File size: 4,622 Bytes
f181b18 8e782e8 f181b18 8e782e8 f181b18 8e782e8 f181b18 8e782e8 f181b18 8e782e8 f181b18 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 |
---
base_model: gpt2
datasets:
- wikimedia/wikipedia
library_name: Distily
license: mit
tags:
- bitnet
- 1.58b
- generated_from_trainer
model-index:
- name: distily_multi_attn_experiment_ortho
results: []
---
# Summary
Distilled with [Distily](https://github.com/lapp0/distily) library
using teacher model [gpt2](https://huggingface.co/gpt2)
on dataset [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia).
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment.
# Model description
More information needed
# Intended uses & limitations
More information needed
-->
# Model Architecture:
- **Architecture**: `GPT2LMHeadModel`
- **Total Parameters**: 124,439,808
- **Data Type (dtype)**: torch.bfloat16
- **Model Size**: 0.24 GB
# Benchmark Metrics Comparison
| Metric | attn_layer_mapper=all, attn_loss_fn=cos, attn_projector=orthogonal, attn_weight=5 | attn_layer_mapper=layer-2, attn_loss_fn=raw_mse, attn_projector=orthogonal, attn_weight=25.0 | teacher |
| :--- | :--- | :--- | :--- |
| ai2_arc (acc) | 0.313 | 0.305 | 0.354 |
| ai2_arc (acc_norm) | 0.31 | 0.302 | 0.339 |
| arc_challenge (acc) | 0.181 | 0.173 | 0.188 |
| arc_challenge (acc_norm) | 0.224 | 0.223 | 0.222 |
| arc_easy (acc) | 0.378 | 0.37 | 0.436 |
| arc_easy (acc_norm) | 0.353 | 0.34 | 0.396 |
| boolq (acc) | 0.49 | 0.387 | 0.51 |
| cola (mcc) | -0.041 | 0.044 | 0.01 |
| glue (acc) | 0.396 | 0.412 | 0.403 |
| glue (f1) | 0.516 | 0.451 | 0.529 |
| glue (mcc) | -0.041 | 0.044 | 0.01 |
| hellaswag (acc) | 0.32 | 0.315 | 0.343 |
| hellaswag (acc_norm) | 0.348 | 0.344 | 0.393 |
| mnli (acc) | 0.336 | 0.338 | 0.338 |
| mnli_mismatch (acc) | 0.343 | 0.351 | 0.346 |
| mrpc (acc) | 0.444 | 0.353 | 0.515 |
| mrpc (f1) | 0.478 | 0.143 | 0.631 |
| qnli (acc) | 0.488 | 0.497 | 0.491 |
| qqp (acc) | 0.356 | 0.406 | 0.367 |
| qqp (f1) | 0.522 | 0.501 | 0.512 |
| rte (acc) | 0.56 | 0.549 | 0.516 |
| sst2 (acc) | 0.498 | 0.545 | 0.511 |
| wikitext (bits_per_byte) | 1.118 | 1.127 | 0.98 |
| wikitext (byte_perplexity) | 2.17 | 2.184 | 1.973 |
| wikitext (word_perplexity) | 63.05 | 65.25 | 37.82 |
| wnli (acc) | 0.408 | 0.451 | 0.451 |
# Resource Usage Comparison
- VRAM Use: 8.2855 GB
# Distillation (Teacher -> Student) Architecture Difference:
- **Architecture**: `GPT2LMHeadModel` -> `GPT2LMHeadModel`
- **Total Parameters**: 124,439,808 -> 124,439,808
- **Data Type (dtype)**: torch.bfloat16 -> torch.bfloat16
- **Model Size**: 0.24 GB -> 0.24 GB
<details>
<summary>Module Diff Details</summary>
```diff
```
</details>
<br/>
# Train Dataset
Trained on 145,724,804 tokens from the [wikimedia/wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) dataset.
- Num Samples: `247,500`
- Subset: `20231101.en`
- Split: `train`
# Training Objective
```
DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=5, loss_fn=cos, layer_mapper=all))
```
# Hyperparameters
The following hyperparameters were used during training:
<details>
<summary>Expand</summary>
- learning_rate: `0.0001`
- train_batch_size: `4`
- eval_batch_size: `8`
- seed: `42`
- optimizer: `Adam with betas=(0.9,0.999) and epsilon=1e-08`
- lr_scheduler_type: `cosine_with_min_lr`
- lr_scheduler_warmup_ratio: `0.5`
- num_epochs: `1.0`
- distillation_objective: `DistillationObjective(logits_loss_component=LossComponent(label=logits, weight=1, loss_fn=kl), attn_loss_component=LossComponent(label=attn, weight=5, loss_fn=cos, layer_mapper=all))`
- train_embeddings: `True`
- lr_scheduler: `<torch.optim.lr_scheduler.LambdaLR object at 0x7f05c40e2050>`
- student_model_name_or_path: `None`
- student_config_name_or_path: `None`
- student_model_config: `None`
- reinitialize_weights: `None`
- copy_teacher_modules: `[('lm_head', False)]`
- student_model_as_bitnet: `True`
- student_model_compile: `False`
- dropout: `None`
- teacher_model_name_or_path: `gpt2`
- teacher_load_in_8bit: `False`
- teacher_load_in_4bit: `False`
- teacher_model_compile: `False`
- dataset_uri: `wikimedia/wikipedia`
- dataset_subset: `20231101.en`
- dataset_split: `train`
- dataset_column_name: `text`
- dataset_sample_size: `250000`
- dataset_test_size: `0.01`
- gradient_accumulation_steps: `1`
- weight_decay: `0.0`
- max_grad_norm: `1.0`
- warmup_ratio: `0.5`
- warmup_steps: `0`
- gradient_checkpointing: `True`
</details>
<br/>
# Framework Versions
- Distily 0.3.0
- Transformers 4.44.0
- Pytorch 2.3.0
- Datasets 2.21.0
|