End of training
Browse files- README.md +12 -11
- all_results.json +20 -0
- eval_results.json +15 -0
- train_results.json +8 -0
- trainer_state.json +423 -0
- training_eval_loss.png +0 -0
- training_loss.png +0 -0
- training_rewards_accuracies.png +0 -0
README.md
CHANGED
|
@@ -3,9 +3,10 @@ library_name: peft
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
|
|
|
|
|
|
| 6 |
- trl
|
| 7 |
- dpo
|
| 8 |
-
- llama-factory
|
| 9 |
- generated_from_trainer
|
| 10 |
model-index:
|
| 11 |
- name: Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_gold
|
|
@@ -17,17 +18,17 @@ should probably proofread and complete it, then remove this comment. -->
|
|
| 17 |
|
| 18 |
# Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_gold
|
| 19 |
|
| 20 |
-
This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on
|
| 21 |
It achieves the following results on the evaluation set:
|
| 22 |
-
- Loss: 0.
|
| 23 |
-
- Rewards/chosen: -0.
|
| 24 |
-
- Rewards/rejected: -3.
|
| 25 |
-
- Rewards/accuracies: 0.
|
| 26 |
-
- Rewards/margins: 3.
|
| 27 |
-
- Logps/chosen: -5.
|
| 28 |
-
- Logps/rejected: -
|
| 29 |
-
- Logits/chosen: -0.
|
| 30 |
-
- Logits/rejected: -0.
|
| 31 |
|
| 32 |
## Model description
|
| 33 |
|
|
|
|
| 3 |
license: llama3.1
|
| 4 |
base_model: meta-llama/Llama-3.1-8B-Instruct
|
| 5 |
tags:
|
| 6 |
+
- llama-factory
|
| 7 |
+
- lora
|
| 8 |
- trl
|
| 9 |
- dpo
|
|
|
|
| 10 |
- generated_from_trainer
|
| 11 |
model-index:
|
| 12 |
- name: Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_gold
|
|
|
|
| 18 |
|
| 19 |
# Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_gold
|
| 20 |
|
| 21 |
+
This model is a fine-tuned version of [meta-llama/Llama-3.1-8B-Instruct](https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct) on the dpo_sg_values_p025_OA_gold dataset.
|
| 22 |
It achieves the following results on the evaluation set:
|
| 23 |
+
- Loss: 0.1369
|
| 24 |
+
- Rewards/chosen: -0.3243
|
| 25 |
+
- Rewards/rejected: -3.7445
|
| 26 |
+
- Rewards/accuracies: 0.9400
|
| 27 |
+
- Rewards/margins: 3.4202
|
| 28 |
+
- Logps/chosen: -5.3343
|
| 29 |
+
- Logps/rejected: -44.0676
|
| 30 |
+
- Logits/chosen: -0.8235
|
| 31 |
+
- Logits/rejected: -0.8529
|
| 32 |
|
| 33 |
## Model description
|
| 34 |
|
all_results.json
ADDED
|
@@ -0,0 +1,20 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"eval_logits/chosen": -0.8234652280807495,
|
| 4 |
+
"eval_logits/rejected": -0.8529273271560669,
|
| 5 |
+
"eval_logps/chosen": -5.334253311157227,
|
| 6 |
+
"eval_logps/rejected": -44.067596435546875,
|
| 7 |
+
"eval_loss": 0.1368969976902008,
|
| 8 |
+
"eval_rewards/accuracies": 0.940000057220459,
|
| 9 |
+
"eval_rewards/chosen": -0.3242790699005127,
|
| 10 |
+
"eval_rewards/margins": 3.4202423095703125,
