Text Generation
Transformers
PyTorch
gpt_neox
text-generation-inference
File size: 784 Bytes
5f228b7
 
 
9d2aa15
5f228b7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9d2aa15
5f228b7
 
 
 
 
 
 
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
---
license: bigscience-openrail-m
datasets:
- iamplus/Instruction_Tuning
---
Instruction Tuned GPT-NeoXT-20B model on Instruction Tuning dataset as listed below (~560k data) using ***Colossal AI***

**Base Model:** togethercomputer/GPT-NeoXT-Chat-Base-20B (GPT-NeoXT-Chat-Base-20B-v0.16 - fine-tuned on feedback data)

**Training Details :** 
* Epochs: 5
* Batch Size : 16 instantaneous per device x 1 gradient accumulation steps x 8 gpus = 128
* Max Length : 1024
* Weight Decay : 0
* Learning Rate : 2e-5
* Learning Rate Scheduler Type : Cosine
* Number of warmup steps : 240
* Machine : 8xA100 80GB

**Dataset Details :**

Dataset : iamplus/Instruction_Tuning

Files : 
* stanford_alpaca_it_v2.csv
* ColossalChat.csv
* unified_chip2.csv
* iamai_summarization_v1.csv
* iamai_v1.csv