End of training

README.md

@@ -3,9 +3,16 @@ library_name: transformers
 license: other
 base_model: flydust/Llama-3.1-Minitron-4B-Magpie-Gemma2-9B-550K
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- flydust/Llama-3-8B-PO-100K-Gemma-2-9B-armorm
+- Magpie-Align/Magpie-Llama-3.1-Pro-DPO-100K-v0.1
 model-index:
 - name: Llama-3.1-Minitron-4B-Magpie-SFT-G550K-MT-Magpo-3.1-Pro-015Mix
   results: []
@@ -16,17 +23,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # Llama-3.1-Minitron-4B-Magpie-SFT-G550K-MT-Magpo-3.1-Pro-015Mix
 
-This model is a fine-tuned version of [flydust/Llama-3.1-Minitron-4B-Magpie-Gemma2-9B-550K](https://huggingface.co/flydust/Llama-3.1-Minitron-4B-Magpie-Gemma2-9B-550K) on the None dataset.
+This model is a fine-tuned version of [flydust/Llama-3.1-Minitron-4B-Magpie-Gemma2-9B-550K](https://huggingface.co/flydust/Llama-3.1-Minitron-4B-Magpie-Gemma2-9B-550K) on the flydust/Llama-3-8B-PO-100K-Gemma-2-9B-armorm and the Magpie-Align/Magpie-Llama-3.1-Pro-DPO-100K-v0.1 datasets.
 It achieves the following results on the evaluation set:
 - Loss: 0.4942
-- Rewards/chosen: -3.3377
-- Rewards/rejected: -4.2603
+- Rewards/chosen: -3.3368
+- Rewards/rejected: -4.2595
 - Rewards/accuracies: 0.7620
-- Rewards/margins: 0.9226
-- Logps/rejected: -928.2655
-- Logps/chosen: -844.1144
-- Logits/rejected: -1.4591
-- Logits/chosen: -1.4783
+- Rewards/margins: 0.9228
+- Logps/rejected: -928.1929
+- Logps/chosen: -844.0221
+- Logits/rejected: -1.4601
+- Logits/chosen: -1.4791
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 0.9999183606825047,
+    "eval_logits/chosen": -1.4790972471237183,
+    "eval_logits/rejected": -1.4601374864578247,
+    "eval_logps/chosen": -844.0220947265625,
+    "eval_logps/rejected": -928.1929321289062,
+    "eval_loss": 0.4941699206829071,
+    "eval_rewards/accuracies": 0.7620000243186951,
+    "eval_rewards/chosen": -3.3367679119110107,
+    "eval_rewards/margins": 0.9227678179740906,
+    "eval_rewards/rejected": -4.259535789489746,
+    "eval_runtime": 300.0346,
+    "eval_samples": 4000,
+    "eval_samples_per_second": 13.332,
+    "eval_steps_per_second": 0.833,
     "total_flos": 0.0,
     "train_loss": 0.5524635882372952,
     "train_runtime": 40329.8241,

config.json

@@ -31,6 +31,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.45.0.dev0",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 128256
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 0.9999183606825047,
+    "eval_logits/chosen": -1.4790972471237183,
+    "eval_logits/rejected": -1.4601374864578247,
+    "eval_logps/chosen": -844.0220947265625,
+    "eval_logps/rejected": -928.1929321289062,
+    "eval_loss": 0.4941699206829071,
+    "eval_rewards/accuracies": 0.7620000243186951,
+    "eval_rewards/chosen": -3.3367679119110107,
+    "eval_rewards/margins": 0.9227678179740906,
+    "eval_rewards/rejected": -4.259535789489746,
+    "eval_runtime": 300.0346,
+    "eval_samples": 4000,
+    "eval_samples_per_second": 13.332,
+    "eval_steps_per_second": 0.833
+}