Waseem AlShikh
commited on
Commit
•
09bc830
1
Parent(s):
ea7d771
128M model
Browse files- README.md +121 -1
- config.json +32 -0
- merges.txt +0 -0
- pytorch_model.bin +3 -0
- special_tokens_map.json +5 -0
- tokenizer.json +0 -0
- tokenizer_config.json +10 -0
- vocab.json +0 -0
README.md
CHANGED
@@ -1,3 +1,123 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
datasets:
|
5 |
+
- English
|
6 |
+
tags:
|
7 |
+
- text generation
|
8 |
+
- pytorch
|
9 |
+
- causal-lm
|
10 |
+
pipeline_tag: text-generation
|
11 |
+
library_name: transformers
|
12 |
---
|
13 |
+
|
14 |
+
license: cc-by-4.0
|
15 |
+
|
16 |
+
|
17 |
+
# Writer-small 128M
|
18 |
+
|
19 |
+
<style>
|
20 |
+
img {
|
21 |
+
display: inline;
|
22 |
+
}
|
23 |
+
</style>
|
24 |
+
|
25 |
+
|[![Model architecture](https://img.shields.io/badge/Model%20Arch-Transformer%20Decoder-green)](#model-architecture)|[![Model size](https://img.shields.io/badge/Params-126M-green)](#model-architecture)|[![Language](https://img.shields.io/badge/Language-en--US-lightgrey#model-badge)](#datasets)
|
26 |
+
|
27 |
+
|
28 |
+
## Model Description
|
29 |
+
|
30 |
+
Writer-small 128M is a transformer-based language model. GPT refers to a class of transformer decoder-only models similar to GPT-2 and 3 while. It has Tensor Parallelism (TP) of 1, Pipeline Parallelism (PP) of 1 and should fit on a single NVIDIA GPU.
|
31 |
+
|
32 |
+
|
33 |
+
## Getting started
|
34 |
+
|
35 |
+
### Step 1: Install Writer-small and dependencies
|
36 |
+
|
37 |
+
You will need to install NVIDIA Apex.
|
38 |
+
|
39 |
+
```
|
40 |
+
git clone https://github.com/ericharper/apex.git
|
41 |
+
cd apex
|
42 |
+
git checkout nm_v1.11.0
|
43 |
+
pip install -v --disable-pip-version-check --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" --global-option="--fast_layer_norm" --global-option="--distributed_adam" --global-option="--deprecated_fused_adam" ./
|
44 |
+
```
|
45 |
+
|
46 |
+
```
|
47 |
+
pip install nemo_toolkit['nlp']==1.11.0
|
48 |
+
```
|
49 |
+
|
50 |
+
### Step 2: Launch eval server
|
51 |
+
|
52 |
+
**Note.** The model has been trained with Tensor Parallelism (TP) of 1 and Pipeline Parallelism (PP) of 1 and should fit on a single NVIDIA GPU.
|
53 |
+
|
54 |
+
```
|
55 |
+
git clone https://github.com/NVIDIA/NeMo.git
|
56 |
+
cd NeMo/examples/nlp/language_modeling
|
57 |
+
git checkout v1.11.0
|
58 |
+
python megatron_gpt_eval.py gpt_model_file=Writer-gpt-small.nemo server=True tensor_model_parallel_size=1 trainer.devices=1
|
59 |
+
```
|
60 |
+
|
61 |
+
### Step 3: Send prompts to your model!
|
62 |
+
```python
|
63 |
+
import json
|
64 |
+
import requests
|
65 |
+
|
66 |
+
port_num = 5555
|
67 |
+
headers = {"Content-Type": "application/json"}
|
68 |
+
|
69 |
+
def request_data(data):
|
70 |
+
resp = requests.put('http://localhost:{}/generate'.format(port_num),
|
71 |
+
data=json.dumps(data),
|
72 |
+
headers=headers)
|
73 |
+
sentences = resp.json()['sentences']
|
74 |
+
return sentences
|
75 |
+
|
76 |
+
|
77 |
+
data = {
|
78 |
+
"sentences": ["Tell me an interesting fact about space travel."]*1,
|
79 |
+
"tokens_to_generate": 50,
|
80 |
+
"temperature": 1.0,
|
81 |
+
"add_BOS": True,
|
82 |
+
"top_k": 0,
|
83 |
+
"top_p": 0.9,
|
84 |
+
"greedy": False,
|
85 |
+
"all_probs": False,
|
86 |
+
"repetition_penalty": 1.2,
|
87 |
+
"min_tokens_to_generate": 2,
|
88 |
+
}
|
89 |
+
|
90 |
+
sentences = request_data(data)
|
91 |
+
print(sentences)
|
92 |
+
```
|
93 |
+
|
94 |
+
|
95 |
+
## Training Data
|
96 |
+
|
97 |
+
The model was trained on ["The Piles" dataset prepared by Eleuther.AI](https://pile.eleuther.ai/). [4]
|
98 |
+
|
99 |
+
## Evaluation results
|
100 |
+
|
101 |
+
*Zero-shot performance.* Evaluated using [LM Evaluation Test Suite from AI21](https://github.com/AI21Labs/lm-evaluation)
|
102 |
+
|
103 |
+
| ARC-Challenge | ARC-Easy | RACE-middle | RACE-high | Winogrande | RTE | BoolQA | HellaSwag | PiQA |
|
104 |
+
| ------------- | -------- | ----------- | --------- | ---------- | --- | ------ | --------- | ---- |
|
105 |
+
| 0.3012 | 0.4596 | 0.459 | 0.3797 | 0.5343 | 0.5451 | 0.5979 | 0.4443 | 0.6834 |
|
106 |
+
|
107 |
+
## Limitations
|
108 |
+
|
109 |
+
The model was trained on the data originally crawled from the Internet. This data contains toxic language and societal biases. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts.