|
| 11 |
+
"eval_rewards/rejected": -3.7445216178894043,
|
| 12 |
+
"eval_runtime": 90.6755,
|
| 13 |
+
"eval_samples_per_second": 5.514,
|
| 14 |
+
"eval_steps_per_second": 2.757,
|
| 15 |
+
"total_flos": 3.5989630324781875e+17,
|
| 16 |
+
"train_loss": 0.25213732730828975,
|
| 17 |
+
"train_runtime": 8848.1085,
|
| 18 |
+
"train_samples_per_second": 1.512,
|
| 19 |
+
"train_steps_per_second": 0.189
|
| 20 |
+
}
|
eval_results.json
ADDED
|
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"eval_logits/chosen": -0.8234652280807495,
|
| 4 |
+
"eval_logits/rejected": -0.8529273271560669,
|
| 5 |
+
"eval_logps/chosen": -5.334253311157227,
|
| 6 |
+
"eval_logps/rejected": -44.067596435546875,
|
| 7 |
+
"eval_loss": 0.1368969976902008,
|
| 8 |
+
"eval_rewards/accuracies": 0.940000057220459,
|
| 9 |
+
"eval_rewards/chosen": -0.3242790699005127,
|
| 10 |
+
"eval_rewards/margins": 3.4202423095703125,
|
| 11 |
+
"eval_rewards/rejected": -3.7445216178894043,
|
| 12 |
+
"eval_runtime": 90.6755,
|
| 13 |
+
"eval_samples_per_second": 5.514,
|
| 14 |
+
"eval_steps_per_second": 2.757
|
| 15 |
+
}
|
train_results.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"epoch": 1.0,
|
| 3 |
+
"total_flos": 3.5989630324781875e+17,
|
| 4 |
+
"train_loss": 0.25213732730828975,
|
| 5 |
+
"train_runtime": 8848.1085,
|
| 6 |
+
"train_samples_per_second": 1.512,
|
| 7 |
+
"train_steps_per_second": 0.189
|
| 8 |
+
}
|
trainer_state.json
ADDED
|
@@ -0,0 +1,423 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_metric": null,
|
| 3 |
+
"best_model_checkpoint": null,
|
| 4 |
+
"epoch": 1.0,
|
| 5 |
+
"eval_steps": 250,
|
| 6 |
+
"global_step": 1672,
|
| 7 |
+
"is_hyper_param_search": false,
|
| 8 |
+
"is_local_process_zero": true,
|
| 9 |
+
"is_world_process_zero": true,
|
| 10 |
+
"log_history": [
|
| 11 |
+
{
|
| 12 |
+
"epoch": 0.050239234449760764,
|
| 13 |
+
"grad_norm": 2.739856243133545,
|
| 14 |
+
"learning_rate": 5e-07,
|
| 15 |
+
"logits/chosen": -0.5215702056884766,
|
| 16 |
+
"logits/rejected": -0.5743793249130249,
|
| 17 |
+
"logps/chosen": -55.2584228515625,
|
| 18 |
+
"logps/rejected": -27.386730194091797,
|
| 19 |
+
"loss": 0.6931,
|
| 20 |
+
"rewards/accuracies": 0.4627976417541504,
|
| 21 |
+
"rewards/chosen": -0.0007224595756269991,
|
| 22 |
+
"rewards/margins": 0.00034199823858216405,
|
| 23 |
+
"rewards/rejected": -0.001064457930624485,
|
| 24 |
+
"step": 84
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 0.10047846889952153,
|
| 28 |
+
"grad_norm": 5.56309175491333,
|
| 29 |
+
"learning_rate": 1e-06,
|
| 30 |
+
"logits/chosen": -0.5266280174255371,
|
| 31 |
+
"logits/rejected": -0.5796422362327576,
|
| 32 |
+
"logps/chosen": -61.078453063964844,
|
| 33 |
+
