End of training

README.md

@@ -2,9 +2,15 @@
 library_name: transformers
 base_model: allenai/tulu-2-7b
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: uf-tulu-2-7b-dpo
   results: []
@@ -15,7 +21,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # uf-tulu-2-7b-dpo
 
-This model is a fine-tuned version of [allenai/tulu-2-7b](https://huggingface.co/allenai/tulu-2-7b) on an unknown dataset.
+This model is a fine-tuned version of [allenai/tulu-2-7b](https://huggingface.co/allenai/tulu-2-7b) on the HuggingFaceH4/ultrafeedback_binarized dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.6635
+- Rewards/chosen: 0.0484
+- Rewards/rejected: -0.0274
+- Rewards/accuracies: 0.6797
+- Rewards/margins: 0.0759
+- Logps/rejected: -325.4168
+- Logps/chosen: -316.8686
+- Logits/rejected: -1.2824
+- Logits/chosen: -1.2603
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 1.0,
+    "eval_logits/chosen": -1.2603442668914795,
+    "eval_logits/rejected": -1.2824182510375977,
+    "eval_logps/chosen": -316.86859130859375,
+    "eval_logps/rejected": -325.4167785644531,
+    "eval_loss": 0.6635114550590515,
+    "eval_rewards/accuracies": 0.6796875,
+    "eval_rewards/chosen": 0.04844370484352112,
+    "eval_rewards/margins": 0.0758848488330841,
+    "eval_rewards/rejected": -0.02744114398956299,
+    "eval_runtime": 91.2633,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 21.915,
+    "eval_steps_per_second": 0.351,
     "total_flos": 0.0,
     "train_loss": 0.6731121569996599,
     "train_runtime": 6739.8654,

config.json

@@ -24,6 +24,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.44.1",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 32000
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 1.0,
+    "eval_logits/chosen": -1.2603442668914795,
+    "eval_logits/rejected": -1.2824182510375977,
+    "eval_logps/chosen": -316.86859130859375,
+    "eval_logps/rejected": -325.4167785644531,
+    "eval_loss": 0.6635114550590515,
+    "eval_rewards/accuracies": 0.6796875,
+    "eval_rewards/chosen": 0.04844370484352112,
+    "eval_rewards/margins": 0.0758848488330841,
+    "eval_rewards/rejected": -0.02744114398956299,
+    "eval_runtime": 91.2633,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 21.915,
+    "eval_steps_per_second": 0.351
+}