End of training

README.md

@@ -2,9 +2,15 @@
 license: apache-2.0
 base_model: alignment-handbook/zephyr-7b-sft-full
 tags:
+- alignment-handbook
 - trl
 - dpo
 - generated_from_trainer
+- trl
+- dpo
+- generated_from_trainer
+datasets:
+- HuggingFaceH4/ultrafeedback_binarized
 model-index:
 - name: zephyr-7b-dpo-full
   results: []
@@ -15,17 +21,17 @@ should probably proofread and complete it, then remove this comment. -->
 
 # zephyr-7b-dpo-full
 
-This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
+This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the HuggingFaceH4/ultrafeedback_binarized dataset.
 It achieves the following results on the evaluation set:
-- Loss: 0.5348
-- Rewards/chosen: 0.6286
-- Rewards/rejected: -0.2212
-- Rewards/accuracies: 0.7539
-- Rewards/margins: 0.8498
-- Logps/rejected: -267.0915
-- Logps/chosen: -250.0218
-- Logits/rejected: -2.4324
-- Logits/chosen: -2.4731
+- Loss: 0.5352
+- Rewards/chosen: 0.5766
+- Rewards/rejected: -0.3207
+- Rewards/accuracies: 0.7617
+- Rewards/margins: 0.8972
+- Logps/rejected: -269.0807
+- Logps/chosen: -251.0624
+- Logits/rejected: -2.4374
+- Logits/chosen: -2.4784
 
 ## Model description
 

all_results.json

@@ -1,5 +1,18 @@
 {
     "epoch": 1.0,
+    "eval_logits/chosen": -2.4783902168273926,
+    "eval_logits/rejected": -2.4373888969421387,
+    "eval_logps/chosen": -251.0624237060547,
+    "eval_logps/rejected": -269.08074951171875,
+    "eval_loss": 0.5351515412330627,
+    "eval_rewards/accuracies": 0.76171875,
+    "eval_rewards/chosen": 0.5765520930290222,
+    "eval_rewards/margins": 0.8972306251525879,
+    "eval_rewards/rejected": -0.3206784725189209,
+    "eval_runtime": 96.1948,
+    "eval_samples": 2000,
+    "eval_samples_per_second": 20.791,
+    "eval_steps_per_second": 0.333,
     "train_loss": 0.5478911828795238,
     "train_runtime": 7553.9268,
     "train_samples": 61134,

config.json

@@ -21,6 +21,6 @@
   "tie_word_embeddings": false,
   "torch_dtype": "bfloat16",
   "transformers_version": "4.38.2",
-  "use_cache": false,
+  "use_cache": true,
   "vocab_size": 32000
 }

configs.yaml

@@ -0,0 +1,13 @@
+zephyr-7b-beta:
+  prompt_template: "/mnt/bn/xuruijie-llm/eval/dpo/prompt.txt"
+  fn_completions: "huggingface_local_completions"
+  completions_kwargs:
+    model_name: "/mnt/bn/xuruijie-llm/checkpoints/dpo/zephyr-7b-dpo-full-different-0.05-0.04-dpop/"
+    model_kwargs:
+      torch_dtype: 'bfloat16'
+    max_new_tokens: 2048
+    temperature: 0.7
+    top_p: 1.0
+    do_sample: True
+  pretty_name: "Zephyr 7B Beta"
+  link: "https://huggingface.co/HuggingFaceH4/zephyr-7b-beta"

eval_results.json

@@ -1,16 +1,16 @@
 {
     "epoch": 1.0,
-    "eval_logits/chosen": -1.7276861667633057,
-    "eval_logits/rejected": -1.6730353832244873,
-    "eval_logps/chosen": -948.7984619140625,
-    "eval_logps/rejected": -961.3382568359375,
-    "eval_loss": 0.03925349563360214,
-    "eval_rewards/accuracies": 1.0,
-    "eval_rewards/chosen": -30.879230499267578,
-    "eval_rewards/margins": 4.054325103759766,
-    "eval_rewards/rejected": -34.933555603027344,
-    "eval_runtime": 96.1529,
+    "eval_logits/chosen": -2.4783902168273926,
+    "eval_logits/rejected": -2.4373888969421387,
+    "eval_logps/chosen": -251.0624237060547,
+    "eval_logps/rejected": -269.08074951171875,
+    "eval_loss": 0.5351515412330627,
+    "eval_rewards/accuracies": 0.76171875,
+    "eval_rewards/chosen": 0.5765520930290222,
+    "eval_rewards/margins": 0.8972306251525879,
+    "eval_rewards/rejected": -0.3206784725189209,
+    "eval_runtime": 96.1948,
     "eval_samples": 2000,
-    "eval_samples_per_second": 20.8,
+    "eval_samples_per_second": 20.791,
     "eval_steps_per_second": 0.333
 }