"logps/rejected": -28.424898147583008,
|
| 34 |
+
"loss": 0.677,
|
| 35 |
+
"rewards/accuracies": 0.644345223903656,
|
| 36 |
+
"rewards/chosen": 0.01867148093879223,
|
| 37 |
+
"rewards/margins": 0.034289903938770294,
|
| 38 |
+
"rewards/rejected": -0.015618422068655491,
|
| 39 |
+
"step": 168
|
| 40 |
+
},
|
| 41 |
+
{
|
| 42 |
+
"epoch": 0.14952153110047847,
|
| 43 |
+
"eval_logits/chosen": -0.6673032641410828,
|
| 44 |
+
"eval_logits/rejected": -0.6792050004005432,
|
| 45 |
+
"eval_logps/chosen": -2.12101674079895,
|
| 46 |
+
"eval_logps/rejected": -7.310378074645996,
|
| 47 |
+
"eval_loss": 0.6611143946647644,
|
| 48 |
+
"eval_rewards/accuracies": 0.878000020980835,
|
| 49 |
+
"eval_rewards/chosen": -0.0029554006177932024,
|
| 50 |
+
"eval_rewards/margins": 0.06584452092647552,
|
| 51 |
+
"eval_rewards/rejected": -0.06879992038011551,
|
| 52 |
+
"eval_runtime": 90.3233,
|
| 53 |
+
"eval_samples_per_second": 5.536,
|
| 54 |
+
"eval_steps_per_second": 2.768,
|
| 55 |
+
"step": 250
|
| 56 |
+
},
|
| 57 |
+
{
|
| 58 |
+
"epoch": 0.1507177033492823,
|
| 59 |
+
"grad_norm": 1.9286553859710693,
|
| 60 |
+
"learning_rate": 9.441489361702127e-07,
|
| 61 |
+
"logits/chosen": -0.5305180549621582,
|
| 62 |
+
"logits/rejected": -0.5869216918945312,
|
| 63 |
+
"logps/chosen": -55.35295867919922,
|
| 64 |
+
"logps/rejected": -28.797080993652344,
|
| 65 |
+
"loss": 0.6057,
|
| 66 |
+
"rewards/accuracies": 0.8898809552192688,
|
| 67 |
+
"rewards/chosen": 0.11930066347122192,
|
| 68 |
+
"rewards/margins": 0.22872012853622437,
|
| 69 |
+
"rewards/rejected": -0.10941947251558304,
|
| 70 |
+
"step": 252
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"epoch": 0.20095693779904306,
|
| 74 |
+
"grad_norm": 2.7939305305480957,
|
| 75 |
+
"learning_rate": 8.882978723404255e-07,
|
| 76 |
+
"logits/chosen": -0.5447388887405396,
|
| 77 |
+
"logits/rejected": -0.5977838039398193,
|
| 78 |
+
"logps/chosen": -50.50016784667969,
|
| 79 |
+
"logps/rejected": -30.050025939941406,
|
| 80 |
+
"loss": 0.5101,
|
| 81 |
+
"rewards/accuracies": 0.90625,
|
| 82 |
+
"rewards/chosen": 0.2724921405315399,
|
| 83 |
+
"rewards/margins": 0.6402528285980225,
|
| 84 |
+
"rewards/rejected": -0.36776062846183777,
|
| 85 |
+
"step": 336
|
| 86 |
+
},
|
| 87 |
+
{
|
| 88 |
+
"epoch": 0.2511961722488038,
|
| 89 |
+
"grad_norm": 2.6437010765075684,
|
| 90 |
+
"learning_rate": 8.324468085106383e-07,
|
| 91 |
+
"logits/chosen": -0.5628253817558289,
|
| 92 |
+
"logits/rejected": -0.6165186762809753,
|
| 93 |
+
"logps/chosen": -47.81046676635742,
|
| 94 |
+
"logps/rejected": -33.11960220336914,
|
| 95 |
+
"loss": 0.4141,
|
| 96 |
+
"rewards/accuracies": 0.8913690447807312,
|
| 97 |
+
"rewards/chosen": 0.3376307785511017,
|
| 98 |
+
