metadata
license: llama3
base_model: tsavage68/UTI_L3_1000steps_1e5rate_SFT
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: UTI2_L3_1000steps_1e6rate_01beta_CSFTDPO
results: []
UTI2_L3_1000steps_1e6rate_01beta_CSFTDPO
This model is a fine-tuned version of tsavage68/UTI_L3_1000steps_1e5rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.2741
- Rewards/chosen: -0.0170
- Rewards/rejected: -6.7809
- Rewards/accuracies: 0.6400
- Rewards/margins: 6.7639
- Logps/rejected: -96.2941
- Logps/chosen: -19.2736
- Logits/rejected: -1.2664
- Logits/chosen: -1.2475
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 1e-06
- train_batch_size: 2
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 4
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1000
Training results
Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
---|---|---|---|---|---|---|---|---|---|---|---|
0.67 | 0.3333 | 25 | 0.6075 | 0.1072 | -0.0786 | 0.6400 | 0.1858 | -29.2710 | -18.0315 | -1.1541 | -1.1497 |
0.3388 | 0.6667 | 50 | 0.3079 | 0.3701 | -1.1689 | 0.6500 | 1.5390 | -40.1739 | -15.4027 | -1.1704 | -1.1602 |
0.1782 | 1.0 | 75 | 0.2489 | 0.3405 | -3.3088 | 0.6500 | 3.6493 | -61.5725 | -15.6982 | -1.2173 | -1.2009 |
0.1047 | 1.3333 | 100 | 0.2514 | 0.3299 | -4.1473 | 0.6500 | 4.4772 | -69.9577 | -15.8048 | -1.2277 | -1.2096 |
0.1909 | 1.6667 | 125 | 0.2649 | 0.2370 | -4.5013 | 0.6400 | 4.7383 | -73.4979 | -16.7332 | -1.2311 | -1.2144 |
0.364 | 2.0 | 150 | 0.2617 | 0.2324 | -4.8873 | 0.6400 | 5.1197 | -77.3577 | -16.7794 | -1.2337 | -1.2169 |
0.26 | 2.3333 | 175 | 0.2628 | 0.1974 | -5.1469 | 0.6400 | 5.3443 | -79.9539 | -17.1290 | -1.2363 | -1.2194 |
0.2253 | 2.6667 | 200 | 0.2643 | 0.1698 | -5.3745 | 0.6400 | 5.5443 | -82.2301 | -17.4054 | -1.2386 | -1.2217 |
0.208 | 3.0 | 225 | 0.2660 | 0.1513 | -5.5214 | 0.6400 | 5.6727 | -83.6984 | -17.5904 | -1.2407 | -1.2238 |
0.2253 | 3.3333 | 250 | 0.2667 | 0.1290 | -5.6833 | 0.6400 | 5.8124 | -85.3180 | -17.8128 | -1.2430 | -1.2261 |
0.1733 | 3.6667 | 275 | 0.2681 | 0.1116 | -5.8186 | 0.6400 | 5.9301 | -86.6704 | -17.9877 | -1.2452 | -1.2281 |
0.2773 | 4.0 | 300 | 0.2686 | 0.1005 | -5.9317 | 0.6400 | 6.0322 | -87.8013 | -18.0979 | -1.2472 | -1.2299 |
0.2426 | 4.3333 | 325 | 0.2690 | 0.0844 | -6.0431 | 0.6400 | 6.1276 | -88.9161 | -18.2589 | -1.2493 | -1.2319 |
0.156 | 4.6667 | 350 | 0.2692 | 0.0741 | -6.1302 | 0.6400 | 6.2043 | -89.7871 | -18.3627 | -1.2509 | -1.2333 |
0.2253 | 5.0 | 375 | 0.2715 | 0.0625 | -6.2127 | 0.6400 | 6.2752 | -90.6117 | -18.4779 | -1.2530 | -1.2353 |
0.2253 | 5.3333 | 400 | 0.2713 | 0.0535 | -6.2910 | 0.6400 | 6.3446 | -91.3949 | -18.5679 | -1.2545 | -1.2367 |
0.2253 | 5.6667 | 425 | 0.2724 | 0.0411 | -6.3668 | 0.6400 | 6.4079 | -92.1528 | -18.6919 | -1.2563 | -1.2383 |
