dpo_f6rfgz12

README.md

@@ -17,15 +17,15 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model was trained from scratch on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.7021
-- Rewards/chosen: 0.5480
-- Rewards/rejected: 0.4854
-- Rewards/accuracies: 0.1242
-- Rewards/margins: 0.0626
-- Logps/rejected: -26.1813
-- Logps/chosen: -28.6110
-- Logits/rejected: -1.0475
-- Logits/chosen: -1.0554
+- Loss: 0.7471
+- Rewards/chosen: 1.4869
+- Rewards/rejected: 1.1641
+- Rewards/accuracies: 0.4383
+- Rewards/margins: 0.3229
+- Logps/rejected: -140.3850
+- Logps/chosen: -153.8973
+- Logits/rejected: -0.6448
+- Logits/chosen: -0.6851
 
 ## Model description
 
@@ -48,10 +48,12 @@ The following hyperparameters were used during training:
 - train_batch_size: 1
 - eval_batch_size: 1
 - seed: 0
+- gradient_accumulation_steps: 2
+- total_train_batch_size: 2
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_steps: 100
-- training_steps: 2500
+- training_steps: 2000
 
 ### Training results
 

adapter_config.json

@@ -16,17 +16,17 @@
   "megatron_core": "megatron.core",
   "modules_to_save": null,
   "peft_type": "LORA",
-  "r": 64,
+  "r": 32,
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "k_proj",
-    "wte",
-    "q_proj",
     "fc_out",
+    "k_proj",
+    "fc_in",
     "out_proj",
+    "q_proj",
     "v_proj",
-    "fc_in"
+    "wte"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:37c713439897aefab1d63b1f1938ec6a3ff714264132976bebe1d9780fd5dfe3
-size 75523504
+oid sha256:d3336b30496cedb477a68def19166b41054123f3cbd0f48e23a18bc33dafcf79
+size 37774720

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b116eedb32c45e3728095b6935fbdfd4c2aa57a3764d638aa33e5a1d3237a04
+oid sha256:2e8d935741839e6606b91e10fcc75f5571638a97b5871cec541b7b522e6905d6
 size 5560