"rewards/margins": 1.0564574003219604,
|
| 99 |
+
"rewards/rejected": -0.7188266515731812,
|
| 100 |
+
"step": 420
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"epoch": 0.29904306220095694,
|
| 104 |
+
"eval_logits/chosen": -0.7404754161834717,
|
| 105 |
+
"eval_logits/rejected": -0.7604221105575562,
|
| 106 |
+
"eval_logps/chosen": -3.4617385864257812,
|
| 107 |
+
"eval_logps/rejected": -19.556419372558594,
|
| 108 |
+
"eval_loss": 0.3327634930610657,
|
| 109 |
+
"eval_rewards/accuracies": 0.8960000276565552,
|
| 110 |
+
"eval_rewards/chosen": -0.13702762126922607,
|
| 111 |
+
"eval_rewards/margins": 1.1563764810562134,
|
| 112 |
+
"eval_rewards/rejected": -1.2934041023254395,
|
| 113 |
+
"eval_runtime": 90.3284,
|
| 114 |
+
"eval_samples_per_second": 5.535,
|
| 115 |
+
"eval_steps_per_second": 2.768,
|
| 116 |
+
"step": 500
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"epoch": 0.3014354066985646,
|
| 120 |
+
"grad_norm": 2.6485891342163086,
|
| 121 |
+
"learning_rate": 7.765957446808509e-07,
|
| 122 |
+
"logits/chosen": -0.5817141532897949,
|
| 123 |
+
"logits/rejected": -0.6466739177703857,
|
| 124 |
+
"logps/chosen": -46.02522277832031,
|
| 125 |
+
"logps/rejected": -36.878692626953125,
|
| 126 |
+
"loss": 0.3166,
|
| 127 |
+
"rewards/accuracies": 0.9032738208770752,
|
| 128 |
+
"rewards/chosen": 0.3111897110939026,
|
| 129 |
+
"rewards/margins": 1.5069682598114014,
|
| 130 |
+
"rewards/rejected": -1.1957786083221436,
|
| 131 |
+
"step": 504
|
| 132 |
+
},
|
| 133 |
+
{
|
| 134 |
+
"epoch": 0.35167464114832536,
|
| 135 |
+
"grad_norm": 3.9122705459594727,
|
| 136 |
+
"learning_rate": 7.207446808510638e-07,
|
| 137 |
+
"logits/chosen": -0.6006772518157959,
|
| 138 |
+
"logits/rejected": -0.6688662767410278,
|
| 139 |
+
"logps/chosen": -47.613197326660156,
|
| 140 |
+
"logps/rejected": -42.94892120361328,
|
| 141 |
+
"loss": 0.2404,
|
| 142 |
+
"rewards/accuracies": 0.9092261791229248,
|
| 143 |
+
"rewards/chosen": 0.2409990429878235,
|
| 144 |
+
"rewards/margins": 2.0163769721984863,
|
| 145 |
+
"rewards/rejected": -1.7753779888153076,
|
| 146 |
+
"step": 588
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"epoch": 0.4019138755980861,
|
| 150 |
+
"grad_norm": 7.711537837982178,
|
| 151 |
+
"learning_rate": 6.648936170212765e-07,
|
| 152 |
+
"logits/chosen": -0.6143811345100403,
|
| 153 |
+
"logits/rejected": -0.6911322474479675,
|
| 154 |
+
"logps/chosen": -54.04777526855469,
|
| 155 |
+
"logps/rejected": -50.569236755371094,
|
| 156 |
+
"loss": 0.1869,
|
| 157 |
+
"rewards/accuracies": 0.9330357313156128,
|
| 158 |
+
"rewards/chosen": 0.23013024032115936,
|
| 159 |
+
"rewards/margins": 2.570420026779175,
|
| 160 |
+
"rewards/rejected": -2.340289831161499,
