Training in progress, step 1400

README.md

@@ -20,15 +20,15 @@ should probably proofread and complete it, then remove this comment. -->
 
 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: 0.1934
-- 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
+- Loss: 1.0817
+- Rewards/chosen: 0.0048
+- Rewards/rejected: -0.0009
+- Rewards/accuracies: 0.5104
+- Rewards/margins: 0.0057
+- Logps/rejected: -37.5184
+- Logps/chosen: -34.0250
+- Logits/rejected: -2.2281
+- Logits/chosen: -2.2329
 
 ## Model description
 
@@ -57,33 +57,21 @@ The following hyperparameters were used during training:
 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
 - lr_scheduler_type: cosine
 - lr_scheduler_warmup_ratio: 0.1
-- num_epochs: 4
+- num_epochs: 1
 
 ### Training results
 
-| Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
-|:-------------:|:-----:|:----:|:-------------:|:---------------:|:------------:|:--------------:|:---------------:|:------------------:|:--------------:|:---------------:|:----------------:|
-| 0.4794        | 0.26  | 100  | -2.2375       | -2.2327         | -33.8699     | -37.3767       | 0.4971          | 0.5573             | 0.0824         | 0.0124          | 0.0699           |
-| 0.4561        | 0.52  | 200  | -2.2342       | -2.2294         | -33.7761     | -37.2930       | 0.4959          | 0.5424             | 0.1292         | 0.0174          | 0.1118           |
-| 0.4202        | 0.78  | 300  | -2.2333       | -2.2285         | -33.7508     | -37.2789       | 0.4944          | 0.5307             | 0.1419         | 0.0230          | 0.1189           |
-| 0.3592        | 1.04  | 400  | 0.4901        | 0.1964          | 0.1560       | 0.5768         | 0.0404          | -37.2046           | -33.6418       | -2.2025         | -2.2073          |
-| 0.3252        | 1.3   | 500  | 0.4906        | 0.2174          | 0.1800       | 0.5162         | 0.0374          | -37.1566           | -33.5997       | -2.2010         | -2.2057          |
-| 0.3006        | 1.56  | 600  | 0.4827        | 0.2385          | 0.1637       | 0.5710         | 0.0747          | -37.1891           | -33.5576       | -2.1813         | -2.1861          |
-| 0.3054        | 1.82  | 700  | 0.4810        | 0.2268          | 0.1440       | 0.5976         | 0.0828          | -37.2286           | -33.5809       | -2.1687         | -2.1734          |
-| 0.2314        | 2.08  | 800  | 0.4799        | 0.1958          | 0.1042       | 0.5743         | 0.0916          | -37.3083           | -33.6429       | -2.1556         | -2.1603          |
-| 0.2187        | 2.34  | 900  | 0.4799        | 0.1986          | 0.1061       | 0.5860         | 0.0925          | -37.3044           | -33.6374       | -2.1490         | -2.1537          |
-| 0.2382        | 2.6   | 1000 | 0.4784        | 0.1879          | 0.0856       | 0.6121         | 0.1023          | -37.3454           | -33.6588       | -2.1419         | -2.1466          |
-| 0.2041        | 2.86  | 1100 | 0.4782        | 0.1909          | 0.0885       | 0.5889         | 0.1024          | -37.3396           | -33.6528       | -2.1392         | -2.1438          |
-| 0.2467        | 3.12  | 1200 | 0.4766        | 0.1941          | 0.0835       | 0.5893         | 0.1106          | -37.3496           | -33.6463       | -2.1390         | -2.1437          |
-| 0.1951        | 3.38  | 1300 | 0.4767        | 0.1928          | 0.0830       | 0.5889         | 0.1098          | -37.3506           | -33.6490       | -2.1394         | -2.1441          |
-| 0.1994        | 3.64  | 1400 | 0.4777        | 0.1881          | 0.0826       | 0.5743         | 0.1055          | -37.3515           | -33.6584       | -2.1386         | -2.1433          |
-| 0.2272        | 3.9   | 1500 | 0.4770        | 0.1914          | 0.0827       | 0.5922         | 0.1087          | -37.3512           | -33.6518       | -2.1389         | -2.1436          |
+| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| 0.8581        | 0.26  | 100  | 1.0730          | 0.0088         | 0.0069           | 0.5199             | 0.0019          | -37.5028       | -34.0170     | -2.2326         | -2.2374       |
+| 0.899         | 0.52  | 200  | 1.0716          | -0.0015        | -0.0088          | 0.5071             | 0.0072          | -37.5342       | -34.0376     | -2.2306         | -2.2355       |
+| 0.6105        | 0.78  | 300  | 1.0315          | 0.0102         | -0.0099          | 0.5602             | 0.0201          | -37.5365       | -34.0141     | -2.2289         | -2.2338       |
 