|
110 |
+
|
111 |
+
## References
|
112 |
+
|
113 |
+
[1] [Improving Language Understanding by Generative Pre-Training](https://s3-us-west-2.amazonaws.com/openai-assets/research-covers/language-unsupervised/language_understanding_paper.pdf)
|
114 |
+
|
115 |
+
[2] [Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism](https://arxiv.org/pdf/1909.08053.pdf)
|
116 |
+
|
117 |
+
[3] [NVIDIA NeMo Toolkit](https://github.com/NVIDIA/NeMo)
|
118 |
+
|
119 |
+
[4] [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://arxiv.org/abs/2101.00027)
|
120 |
+
|
121 |
+
## Licence
|
122 |
+
|
123 |
+
License to use this model is covered by the [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/). By downloading the public and release version of the model, you accept the terms and conditions of the [CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/) license.
|
config.json
ADDED
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "Writer/Writer-LLM-small",
|
3 |
+
"activation_function": "gelu",
|
4 |
+
"architectures": [
|
5 |
+
"GPT2LMHeadModel"
|
6 |
+
],
|
7 |
+
"attn_pdrop": 0.1,
|
8 |
+
"bos_token_id": 50256,
|
9 |
+
"embd_pdrop": 0.1,
|
10 |
+
"eos_token_id": 50256,
|
11 |
+
"initializer_range": 0.023,
|
12 |
+
"layer_norm_epsilon": 1e-05,
|
13 |
+
"model_type": "gpt2",
|
14 |
+
"n_embd": 768,
|
15 |
+
"n_head": 12,
|
16 |
+
"n_inner": 3072,
|
17 |
+
"n_layer": 12,
|
18 |
+
"n_positions": 2048,
|
19 |
+
"reorder_and_upcast_attn": false,
|
20 |
+
"resid_pdrop": 0.1,
|
21 |
+
"scale_attn_by_inverse_layer_idx": false,
|
22 |
+
"scale_attn_weights": true,
|
23 |
+
"summary_activation": null,
|
24 |
+
"summary_first_dropout": 0.1,
|
25 |
+
"summary_proj_to_labels": true,
|
26 |
+
"summary_type": "cls_index",
|
27 |
+
"summary_use_proj": true,
|
28 |
+
"torch_dtype": "float32",
|
29 |
+
"transformers_version": "4.24.0",
|
30 |
+
"use_cache": true,
|
31 |
+
"vocab_size": 50257
|
32 |
+
}
|
merges.txt
ADDED
The diff for this file is too large to render.
See raw diff
|
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:6921cac57faf7302a5a7614be16602556eb58dc2055fb48fd2ebf905aa5d365e
|
3 |
+
size 551292477
|
special_tokens_map.json
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"bos_token": "<|endoftext|>",
|
3 |
+
"eos_token": "<|endoftext|>",
|
4 |
+
"unk_token": "<|endoftext|>"
|
5 |
+
}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
tokenizer_config.json
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"add_prefix_space": false,
|
3 |
+
"bos_token": "<|endoftext|>",
|
4 |
+
"eos_token": "<|endoftext|>",
|
5 |
+
"model_max_length": 1024,
|
6 |
+
"name_or_path": "gpt2",
|
7 |
+
"special_tokens_map_file": null,
|
8 |
+
"tokenizer_class": "GPT2Tokenizer",
|
9 |
+
"unk_token": "<|endoftext|>"
|
10 |
+
}
|
vocab.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|