0.208 | 6.0 | 450 | 0.2729 | 0.0353 | -6.4187 | 0.6400 | 6.4541 | -92.6719 | -18.7501 | -1.2573 | -1.2392 |
0.2773 | 6.3333 | 475 | 0.2736 | 0.0283 | -6.4704 | 0.6400 | 6.4987 | -93.1886 | -18.8205 | -1.2582 | -1.2400 |
0.3119 | 6.6667 | 500 | 0.2725 | 0.0224 | -6.5105 | 0.6400 | 6.5329 | -93.5893 | -18.8791 | -1.2592 | -1.2409 |
0.208 | 7.0 | 525 | 0.2719 | 0.0140 | -6.5739 | 0.6400 | 6.5880 | -94.2240 | -18.9630 | -1.2606 | -1.2422 |
0.1733 | 7.3333 | 550 | 0.2740 | 0.0094 | -6.6118 | 0.6400 | 6.6212 | -94.6024 | -19.0092 | -1.2618 | -1.2433 |
0.2599 | 7.6667 | 575 | 0.2728 | 0.0021 | -6.6411 | 0.6400 | 6.6432 | -94.8961 | -19.0825 | -1.2622 | -1.2436 |
0.2599 | 8.0 | 600 | 0.2736 | -0.0003 | -6.6671 | 0.6400 | 6.6668 | -95.1557 | -19.1060 | -1.2631 | -1.2444 |
0.2253 | 8.3333 | 625 | 0.2728 | -0.0010 | -6.6895 | 0.6400 | 6.6884 | -95.3796 | -19.1137 | -1.2634 | -1.2447 |
0.104 | 8.6667 | 650 | 0.2735 | -0.0019 | -6.7075 | 0.6400 | 6.7056 | -95.5598 | -19.1222 | -1.2641 | -1.2453 |
0.2253 | 9.0 | 675 | 0.2726 | -0.0051 | -6.7243 | 0.6400 | 6.7192 | -95.7281 | -19.1544 | -1.2648 | -1.2460 |
0.2253 | 9.3333 | 700 | 0.2736 | -0.0097 | -6.7446 | 0.6400 | 6.7348 | -95.9304 | -19.2006 | -1.2653 | -1.2465 |
0.2253 | 9.6667 | 725 | 0.2740 | -0.0130 | -6.7590 | 0.6400 | 6.7460 | -96.0751 | -19.2334 | -1.2655 | -1.2466 |
0.3119 | 10.0 | 750 | 0.2742 | -0.0140 | -6.7661 | 0.6400 | 6.7520 | -96.1452 | -19.2434 | -1.2656 | -1.2466 |
0.208 | 10.3333 | 775 | 0.2741 | -0.0154 | -6.7688 | 0.6400 | 6.7534 | -96.1727 | -19.2569 | -1.2660 | -1.2470 |
0.2253 | 10.6667 | 800 | 0.2728 | -0.0133 | -6.7751 | 0.6400 | 6.7618 | -96.2353 | -19.2360 | -1.2661 | -1.2471 |
0.2426 | 11.0 | 825 | 0.2734 | -0.0133 | -6.7787 | 0.6400 | 6.7654 | -96.2719 | -19.2365 | -1.2662 | -1.2473 |
0.2946 | 11.3333 | 850 | 0.2743 | -0.0138 | -6.7737 | 0.6400 | 6.7599 | -96.2217 | -19.2417 | -1.2663 | -1.2474 |
0.1733 | 11.6667 | 875 | 0.2739 | -0.0147 | -6.7807 | 0.6400 | 6.7660 | -96.2913 | -19.2500 | -1.2662 | -1.2472 |
0.156 | 12.0 | 900 | 0.2751 | -0.0158 | -6.7820 | 0.6400 | 6.7661 | -96.3044 | -19.2615 | -1.2664 | -1.2475 |
0.1906 | 12.3333 | 925 | 0.2747 | -0.0152 | -6.7835 | 0.6400 | 6.7682 | -96.3194 | -19.2557 | -1.2663 | -1.2474 |
0.2426 | 12.6667 | 950 | 0.2741 | -0.0190 | -6.7817 | 0.6400 | 6.7627 | -96.3018 | -19.2932 | -1.2665 | -1.2475 |
0.208 | 13.0 | 975 | 0.2741 | -0.0170 | -6.7809 | 0.6400 | 6.7639 | -96.2941 | -19.2736 | -1.2664 | -1.2475 |
0.3119 | 13.3333 | 1000 | 0.2741 | -0.0170 | -6.7809 | 0.6400 | 6.7639 | -96.2941 | -19.2736 | -1.2664 | -1.2475 |
Framework versions
- Transformers 4.41.2
- Pytorch 2.0.0+cu117
- Datasets 2.19.2
- Tokenizers 0.19.1