Commit
•
97b64ee
1
Parent(s):
e07cf4b
uploaded readme
Browse files
README.md
ADDED
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
Quantization made by Richard Erkhov.
|
2 |
+
|
3 |
+
[Github](https://github.com/RichardErkhov)
|
4 |
+
|
5 |
+
[Discord](https://discord.gg/pvy7H8DZMG)
|
6 |
+
|
7 |
+
[Request more models](https://github.com/RichardErkhov/quant_request)
|
8 |
+
|
9 |
+
|
10 |
+
pythia-31m-goodwiki-deduped-2048-scratch - bnb 8bits
|
11 |
+
- Model creator: https://huggingface.co/pszemraj/
|
12 |
+
- Original model: https://huggingface.co/pszemraj/pythia-31m-goodwiki-deduped-2048-scratch/
|
13 |
+
|
14 |
+
|
15 |
+
|
16 |
+
|
17 |
+
Original model description:
|
18 |
+
---
|
19 |
+
tags:
|
20 |
+
- generated_from_trainer
|
21 |
+
metrics:
|
22 |
+
- accuracy
|
23 |
+
inference:
|
24 |
+
parameters:
|
25 |
+
max_new_tokens: 64
|
26 |
+
do_sample: true
|
27 |
+
repetition_penalty: 1.1
|
28 |
+
no_repeat_ngram_size: 5
|
29 |
+
guidance_scale: 1.01
|
30 |
+
eta_cutoff: 0.001
|
31 |
+
widget:
|
32 |
+
- text: My name is El Microondas the Wise and
|
33 |
+
example_title: El Microondas
|
34 |
+
- text: A meme is
|
35 |
+
example_title: meme
|
36 |
+
- text: >-
|
37 |
+
Barack Obama nominated Hilary Clinton as his secretary of state on Monday.
|
38 |
+
He chose her because she had
|
39 |
+
example_title: Coreference resolution
|
40 |
+
- text: >-
|
41 |
+
On a shelf, there are five books: a gray book, a red book, a purple book, a
|
42 |
+
blue book, and a black book
|
43 |
+
example_title: Logic puzzles
|
44 |
+
- text: >-
|
45 |
+
The two men running to become New York City's next mayor will face off in
|
46 |
+
their first debate Wednesday night
|
47 |
+
example_title: Reading comprehension
|
48 |
+
pipeline_tag: text-generation
|
49 |
+
license: apache-2.0
|
50 |
+
datasets:
|
51 |
+
- euirim/goodwiki
|
52 |
+
language:
|
53 |
+
- en
|
54 |
+
---
|
55 |
+
|
56 |
+
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
57 |
+
should probably proofread and complete it, then remove this comment. -->
|
58 |
+
|
59 |
+
# pythia-31m-goodwiki-deduped-2048-scratch
|
60 |
+
|
61 |
+
Train from scratch based on config of [EleutherAI/pythia-31m](https://huggingface.co/EleutherAI/pythia-31m) for 3 epochs.
