End of training

README.md

@@ -3,9 +3,15 @@ library_name: transformers
 license: llama3.1
 base_model: Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- Magpie-Align/MagpieLM-4B-DPO-Data-v0.1
 model-index:
 - name: Llama-3.1-8B-Magpie-SFT-GMix-550K-DPO-02Mix
   results: []
@@ -16,17 +22,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # Llama-3.1-8B-Magpie-SFT-GMix-550K-DPO-02Mix
 
-This model is a fine-tuned version of [Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K](https://huggingface.co/Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K) on an unknown dataset.
+This model is a fine-tuned version of [Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K](https://huggingface.co/Magpie-Align/Llama-3.1-8B-Magpie-SFT-GMix-550K) on the Magpie-Align/MagpieLM-4B-DPO-Data-v0.1 dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.3867
-- Rewards/chosen: -5.1502
-- Rewards/rejected: -6.8793
-- Rewards/accuracies: 0.8080
-- Rewards/margins: 1.7291
-- Logps/rejected: -1153.0979
-- Logps/chosen: -988.9237
-- Logits/rejected: -0.6120
-- Logits/chosen: -0.6722
+- Loss: 0.3866
+- Rewards/chosen: -5.1623
+- Rewards/rejected: -6.8930
+- Rewards/accuracies: 0.8060
+- Rewards/margins: 1.7307
+- Logps/rejected: -1154.4679
+- Logps/chosen: -990.1328
+- Logits/rejected: -0.6102
+- Logits/chosen: -0.6705
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 0.9999183606825047,
+    "eval_logits/chosen": -0.67052161693573,
+    "eval_logits/rejected": -0.61020427942276,
+    "eval_logps/chosen": -990.1327514648438,
+    "eval_logps/rejected": -1154.4678955078125,
+    "eval_loss": 0.38660529255867004,
+    "eval_rewards/accuracies": 0.8059999942779541,
+    "eval_rewards/chosen": -5.162255764007568,
+    "eval_rewards/margins": 1.7307368516921997,
+    "eval_rewards/rejected": -6.89299201965332,
+    "eval_runtime": 614.9178,
+    "eval_samples": 4000,
+    "eval_samples_per_second": 6.505,
+    "eval_steps_per_second": 0.407,
     "total_flos": 0.0,
     "train_loss": 0.46421666473520107,
     "train_runtime": 86694.5826,

config.json

@@ -31,6 +31,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.44.2",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 128256
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 0.9999183606825047,
+    "eval_logits/chosen": -0.67052161693573,
+    "eval_logits/rejected": -0.61020427942276,
+    "eval_logps/chosen": -990.1327514648438,
+    "eval_logps/rejected": -1154.4678955078125,
+    "eval_loss": 0.38660529255867004,
+    "eval_rewards/accuracies": 0.8059999942779541,
+    "eval_rewards/chosen": -5.162255764007568,
+    "eval_rewards/margins": 1.7307368516921997,
+    "eval_rewards/rejected": -6.89299201965332,
+    "eval_runtime": 614.9178,
+    "eval_samples": 4000,
+    "eval_samples_per_second": 6.505,
+    "eval_steps_per_second": 0.407
+}