|
| 161 |
+
"step": 672
|
| 162 |
+
},
|
| 163 |
+
{
|
| 164 |
+
"epoch": 0.4485645933014354,
|
| 165 |
+
"eval_logits/chosen": -0.827215313911438,
|
| 166 |
+
"eval_logits/rejected": -0.8525059819221497,
|
| 167 |
+
"eval_logps/chosen": -5.357245445251465,
|
| 168 |
+
"eval_logps/rejected": -34.60578918457031,
|
| 169 |
+
"eval_loss": 0.19428196549415588,
|
| 170 |
+
"eval_rewards/accuracies": 0.9280000329017639,
|
| 171 |
+
"eval_rewards/chosen": -0.3265782594680786,
|
| 172 |
+
"eval_rewards/margins": 2.4717628955841064,
|
| 173 |
+
"eval_rewards/rejected": -2.7983410358428955,
|
| 174 |
+
"eval_runtime": 90.3138,
|
| 175 |
+
"eval_samples_per_second": 5.536,
|
| 176 |
+
"eval_steps_per_second": 2.768,
|
| 177 |
+
"step": 750
|
| 178 |
+
},
|
| 179 |
+
{
|
| 180 |
+
"epoch": 0.45215311004784686,
|
| 181 |
+
"grad_norm": 4.005836486816406,
|
| 182 |
+
"learning_rate": 6.090425531914894e-07,
|
| 183 |
+
"logits/chosen": -0.6301328539848328,
|
| 184 |
+
"logits/rejected": -0.7088668346405029,
|
| 185 |
+
"logps/chosen": -42.79747772216797,
|
| 186 |
+
"logps/rejected": -50.45292663574219,
|
| 187 |
+
"loss": 0.1786,
|
| 188 |
+
"rewards/accuracies": 0.9345238208770752,
|
| 189 |
+
"rewards/chosen": 0.04424615204334259,
|
| 190 |
+
"rewards/margins": 2.7416493892669678,
|
| 191 |
+
"rewards/rejected": -2.6974034309387207,
|
| 192 |
+
"step": 756
|
| 193 |
+
},
|
| 194 |
+
{
|
| 195 |
+
"epoch": 0.5023923444976076,
|
| 196 |
+
"grad_norm": 3.9601917266845703,
|
| 197 |
+
"learning_rate": 5.531914893617021e-07,
|
| 198 |
+
"logits/chosen": -0.6196833252906799,
|
| 199 |
+
"logits/rejected": -0.7095603346824646,
|
| 200 |
+
"logps/chosen": -47.29566192626953,
|
| 201 |
+
"logps/rejected": -54.954986572265625,
|
| 202 |
+
"loss": 0.1614,
|
| 203 |
+
"rewards/accuracies": 0.9345238208770752,
|
| 204 |
+
"rewards/chosen": 0.14783993363380432,
|
| 205 |
+
"rewards/margins": 3.119297742843628,
|
| 206 |
+
"rewards/rejected": -2.9714577198028564,
|
| 207 |
+
"step": 840
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"epoch": 0.5526315789473685,
|
| 211 |
+
"grad_norm": 4.151618480682373,
|
| 212 |
+
"learning_rate": 4.973404255319149e-07,
|
| 213 |
+
"logits/chosen": -0.5776461958885193,
|
| 214 |
+
"logits/rejected": -0.6859668493270874,
|
| 215 |
+
"logps/chosen": -56.877925872802734,
|
| 216 |
+
"logps/rejected": -61.5712890625,
|
| 217 |
+
"loss": 0.1234,
|
| 218 |
+
"rewards/accuracies": 0.9508928656578064,
|
| 219 |
+
"rewards/chosen": 0.2916085124015808,
|
| 220 |
+
"rewards/margins": 3.574091672897339,
|
| 221 |
+
"rewards/rejected": -3.2824831008911133,
|
| 222 |
+
"step": 924
|
| 223 |
+
},
|
| 224 |
+
{
|
| 225 |
+
"epoch": 0.5980861244019139,
|
| 226 |
+
