gemma_2b_oasst1_reward_model

README.md

@@ -20,8 +20,8 @@ should probably proofread and complete it, then remove this comment. -->
 
 This model is a fine-tuned version of [google/gemma-2b](https://huggingface.co/google/gemma-2b) on an unknown dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.4481
-- Accuracy: 0.8036
+- Loss: 0.4250
+- Accuracy: 0.7881
 
 ## Model description
 
@@ -55,13 +55,9 @@ The following hyperparameters were used during training:
 
 | Training Loss | Epoch | Step | Validation Loss | Accuracy |
 |:-------------:|:-----:|:----:|:---------------:|:--------:|
-| 0.4696        | 0.38  | 100  | 0.5181          | 0.7542   |
-| 0.4327        | 0.76  | 200  | 0.4738          | 0.8025   |
-| 0.3946        | 1.15  | 300  | 0.5145          | 0.7924   |
-| 0.3372        | 1.53  | 400  | 0.5370          | 0.7890   |
-| 0.3618        | 1.91  | 500  | 0.4481          | 0.8036   |
-| 0.3292        | 2.29  | 600  | 0.4799          | 0.7991   |
-| 0.4514        | 2.68  | 700  | 0.4763          | 0.8013   |
+| 0.443         | 1.0   | 100  | 0.5045          | 0.7458   |
+| 0.4098        | 2.0   | 200  | 0.4312          | 0.7938   |
+| 0.5036        | 2.99  | 300  | 0.4250          | 0.7881   |
 
 
 ### Framework versions

adapter_config.json

@@ -14,15 +14,15 @@
   "megatron_config": null,
   "megatron_core": "megatron.core",
   "modules_to_save": [
-    "scores"
+    "score"
   ],
   "peft_type": "LORA",
   "r": 16,
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "q_proj",
-    "v_proj"
+    "v_proj",
+    "q_proj"
   ],
   "task_type": "SEQ_CLS",
   "use_dora": false,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bd018fa832746abef320d15daa2a18ffa499ac099d65fa6cd61eecd8541ef664
+oid sha256:6fbbe5c5a3cbdef631341ec6ca752168bcbfb155d7263b87eda0c2a0de80bc47
 size 7390624

tokenizer_config.json

@@ -40,7 +40,7 @@
   "device": "cuda",
   "eos_token": "<eos>",
   "legacy": null,
-  "max_length": 512,
+  "max_length": 128,
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<eos>",
   "padding": "max_length",

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5b47e8c2a64e60d475eacab92795d3306f950bae04351265e2a2e58e609fba95
+oid sha256:360893070a94058414051ef16b43089429c4f91060f408d7d06a8d39753745ff
 size 4856