval_rewards/margins": 1.6644223928451538,
-    "eval_rewards/rejected": -5.211816787719727,
-    "eval_runtime": 124.9704,
+    "eval_logits/chosen": -13.29525375366211,
+    "eval_logits/rejected": -12.665611267089844,
+    "eval_logps/chosen": -371.9910583496094,
+    "eval_logps/rejected": -373.3665466308594,
+    "eval_loss": 0.596422553062439,
+    "eval_rewards/accuracies": 0.65625,
+    "eval_rewards/chosen": 0.3541353642940521,
+    "eval_rewards/margins": 0.24235813319683075,
+    "eval_rewards/rejected": 0.11177725344896317,
+    "eval_runtime": 14.5207,
     "eval_samples": 750,
-    "eval_samples_per_second": 6.001,
-    "eval_steps_per_second": 0.192
+    "eval_samples_per_second": 51.65,
+    "eval_steps_per_second": 1.653
 }

runs/Apr27_14-56-02_660111d13776/events.out.tfevents.1714227357.660111d13776.25904.1

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5397c3962fd4361d6a362bd65f7ee18c533ab47c0bf05cb1af37e91ec945dcf8
+size 815