 
 ### Framework versions
 
-- PEFT 0.10.0
-- Transformers 4.39.0.dev0
+- PEFT 0.8.2
+- Transformers 4.37.2
 - Pytorch 2.1.2+cu121
-- Datasets 2.14.6
+- Datasets 2.17.0
 - Tokenizers 0.15.1

adapter_config.json

@@ -20,13 +20,13 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "down_proj",
+    "up_proj",
+    "gate_proj",
     "o_proj",
+    "v_proj",
     "q_proj",
     "k_proj",
-    "v_proj",
-    "up_proj",
-    "gate_proj"
+    "down_proj"
   ],
   "task_type": "CAUSAL_LM",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6f46c85f0fcd659e30195c2a03dc670cbb5bf2695d2760e07499028db39b52e8
+oid sha256:12f6d09aa3ac048cb53037f14ce14baac1d201ebceec3988e0749fb178db23ef
 size 176183216

all_results.json

@@ -1,21 +1,21 @@
 {
-    "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,
+    "epoch": 1.0,
+    "eval_logits/chosen": -2.232933521270752,
+    "eval_logits/rejected": -2.2280824184417725,
+    "eval_logps/chosen": -34.02503967285156,
+    "eval_logps/rejected": -37.51841735839844,
+    "eval_loss": 1.0816813707351685,
+    "eval_rewards/accuracies": 0.5103820562362671,
+    "eval_rewards/chosen": 0.004756772890686989,
+    "eval_rewards/margins": 0.005652438849210739,
+    "eval_rewards/rejected": -0.0008956658421084285,
+    "eval_runtime": 145.5936,
     "eval_samples": 343,
-    "eval_samples_per_second": 2.352,
+    "eval_samples_per_second": 2.356,
     "eval_steps_per_second": 0.295,
-    "train_loss": 0.2131701453939661,
-    "train_runtime": 10811.0627,
+    "train_loss": 0.8446982934877469,
+    "train_runtime": 3252.3399,
     "train_samples": 3079,
-    "train_samples_per_second": 1.139,
-    "train_steps_per_second": 0.142
+    "train_samples_per_second": 0.947,
+    "train_steps_per_second": 0.118
 }

config.json

@@ -18,8 +18,6 @@
   "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,
@@ -47,7 +45,7 @@
   },
   "tie_word_embeddings": false,
   "tokenizer_class": "GPT2Tokenizer",
-  "transformers_version": "4.39.0.dev0",
+  "transformers_version": "4.37.2",
   "use_cache": true,
   "vocab_size": 50400
 }

eval_results.json

@@ -1,16 +1,16 @@
 {
-    "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,
+    "epoch": 1.0,
+    "eval_logits/chosen": -2.232933521270752,
+    "eval_logits/rejected": -2.2280824184417725,
+    "eval_logps/chosen": -34.02503967285156,
+    "eval_logps/rejected": -37.51841735839844,
+    "eval_loss": 1.0816813707351685,
+    "eval_rewards/accuracies": 0.5103820562362671,
+    "eval_rewards/chosen": 0.004756772890686989,
+    "eval_rewards/margins": 0.005652438849210739,
+    "eval_rewards/rejected": -0.0008956658421084285,
+    "eval_runtime": 145.5936,
     "eval_samples": 343,
-    "eval_samples_per_second": 2.352,
+    "eval_samples_per_second": 2.356,
     "eval_steps_per_second": 0.295
 }

train_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 4.0,
-    "train_loss": 0.2131701453939661,
-    "train_runtime": 10811.0627,
+    "epoch": 1.0,
+    "train_loss": 0.8446982934877469,
+    "train_runtime": 3252.3399,
     "train_samples": 3079,
-    "train_samples_per_second": 1.139,
-    "train_steps_per_second": 0.142
+    "train_samples_per_second": 0.947,
+    "train_steps_per_second": 0.118
 }

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:02ca49827b893bc712e0a301674e01ebffe16d4c7500943127387ef65d9c0346
+oid sha256:cba7cea87426fb79c9f89c8c302642cfb8f90ae77279e90f5d889e1a7e26dc53
 size 5176