Model save

README.md

@@ -15,15 +15,15 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [HuggingFaceH4/mistral-7b-sft-beta](https://huggingface.co/HuggingFaceH4/mistral-7b-sft-beta) on the None dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.4553
-- Rewards/chosen: -0.5876
-- Rewards/rejected: -2.1911
-- Rewards/accuracies: 0.8359
-- Rewards/margins: 1.6035
-- Logps/rejected: -246.6992
-- Logps/chosen: -279.5245
-- Logits/rejected: -2.8331
-- Logits/chosen: -2.8422
+- Loss: 1.0563
+- Rewards/chosen: -6.7505
+- Rewards/rejected: -10.0735
+- Rewards/accuracies: 0.7227
+- Rewards/margins: 3.3230
+- Logps/rejected: -273.7712
+- Logps/chosen: -341.3420
+- Logits/rejected: -2.2189
+- Logits/chosen: -2.3037
 
 ## Model description
 
@@ -59,14 +59,14 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
 |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
-| 0.4771        | 1.0   | 485  | 0.4617          | -0.0843        | -1.3520          | 0.7891             | 1.2677          | -238.3082      | -274.4911    | -2.8501         | -2.8796       |
-| 0.4124        | 2.0   | 970  | 0.4545          | -0.3551        | -1.7590          | 0.8164             | 1.4038          | -242.3781      | -277.1996    | -2.8563         | -2.8659       |
-| 0.3549        | 3.0   | 1455 | 0.4553          | -0.5876        | -2.1911          | 0.8359             | 1.6035          | -246.6992      | -279.5245    | -2.8331         | -2.8422       |
+| 0.0636        | 1.0   | 485  | 0.6042          | -4.0022        | -7.1190          | 0.8086             | 3.1168          | -244.2258      | -313.8589    | -2.3960         | -2.4665       |
+| 0.0443        | 2.0   | 970  | 0.7951          | -5.5853        | -8.9194          | 0.7383             | 3.3341          | -262.2304      | -329.6904    | -2.3026         | -2.3851       |
+| 0.0238        | 3.0   | 1455 | 1.0563          | -6.7505        | -10.0735         | 0.7227             | 3.3230          | -273.7712      | -341.3420    | -2.2189         | -2.3037       |
 
 
 ### Framework versions
 
 - Transformers 4.35.0
-- Pytorch 2.1.0+cu121
+- Pytorch 2.1.1+cu121
 - Datasets 2.14.6
 - Tokenizers 0.14.1

all_results.json

@@ -1,21 +1,21 @@
 {
     "epoch": 3.0,
-    "eval_logits/chosen": -2.842160701751709,
-    "eval_logits/rejected": -2.833141326904297,
-    "eval_logps/chosen": -279.5245056152344,
-    "eval_logps/rejected": -246.69915771484375,
-    "eval_loss": 0.45531293749809265,
-    "eval_rewards/accuracies": 0.8359375,
-    "eval_rewards/chosen": -0.5876308083534241,
-    "eval_rewards/margins": 1.6034575700759888,
-    "eval_rewards/rejected": -2.1910881996154785,
-    "eval_runtime": 252.3832,
+    "eval_logits/chosen": -2.3036882877349854,
+    "eval_logits/rejected": -2.218935012817383,
+    "eval_logps/chosen": -341.342041015625,
+    "eval_logps/rejected": -273.77117919921875,
+    "eval_loss": 1.0562912225723267,
+    "eval_rewards/accuracies": 0.72265625,
+    "eval_rewards/chosen": -6.750503063201904,
+    "eval_rewards/margins": 3.3230087757110596,
+    "eval_rewards/rejected": -10.07351303100586,
+    "eval_runtime": 258.1215,
     "eval_samples": 2000,
-    "eval_samples_per_second": 7.924,
-    "eval_steps_per_second": 0.063,
-    "train_loss": 0.43281792414557074,
-    "train_runtime": 46468.4841,
-    "train_samples": 61966,
-    "train_samples_per_second": 4.001,
+    "eval_samples_per_second": 7.748,
+    "eval_steps_per_second": 0.062,
+    "train_loss": 0.07034083745375122,
+    "train_runtime": 46831.0549,
+    "train_samples": 62064,
+    "train_samples_per_second": 3.976,
     "train_steps_per_second": 0.031
 }

eval_results.json

@@ -1,16 +1,16 @@
 {
     "epoch": 3.0,
-    "eval_logits/chosen": -2.842160701751709,
-    "eval_logits/rejected": -2.833141326904297,
-    "eval_logps/chosen": -279.5245056152344,
-    "eval_logps/rejected": -246.69915771484375,
-    "eval_loss": 0.45531293749809265,
-    "eval_rewards/accuracies": 0.8359375,
-    "eval_rewards/chosen": -0.5876308083534241,
-    "eval_rewards/margins": 1.6034575700759888,
-    "eval_rewards/rejected": -2.1910881996154785,
-    "eval_runtime": 252.3832,
+    "eval_logits/chosen": -2.3036882877349854,
+    "eval_logits/rejected": -2.218935012817383,
+    "eval_logps/chosen": -341.342041015625,
+    "eval_logps/rejected": -273.77117919921875,
+    "eval_loss": 1.0562912225723267,
+    "eval_rewards/accuracies": 0.72265625,
+    "eval_rewards/chosen": -6.750503063201904,
+    "eval_rewards/margins": 3.3230087757110596,
+    "eval_rewards/rejected": -10.07351303100586,
+    "eval_runtime": 258.1215,
     "eval_samples": 2000,
-    "eval_samples_per_second": 7.924,
-    "eval_steps_per_second": 0.063
+    "eval_samples_per_second": 7.748,
+    "eval_steps_per_second": 0.062
 }

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 3.0,
-    "train_loss": 0.43281792414557074,
-    "train_runtime": 46468.4841,
-    "train_samples": 61966,
-    "train_samples_per_second": 4.001,
+    "train_loss": 0.07034083745375122,
+    "train_runtime": 46831.0549,
+    "train_samples": 62064,
+    "train_samples_per_second": 3.976,
     "train_steps_per_second": 0.031
 }