"eval_logits/chosen": -0.833606481552124,
|
| 227 |
+
"eval_logits/rejected": -0.8610065579414368,
|
| 228 |
+
"eval_logps/chosen": -5.521287441253662,
|
| 229 |
+
"eval_logps/rejected": -39.60647201538086,
|
| 230 |
+
"eval_loss": 0.1578649878501892,
|
| 231 |
+
"eval_rewards/accuracies": 0.9380000233650208,
|
| 232 |
+
"eval_rewards/chosen": -0.3429825007915497,
|
| 233 |
+
"eval_rewards/margins": 2.9554266929626465,
|
| 234 |
+
"eval_rewards/rejected": -3.2984092235565186,
|
| 235 |
+
"eval_runtime": 90.3107,
|
| 236 |
+
"eval_samples_per_second": 5.536,
|
| 237 |
+
"eval_steps_per_second": 2.768,
|
| 238 |
+
"step": 1000
|
| 239 |
+
},
|
| 240 |
+
{
|
| 241 |
+
"epoch": 0.6028708133971292,
|
| 242 |
+
"grad_norm": 2.2310285568237305,
|
| 243 |
+
"learning_rate": 4.4148936170212766e-07,
|
| 244 |
+
"logits/chosen": -0.6338366866111755,
|
| 245 |
+
"logits/rejected": -0.7232056260108948,
|
| 246 |
+
"logps/chosen": -42.76348114013672,
|
| 247 |
+
"logps/rejected": -56.131858825683594,
|
| 248 |
+
"loss": 0.1267,
|
| 249 |
+
"rewards/accuracies": 0.9449405074119568,
|
| 250 |
+
"rewards/chosen": 0.11895173043012619,
|
| 251 |
+
"rewards/margins": 3.4666144847869873,
|
| 252 |
+
"rewards/rejected": -3.3476624488830566,
|
| 253 |
+
"step": 1008
|
| 254 |
+
},
|
| 255 |
+
{
|
| 256 |
+
"epoch": 0.65311004784689,
|
| 257 |
+
"grad_norm": 0.4909895062446594,
|
| 258 |
+
"learning_rate": 3.856382978723404e-07,
|
| 259 |
+
"logits/chosen": -0.612333357334137,
|
| 260 |
+
"logits/rejected": -0.7189978957176208,
|
| 261 |
+
"logps/chosen": -52.86572265625,
|
| 262 |
+
"logps/rejected": -62.92350387573242,
|
| 263 |
+
"loss": 0.0975,
|
| 264 |
+
"rewards/accuracies": 0.9702380895614624,
|
| 265 |
+
"rewards/chosen": 0.2107280194759369,
|
| 266 |
+
"rewards/margins": 3.751209259033203,
|
| 267 |
+
"rewards/rejected": -3.5404813289642334,
|
| 268 |
+
"step": 1092
|
| 269 |
+
},
|
| 270 |
+
{
|
| 271 |
+
"epoch": 0.7033492822966507,
|
| 272 |
+
"grad_norm": 1.8026127815246582,
|
| 273 |
+
"learning_rate": 3.2978723404255315e-07,
|
| 274 |
+
"logits/chosen": -0.6208314895629883,
|
| 275 |
+
"logits/rejected": -0.7236352562904358,
|
| 276 |
+
"logps/chosen": -48.80342102050781,
|
| 277 |
+
"logps/rejected": -60.991233825683594,
|
| 278 |
+
"loss": 0.122,
|
| 279 |
+
"rewards/accuracies": 0.949404776096344,
|
| 280 |
+
"rewards/chosen": 0.10279477387666702,
|
| 281 |
+
"rewards/margins": 3.68986439704895,
|
| 282 |
+
"rewards/rejected": -3.587069511413574,
|
| 283 |
+
"step": 1176
|
| 284 |
+
},
|
| 285 |
+
{
|
| 286 |
+
"epoch": 0.7476076555023924,
|
| 287 |
+
"eval_logits/chosen": -0.8265290856361389,
|
| 288 |
+
"eval_logits/rejected": -0.8553086519241333,
|
| 289 |
+
"eval_logps/chosen": -5.278414249420166,
|
