chansung commited on
Commit
5292c8c
1 Parent(s): 83a49e5

Model save

Browse files
Files changed (4) hide show
  1. README.md +18 -14
  2. all_results.json +6 -11
  3. train_results.json +6 -6
  4. trainer_state.json +0 -0
README.md CHANGED
@@ -1,11 +1,10 @@
1
  ---
2
  base_model: meta-llama/Meta-Llama-3-8B
3
  datasets:
4
- - llama-duo/synth_classification_dataset_dedup
5
  library_name: peft
6
  license: llama3
7
  tags:
8
- - alignment-handbook
9
  - trl
10
  - sft
11
  - generated_from_trainer
@@ -19,9 +18,9 @@ should probably proofread and complete it, then remove this comment. -->
19
 
20
  # llama3-8b-classification-gpt4o-100k
21
 
22
- This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B) on the llama-duo/synth_classification_dataset_dedup dataset.
23
  It achieves the following results on the evaluation set:
24
- - Loss: 1.8087
25
 
26
  ## Model description
27
 
@@ -42,27 +41,32 @@ More information needed
42
  The following hyperparameters were used during training:
43
  - learning_rate: 0.0002
44
  - train_batch_size: 4
45
- - eval_batch_size: 2
46
  - seed: 42
47
  - distributed_type: multi-GPU
48
  - num_devices: 4
49
  - gradient_accumulation_steps: 2
50
  - total_train_batch_size: 32
51
- - total_eval_batch_size: 8
52
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
53
  - lr_scheduler_type: cosine
54
  - lr_scheduler_warmup_ratio: 0.1
55
- - num_epochs: 5
56
 
57
  ### Training results
58
 
59
- | Training Loss | Epoch | Step | Validation Loss |
60
- |:-------------:|:------:|:----:|:---------------:|
61
- | 1.4429 | 0.9978 | 225 | 1.7484 |
62
- | 1.3594 | 2.0 | 451 | 1.7719 |
63
- | 1.3154 | 2.9978 | 676 | 1.7679 |
64
- | 1.268 | 4.0 | 902 | 1.7960 |
65
- | 1.2695 | 4.9889 | 1125 | 1.8087 |
 
 
 
 
 
66
 
67
 
68
  ### Framework versions
 
1
  ---
2
  base_model: meta-llama/Meta-Llama-3-8B
3
  datasets:
4
+ - generator
5
  library_name: peft
6
  license: llama3
7
  tags:
 
8
  - trl
9
  - sft
10
  - generated_from_trainer
 
18
 
19
  # llama3-8b-classification-gpt4o-100k
20
 
21
+ This model is a fine-tuned version of [meta-llama/Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B) on the generator dataset.
22
  It achieves the following results on the evaluation set:
23
+ - Loss: 2.0032
24
 
25
  ## Model description
26
 
 
41
  The following hyperparameters were used during training:
42
  - learning_rate: 0.0002
43
  - train_batch_size: 4
44
+ - eval_batch_size: 4
45
  - seed: 42
46
  - distributed_type: multi-GPU
47
  - num_devices: 4
48
  - gradient_accumulation_steps: 2
49
  - total_train_batch_size: 32
50
+ - total_eval_batch_size: 16
51
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
52
  - lr_scheduler_type: cosine
53
  - lr_scheduler_warmup_ratio: 0.1
54
+ - num_epochs: 10
55
 
56
  ### Training results
57
 
58
+ | Training Loss | Epoch | Step | Validation Loss |
59
+ |:-------------:|:-----:|:----:|:---------------:|
60
+ | 0.9183 | 1.0 | 237 | 1.6517 |
61
+ | 0.8583 | 2.0 | 474 | 1.6295 |
62
+ | 0.8179 | 3.0 | 711 | 1.6559 |
63
+ | 0.7533 | 4.0 | 948 | 1.6894 |
64
+ | 0.716 | 5.0 | 1185 | 1.7251 |
65
+ | 0.6876 | 6.0 | 1422 | 1.7830 |
66
+ | 0.6344 | 7.0 | 1659 | 1.8557 |
67
+ | 0.591 | 8.0 | 1896 | 1.9240 |
68
+ | 0.5677 | 9.0 | 2133 | 1.9842 |
69
+ | 0.5648 | 10.0 | 2370 | 2.0032 |
70
 
71
 
72
  ### Framework versions
all_results.json CHANGED
@@ -1,14 +1,9 @@
1
  {
2
- "epoch": 4.988913525498892,
3
- "eval_loss": 1.8087279796600342,
4
- "eval_runtime": 0.3834,
5
- "eval_samples": 16,
6
- "eval_samples_per_second": 2.608,
7
- "eval_steps_per_second": 2.608,
8
- "total_flos": 1.6629843858229821e+18,
9
- "train_loss": 1.390608725865682,
10
- "train_runtime": 3610.9267,
11
  "train_samples": 92634,
12
- "train_samples_per_second": 9.977,
13
- "train_steps_per_second": 0.312
14
  }
 
1
  {
2
+ "epoch": 10.0,
3
+ "total_flos": 3.5097090775444357e+18,
4
+ "train_loss": 0.7390849222110797,
5
+ "train_runtime": 8188.9555,
 
 
 
 
 
6
  "train_samples": 92634,
7
+ "train_samples_per_second": 9.243,
8
+ "train_steps_per_second": 0.289
9
  }
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
- "epoch": 4.988913525498892,
3
- "total_flos": 1.6629843858229821e+18,
4
- "train_loss": 1.390608725865682,
5
- "train_runtime": 3610.9267,
6
  "train_samples": 92634,
7
- "train_samples_per_second": 9.977,
8
- "train_steps_per_second": 0.312
9
  }
 
1
  {
2
+ "epoch": 10.0,
3
+ "total_flos": 3.5097090775444357e+18,
4
+ "train_loss": 0.7390849222110797,
5
+ "train_runtime": 8188.9555,
6
  "train_samples": 92634,
7
+ "train_samples_per_second": 9.243,
8
+ "train_steps_per_second": 0.289
9
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff