gpt2_dpo_lora/1

README.md

@@ -13,6 +13,16 @@ should probably proofread and complete it, then remove this comment. -->
 # training
 
 This model is a fine-tuned version of [lvwerra/gpt2-imdb](https://huggingface.co/lvwerra/gpt2-imdb) on an unknown dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.4649
+- Rewards/chosen: 1.1097
+- Rewards/rejected: 0.3323
+- Rewards/accuracies: 0.8186
+- Rewards/margins: 0.7774
+- Logps/rejected: -143.4800
+- Logps/chosen: -175.0714
+- Logits/rejected: -35.2043
+- Logits/chosen: -32.7114
 
 ## Model description
 
@@ -39,6 +49,17 @@ The following hyperparameters were used during training:
 - lr_scheduler_type: linear
 - num_epochs: 3
 
+### Training results
+
+| Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
+|:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
+| No log        | 0.55  | 400  | 0.6593          | 1.0074         | 0.5904           | 0.7357             | 0.4170          | -140.8990      | -176.0949    | -35.9356        | -33.1922      |
+| 0.7974        | 1.11  | 800  | 0.5807          | 1.1511         | 0.5902           | 0.7634             | 0.5610          | -140.9016      | -174.6575    | -35.9192        | -33.2655      |
+| 0.5983        | 1.66  | 1200 | 0.5200          | 1.0697         | 0.4300           | 0.7979             | 0.6397          | -142.5030      | -175.4720    | -35.5696        | -33.0300      |
+| 0.4982        | 2.21  | 1600 | 0.4807          | 1.1128         | 0.3733           | 0.8158             | 0.7395          | -143.0704      | -175.0409    | -35.2967        | -32.7791      |
+| 0.4663        | 2.77  | 2000 | 0.4649          | 1.1097         | 0.3323           | 0.8186             | 0.7774          | -143.4800      | -175.0714    | -35.2043        | -32.7114      |
+
+
 ### Framework versions
 
 - Transformers 4.35.2

adapter_config.json

@@ -0,0 +1,23 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "lvwerra/gpt2-imdb",
+  "bias": "none",
+  "fan_in_fan_out": true,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "lora_alpha": 0.5,
+  "lora_dropout": 0.0,
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "c_proj",
+    "c_attn"
+  ],
+  "task_type": "CAUSAL_LM"
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d2d7f0ed97d241f21c4537ea68fd873a9428be93f84a318996e3325f30cfa12
+size 3253104

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b35f933a61cdd61ce70b51887d1e525299cc71a65ab97155bcf8125a224ed3c4
-size 4536
+oid sha256:1f1dc624044b41e0b15000b8a7c3dbf22f1849c23f4519a2102e0a315c840a9f
+size 4600