Model save

README.md

@@ -2,13 +2,9 @@
 license: mit
 library_name: peft
 tags:
-- alignment-handbook
-- generated_from_trainer
 - trl
 - dpo
 - generated_from_trainer
-datasets:
-- HuggingFaceH4/ultrafeedback_binarized
 base_model: microsoft/phi-2
 model-index:
 - name: phi-2-gpo-iter-0
@@ -20,7 +16,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # phi-2-gpo-iter-0
 
-This model is a fine-tuned version of [lole25/phi-2-sft-ultrachat-lora](https://huggingface.co/lole25/phi-2-sft-ultrachat-lora) on the HuggingFaceH4/ultrafeedback_binarized dataset.
+This model is a fine-tuned version of [microsoft/phi-2](https://huggingface.co/microsoft/phi-2) on the None dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.0103
+- Rewards/chosen: 0.0006
+- Rewards/rejected: 0.0005
+- Rewards/accuracies: 0.4945
+- Rewards/margins: 0.0001
+- Logps/rejected: -233.4960
+- Logps/chosen: -256.5553
+- Logits/rejected: 0.8946
+- Logits/chosen: 0.8373
 
 ## Model description
 

adapter_config.json

@@ -20,9 +20,9 @@
   "revision": null,
   "target_modules": [
     "dense",
-    "v_proj",
+    "k_proj",
     "q_proj",
-    "k_proj"
+    "v_proj"
   ],
   "task_type": "CAUSAL_LM"
 }

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:33a91970ad32fd59ca792aa2083a2c74b90dac687a074c2c49e8e55236c0a2c0
+oid sha256:279aa434e1b02b80bc6174c710c9fcad6a53203ae32da4a203d6a9e813474420
 size 41977616

all_results.json

@@ -1,8 +1,21 @@
 {
-    "epoch": 1.6,
-    "train_loss": 0.0,
-    "train_runtime": 4.6397,
-    "train_samples": 30567,
-    "train_samples_per_second": 4.311,
-    "train_steps_per_second": 0.431
+    "epoch": 1.99,
+    "eval_logits/chosen": 0.8372604250907898,
+    "eval_logits/rejected": 0.8945639133453369,
+    "eval_logps/chosen": -256.5552978515625,
+    "eval_logps/rejected": -233.49595642089844,
+    "eval_loss": 0.01026910264045,
+    "eval_rewards/accuracies": 0.4945000112056732,
+    "eval_rewards/chosen": 0.0006274624029174447,
+    "eval_rewards/margins": 0.00010108990682056174,
+    "eval_rewards/rejected": 0.0005263725761324167,
+    "eval_runtime": 709.1992,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 2.82,
+    "eval_steps_per_second": 0.705,
+    "train_loss": 0.001956978782771095,
+    "train_runtime": 290.9594,
+    "train_samples": 61135,
+    "train_samples_per_second": 6.874,
+    "train_steps_per_second": 0.426
 }

eval_results.json

@@ -0,0 +1,16 @@
+{
+    "epoch": 1.99,
+    "eval_logits/chosen": 0.8372604250907898,
+    "eval_logits/rejected": 0.8945639133453369,
+    "eval_logps/chosen": -256.5552978515625,
+    "eval_logps/rejected": -233.49595642089844,
+    "eval_loss": 0.01026910264045,
+    "eval_rewards/accuracies": 0.4945000112056732,
+    "eval_rewards/chosen": 0.0006274624029174447,
+    "eval_rewards/margins": 0.00010108990682056174,
+    "eval_rewards/rejected": 0.0005263725761324167,
+    "eval_runtime": 709.1992,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 2.82,
+    "eval_steps_per_second": 0.705
+}

train_results.json

@@ -1,8 +1,8 @@
 {
-    "epoch": 1.6,
-    "train_loss": 0.0,
-    "train_runtime": 4.6397,
-    "train_samples": 30567,
-    "train_samples_per_second": 4.311,
-    "train_steps_per_second": 0.431
+    "epoch": 1.99,
+    "train_loss": 0.001956978782771095,
+    "train_runtime": 290.9594,
+    "train_samples": 61135,
+    "train_samples_per_second": 6.874,
+    "train_steps_per_second": 0.426
 }

trainer_state.json

@@ -1,9 +1,9 @@
 {
   "best_metric": null,
   "best_model_checkpoint": null,
-  "epoch": 1.6,
+  "epoch": 1.992,
   "eval_steps": 100,
-  "global_step": 100,
+  "global_step": 124,
   "is_hyper_param_search": false,
   "is_local_process_zero": true,
   "is_world_process_zero": true,
@@ -163,17 +163,45 @@
       "step": 100
     },
     {
-      "epoch": 1.6,
-      "step": 100,
+      "epoch": 1.77,
+      "learning_rate": 1.937002879188285e-07,
+      "logits/chosen": 0.8532508611679077,
+      "logits/rejected": 0.8138865232467651,
+      "logps/chosen": -175.2100830078125,
+      "logps/rejected": -158.42974853515625,
+      "loss": 0.0102,
+      "rewards/accuracies": 0.48124998807907104,
+      "rewards/chosen": -0.0008482037228532135,
+      "rewards/margins": 0.00030287052504718304,
+      "rewards/rejected": -0.0011510740732774138,
+      "step": 110
+    },
+    {
+      "epoch": 1.93,
+      "learning_rate": 1.6003680950742728e-08,
+      "logits/chosen": 0.8294161558151245,
+      "logits/rejected": 0.826370120048523,
+      "logps/chosen": -158.0736083984375,
+      "logps/rejected": -146.1743927001953,
+      "loss": 0.01,
+      "rewards/accuracies": 0.48750001192092896,
+      "rewards/chosen": -0.0006492567481473088,
+      "rewards/margins": 0.0010562509996816516,
+      "rewards/rejected": -0.0017055077478289604,
+      "step": 120
+    },
+    {
+      "epoch": 1.99,
+      "step": 124,
       "total_flos": 0.0,
-      "train_loss": 0.0,
-      "train_runtime": 4.6397,
-      "train_samples_per_second": 4.311,
-      "train_steps_per_second": 0.431
+      "train_loss": 0.001956978782771095,
+      "train_runtime": 290.9594,
+      "train_samples_per_second": 6.874,
+      "train_steps_per_second": 0.426
     }
   ],
   "logging_steps": 10,
-  "max_steps": 2,
+  "max_steps": 124,
   "num_input_tokens_seen": 0,
   "num_train_epochs": 2,
   "save_steps": 100,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dd2d8b87ad92dc36814226c3df0892a8aee8625e5b676832e31a4039e9f1e50c
+oid sha256:8eae090f69235327d3649619a5e45322d1fd2adadd2e607fcfed49b1f3988ff4
 size 5816