End of training

README.md

@@ -1,11 +1,13 @@
 ---
 library_name: peft
 tags:
+- alignment-handbook
 - trl
 - dpo
-- alignment-handbook
 - generated_from_trainer
 base_model: NbAiLab/nb-gpt-j-6B-v2
+datasets:
+- hugodk-sch/aftonposten_title_prefs
 model-index:
 - name: aftonposten-6b-align-scan
   results: []
@@ -16,17 +18,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # aftonposten-6b-align-scan
 
-This model is a fine-tuned version of [NbAiLab/nb-gpt-j-6B-v2](https://huggingface.co/NbAiLab/nb-gpt-j-6B-v2) on the None dataset.
+This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
 It achieves the following results on the evaluation set:
 - Loss: 0.4770
-- Rewards/chosen: 0.1914
-- Rewards/rejected: 0.0827
-- Rewards/accuracies: 0.5922
-- Rewards/margins: 0.1087
-- Logps/rejected: -37.3512
-- Logps/chosen: -33.6518
-- Logits/rejected: -2.1389
-- Logits/chosen: -2.1436
+- Rewards/chosen: 0.1934
+- Rewards/rejected: 0.0834
+- Rewards/accuracies: 0.5893
+- Rewards/margins: 0.1100
+- Logps/rejected: -37.3498
+- Logps/chosen: -33.6478
+- Logits/rejected: -2.1388
+- Logits/chosen: -2.1435
 
 ## Model description
 

all_results.json

@@ -1,17 +1,17 @@
 {
     "epoch": 4.0,
-    "eval_logits/chosen": -2.232633352279663,
-    "eval_logits/rejected": -2.2278287410736084,
-    "eval_logps/chosen": -33.75718307495117,
-    "eval_logps/rejected": -37.27397155761719,
-    "eval_loss": 0.49580448865890503,
-    "eval_rewards/accuracies": 0.545265793800354,
-    "eval_rewards/chosen": 0.13868312537670135,
-    "eval_rewards/margins": 0.017359821125864983,
-    "eval_rewards/rejected": 0.12132329493761063,
-    "eval_runtime": 145.5703,
+    "eval_logits/chosen": -2.1435129642486572,
+    "eval_logits/rejected": -2.1388447284698486,
+    "eval_logps/chosen": -33.64778518676758,
+    "eval_logps/rejected": -37.34977340698242,
+    "eval_loss": 0.47702479362487793,
+    "eval_rewards/accuracies": 0.5892857313156128,
+    "eval_rewards/chosen": 0.19338257610797882,
+    "eval_rewards/margins": 0.109960176050663,
+    "eval_rewards/rejected": 0.08342239260673523,
+    "eval_runtime": 145.8636,
     "eval_samples": 343,
-    "eval_samples_per_second": 2.356,
+    "eval_samples_per_second": 2.352,
     "eval_steps_per_second": 0.295,
     "train_loss": 0.2131701453939661,
     "train_runtime": 10811.0627,

config.json

@@ -18,6 +18,8 @@
   "n_layer": 28,
   "n_positions": 2048,
   "quantization_config": {
+    "_load_in_4bit": true,
+    "_load_in_8bit": false,
     "bnb_4bit_compute_dtype": "float16",
     "bnb_4bit_quant_type": "nf4",
     "bnb_4bit_use_double_quant": false,
@@ -45,7 +47,7 @@
   },
   "tie_word_embeddings": false,
   "tokenizer_class": "GPT2Tokenizer",
-  "transformers_version": "4.37.2",
+  "transformers_version": "4.39.0.dev0",
   "use_cache": true,
   "vocab_size": 50400
 }

eval_results.json

@@ -1,16 +1,16 @@
 {
-    "epoch": 1.0,
-    "eval_logits/chosen": -2.232633352279663,
-    "eval_logits/rejected": -2.2278287410736084,
-    "eval_logps/chosen": -33.75718307495117,
-    "eval_logps/rejected": -37.27397155761719,
-    "eval_loss": 0.49580448865890503,
-    "eval_rewards/accuracies": 0.545265793800354,
-    "eval_rewards/chosen": 0.13868312537670135,
-    "eval_rewards/margins": 0.017359821125864983,
-    "eval_rewards/rejected": 0.12132329493761063,
-    "eval_runtime": 145.5703,
+    "epoch": 4.0,
+    "eval_logits/chosen": -2.1435129642486572,
+    "eval_logits/rejected": -2.1388447284698486,
+    "eval_logps/chosen": -33.64778518676758,
+    "eval_logps/rejected": -37.34977340698242,
+    "eval_loss": 0.47702479362487793,
+    "eval_rewards/accuracies": 0.5892857313156128,
+    "eval_rewards/chosen": 0.19338257610797882,
+    "eval_rewards/margins": 0.109960176050663,
+    "eval_rewards/rejected": 0.08342239260673523,
+    "eval_runtime": 145.8636,
     "eval_samples": 343,
-    "eval_samples_per_second": 2.356,
+    "eval_samples_per_second": 2.352,
     "eval_steps_per_second": 0.295
 }