| 290 |
+
"eval_logps/rejected": -42.231048583984375,
|
| 291 |
+
"eval_loss": 0.143920436501503,
|
| 292 |
+
"eval_rewards/accuracies": 0.9360000491142273,
|
| 293 |
+
"eval_rewards/chosen": -0.31869521737098694,
|
| 294 |
+
"eval_rewards/margins": 3.2421717643737793,
|
| 295 |
+
"eval_rewards/rejected": -3.5608673095703125,
|
| 296 |
+
"eval_runtime": 90.3194,
|
| 297 |
+
"eval_samples_per_second": 5.536,
|
| 298 |
+
"eval_steps_per_second": 2.768,
|
| 299 |
+
"step": 1250
|
| 300 |
+
},
|
| 301 |
+
{
|
| 302 |
+
"epoch": 0.7535885167464115,
|
| 303 |
+
"grad_norm": 0.7733187079429626,
|
| 304 |
+
"learning_rate": 2.7393617021276595e-07,
|
| 305 |
+
"logits/chosen": -0.595526933670044,
|
| 306 |
+
"logits/rejected": -0.7032517790794373,
|
| 307 |
+
"logps/chosen": -53.06210708618164,
|
| 308 |
+
"logps/rejected": -64.58065795898438,
|
| 309 |
+
"loss": 0.0987,
|
| 310 |
+
"rewards/accuracies": 0.9598214626312256,
|
| 311 |
+
"rewards/chosen": 0.2242652177810669,
|
| 312 |
+
"rewards/margins": 3.922701835632324,
|
| 313 |
+
"rewards/rejected": -3.698436975479126,
|
| 314 |
+
"step": 1260
|
| 315 |
+
},
|
| 316 |
+
{
|
| 317 |
+
"epoch": 0.8038277511961722,
|
| 318 |
+
"grad_norm": 13.78078556060791,
|
| 319 |
+
"learning_rate": 2.1808510638297872e-07,
|
| 320 |
+
"logits/chosen": -0.6165376305580139,
|
| 321 |
+
"logits/rejected": -0.7192720174789429,
|
| 322 |
+
"logps/chosen": -43.811336517333984,
|
| 323 |
+
"logps/rejected": -62.044307708740234,
|
| 324 |
+
"loss": 0.0933,
|
| 325 |
+
"rewards/accuracies": 0.9717261791229248,
|
| 326 |
+
"rewards/chosen": 0.13936470448970795,
|
| 327 |
+
"rewards/margins": 3.9077537059783936,
|
| 328 |
+
"rewards/rejected": -3.7683887481689453,
|
| 329 |
+
"step": 1344
|
| 330 |
+
},
|
| 331 |
+
{
|
| 332 |
+
"epoch": 0.854066985645933,
|
| 333 |
+
"grad_norm": 8.148602485656738,
|
| 334 |
+
"learning_rate": 1.622340425531915e-07,
|
| 335 |
+
"logits/chosen": -0.5712071657180786,
|
| 336 |
+
"logits/rejected": -0.6922958493232727,
|
| 337 |
+
"logps/chosen": -53.978843688964844,
|
| 338 |
+
"logps/rejected": -67.05591583251953,
|
| 339 |
+
"loss": 0.0821,
|
| 340 |
+
"rewards/accuracies": 0.973214328289032,
|
| 341 |
+
"rewards/chosen": 0.26229721307754517,
|
| 342 |
+
"rewards/margins": 4.126736164093018,
|
| 343 |
+
"rewards/rejected": -3.864438772201538,
|
| 344 |
+
"step": 1428
|
| 345 |
+
},
|
| 346 |
+
{
|
| 347 |
+
"epoch": 0.8971291866028708,
|
| 348 |
+
"eval_logits/chosen": -0.8240648508071899,
|
| 349 |
+
"eval_logits/rejected": -0.8534601330757141,
|
| 350 |
+
"eval_logps/chosen": -5.354381084442139,
|
| 351 |
+
"eval_logps/rejected": -43.72341537475586,
|
| 352 |
+
"eval_loss": 0.13976776599884033,
|
| 353 |
+
