End of training

README.md

@@ -1,8 +1,15 @@
 ---
+base_model: data/zephyr-7b-sft-full-accumulation2
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: zephyr-7b-dpo-full-accumulation4
   results: []
@@ -13,17 +20,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # zephyr-7b-dpo-full-accumulation4
 
-This model was trained from scratch on an unknown dataset.
+This model is a fine-tuned version of [data/zephyr-7b-sft-full-accumulation2](https://huggingface.co/data/zephyr-7b-sft-full-accumulation2) on the HuggingFaceH4/ultrafeedback_binarized dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5042
-- Rewards/chosen: -1.0384
-- Rewards/rejected: -2.0520
-- Rewards/accuracies: 0.7773
-- Rewards/margins: 1.0136
-- Logps/rejected: -465.5701
-- Logps/chosen: -363.5876
-- Logits/rejected: 3.4727
-- Logits/chosen: 2.9519
+- Loss: 0.5032
+- Rewards/chosen: -0.9893
+- Rewards/rejected: -2.0234
+- Rewards/accuracies: 0.7812
+- Rewards/margins: 1.0341
+- Logps/rejected: -462.7061
+- Logps/chosen: -358.6745
+- Logits/rejected: 3.3182
+- Logits/chosen: 2.7991
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 0.9984301412872841,
+    "eval_logits/chosen": 2.7991323471069336,
+    "eval_logits/rejected": 3.318207263946533,
+    "eval_logps/chosen": -358.6744689941406,
+    "eval_logps/rejected": -462.7060546875,
+    "eval_loss": 0.5032069683074951,
+    "eval_rewards/accuracies": 0.78125,
+    "eval_rewards/chosen": -0.9892802238464355,
+    "eval_rewards/margins": 1.0340925455093384,
+    "eval_rewards/rejected": -2.0233726501464844,
+    "eval_runtime": 134.6793,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 14.85,
+    "eval_steps_per_second": 0.238,
     "total_flos": 0.0,
     "train_loss": 0.5498417743346976,
     "train_runtime": 10726.7175,

config.json

@@ -21,6 +21,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.40.0",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 32000
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 0.9984301412872841,
+    "eval_logits/chosen": 2.7991323471069336,
+    "eval_logits/rejected": 3.318207263946533,
+    "eval_logps/chosen": -358.6744689941406,
+    "eval_logps/rejected": -462.7060546875,
+    "eval_loss": 0.5032069683074951,
+    "eval_rewards/accuracies": 0.78125,
+    "eval_rewards/chosen": -0.9892802238464355,
+    "eval_rewards/margins": 1.0340925455093384,
+    "eval_rewards/rejected": -2.0233726501464844,
+    "eval_runtime": 134.6793,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 14.85,
+    "eval_steps_per_second": 0.238
+}