|
62 |
+
|
63 |
+
It achieves the following results on the evaluation set:
|
64 |
+
- Loss: 4.5181
|
65 |
+
- Accuracy: 0.2680
|
66 |
+
|
67 |
+
## Model description
|
68 |
+
|
69 |
+
More information needed
|
70 |
+
|
71 |
+
## Intended uses & limitations
|
72 |
+
|
73 |
+
More information needed
|
74 |
+
|
75 |
+
## Training and evaluation data
|
76 |
+
|
77 |
+
```
|
78 |
+
***** eval metrics *****
|
79 |
+
epoch = 3.0
|
80 |
+
eval_accuracy = 0.2694 eval_loss = 4.4986
|
81 |
+
eval_runtime = 0:00:14.62
|
82 |
+
eval_samples = 500 eval_samples_per_second = 34.187 eval_steps_per_second = 17.093
|
83 |
+
perplexity = 89.8934
|
84 |
+
```
|
85 |
+
|
86 |
+
## Training procedure
|
87 |
+
|
88 |
+
### Training hyperparameters
|
89 |
+
|
90 |
+
The following hyperparameters were used during training:
|
91 |
+
- learning_rate: 0.0005
|
92 |
+
- train_batch_size: 2
|
93 |
+
- eval_batch_size: 2
|
94 |
+
- seed: 80085
|
95 |
+
- gradient_accumulation_steps: 64
|
96 |
+
- total_train_batch_size: 128
|
97 |
+
- optimizer: Adam with betas=(0.9,0.99) and epsilon=1e-07
|
98 |
+
- lr_scheduler_type: inverse_sqrt
|
99 |
+
- lr_scheduler_warmup_ratio: 0.05
|
100 |
+
- num_epochs: 3.0
|
101 |
+
|
102 |
+
### Training results
|
103 |
+
|
104 |
+
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|
105 |
+
|:-------------:|:-----:|:----:|:---------------:|:--------:|
|
106 |
+
| 6.8347 | 0.16 | 100 | 6.7683 | 0.1380 |
|
107 |
+
| 6.0732 | 0.32 | 200 | 6.0489 | 0.1712 |
|
108 |
+
| 5.6949 | 0.48 | 300 | 5.6941 | 0.1935 |
|
109 |
+
| 5.4723 | 0.64 | 400 | 5.4411 | 0.2066 |
|
110 |
+
| 5.2672 | 0.8 | 500 | 5.2621 | 0.2162 |
|
111 |
+
| 5.165 | 0.96 | 600 | 5.1339 | 0.2241 |
|
112 |
+
| 5.0693 | 1.12 | 700 | 5.0290 | 0.2304 |
|
113 |
+
| 4.9234 | 1.28 | 800 | 4.9430 | 0.2369 |
|
114 |
+
| 4.886 | 1.44 | 900 | 4.8702 | 0.2413 |
|
115 |
+
| 4.8422 | 1.6 | 1000 | 4.8086 | 0.2458 |
|
116 |
+
| 4.7688 | 1.76 | 1100 | 4.7593 | 0.2488 |
|
117 |
+
| 4.734 | 1.93 | 1200 | 4.7118 | 0.2527 |
|
118 |
+
| 4.6877 | 2.09 | 1300 | 4.6721 | 0.2556 |
|
119 |
+
| 4.6135 | 2.25 | 1400 | 4.6350 | 0.2583 |
|
120 |
+
| 4.6117 | 2.41 | 1500 | 4.6013 | 0.2606 |
|
121 |
+
| 4.5424 | 2.57 | 1600 | 4.5707 | 0.2635 |
|
122 |
+
| 4.5535 | 2.73 | 1700 | 4.5447 | 0.2658 |
|
123 |
+
| 4.4823 | 2.89 | 1800 | 4.5181 | 0.2680 |
|
124 |
+
|
125 |
+
|
126 |
+
### Framework versions
|
127 |
+
|
128 |
+
- Transformers 4.33.1
|
129 |
+
- Pytorch 2.2.0.dev20230907+cu118
|
130 |
+
- Datasets 2.14.5
|
131 |
+
- Tokenizers 0.13.3
|
132 |
+
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
133 |
+
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_pszemraj__pythia-31m-goodwiki-deduped-2048-scratch)
|
134 |
+
|
135 |
+
| Metric | Value |
|
136 |
+
|-----------------------|---------------------------|
|
137 |
+
| Avg. | 24.85 |
|
138 |
+
| ARC (25-shot) | 23.12 |
|
139 |
+
| HellaSwag (10-shot) | 25.66 |
|
140 |
+
| MMLU (5-shot) | 23.11 |
|
141 |
+
| TruthfulQA (0-shot) | 51.32 |
|
142 |
+
| Winogrande (5-shot) | 49.88 |
|
143 |
+
| GSM8K (5-shot) | 0.0 |
|
144 |
+
| DROP (3-shot) | 0.86 |
|
145 |
+
|
146 |
+
|