"eval_rewards/accuracies": 0.9340000152587891,
|
| 354 |
+
"eval_rewards/chosen": -0.3262918293476105,
|
| 355 |
+
"eval_rewards/margins": 3.383812189102173,
|
| 356 |
+
"eval_rewards/rejected": -3.71010422706604,
|
| 357 |
+
"eval_runtime": 90.2962,
|
| 358 |
+
"eval_samples_per_second": 5.537,
|
| 359 |
+
"eval_steps_per_second": 2.769,
|
| 360 |
+
"step": 1500
|
| 361 |
+
},
|
| 362 |
+
{
|
| 363 |
+
"epoch": 0.9043062200956937,
|
| 364 |
+
"grad_norm": 0.7219157218933105,
|
| 365 |
+
"learning_rate": 1.0638297872340425e-07,
|
| 366 |
+
"logits/chosen": -0.5762205719947815,
|
| 367 |
+
"logits/rejected": -0.6943542957305908,
|
| 368 |
+
"logps/chosen": -50.31645965576172,
|
| 369 |
+
"logps/rejected": -66.44419860839844,
|
| 370 |
+
"loss": 0.0927,
|
| 371 |
+
"rewards/accuracies": 0.9657738208770752,
|
| 372 |
+
"rewards/chosen": 0.20835421979427338,
|
| 373 |
+
"rewards/margins": 4.124646186828613,
|
| 374 |
+
"rewards/rejected": -3.916292190551758,
|
| 375 |
+
"step": 1512
|
| 376 |
+
},
|
| 377 |
+
{
|
| 378 |
+
"epoch": 0.9545454545454546,
|
| 379 |
+
"grad_norm": 8.767385482788086,
|
| 380 |
+
"learning_rate": 5.053191489361702e-08,
|
| 381 |
+
"logits/chosen": -0.5936585664749146,
|
| 382 |
+
"logits/rejected": -0.6990638971328735,
|
| 383 |
+
"logps/chosen": -51.6839714050293,
|
| 384 |
+
"logps/rejected": -65.34242248535156,
|
| 385 |
+
"loss": 0.1066,
|
| 386 |
+
"rewards/accuracies": 0.9553571343421936,
|
| 387 |
+
"rewards/chosen": 0.15223945677280426,
|
| 388 |
+
"rewards/margins": 4.063188076019287,
|
| 389 |
+
"rewards/rejected": -3.9109482765197754,
|
| 390 |
+
"step": 1596
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"epoch": 1.0,
|
| 394 |
+
"step": 1672,
|
| 395 |
+
"total_flos": 3.5989630324781875e+17,
|
| 396 |
+
"train_loss": 0.25213732730828975,
|
| 397 |
+
"train_runtime": 8848.1085,
|
| 398 |
+
"train_samples_per_second": 1.512,
|
| 399 |
+
"train_steps_per_second": 0.189
|
| 400 |
+
}
|
| 401 |
+
],
|
| 402 |
+
"logging_steps": 84,
|
| 403 |
+
"max_steps": 1672,
|
| 404 |
+
"num_input_tokens_seen": 0,
|
| 405 |
+
"num_train_epochs": 1,
|
| 406 |
+
"save_steps": 500,
|
| 407 |
+
"stateful_callbacks": {
|
| 408 |
+
"TrainerControl": {
|
| 409 |
+
"args": {
|
| 410 |
+
"should_epoch_stop": false,
|
| 411 |
+
"should_evaluate": false,
|
| 412 |
+
"should_log": false,
|
| 413 |
+
"should_save": true,
|
| 414 |
+
"should_training_stop": true
|
| 415 |
+
},
|
| 416 |
+
"attributes": {}
|
| 417 |
+
}
|
| 418 |
+
},
|
| 419 |
+
"total_flos": 3.5989630324781875e+17,
|
| 420 |
+
"train_batch_size": 2,
|
| 421 |
+
"trial_name": null,
|
| 422 |
+
"trial_params": null
|
| 423 |
+
}
|
training_eval_loss.png
ADDED
|
training_loss.png
ADDED
|
training_rewards_accuracies.png
ADDED
|