a3x/gpt-4chan
1
0
Fork 0
mirror of https://huggingface.co/ykilcher/gpt-4chan synced 2024-12-22 04:00:02 +00:00
Code Issues Projects Releases Wiki Activity

model card update

Browse source
This commit is contained in:
Yannic Kilcher 2022-05-29 20:46:46 +02:00
parent 0680e0197a
commit e18aeebdaa
1 changed files with 8 additions and 6 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
README.md
View file

@ -58,7 +58,7 @@ gen_tokens = model.generate(
gen_text = tokenizer.batch_decode(gen_tokens)[0] gen_text = tokenizer.batch_decode(gen_tokens)[0]
``` ```
For the float16 model(GPU): For the float16 model (GPU):
```python ```python
from transformers import GPTJForCausalLM, AutoTokenizer from transformers import GPTJForCausalLM, AutoTokenizer
import torch import torch
@ -69,6 +69,7 @@ import torch
model = GPTJForCausalLM.from_pretrained( model = GPTJForCausalLM.from_pretrained(
"ykilcher/gpt-4chan", revision="float16", torch_dtype=torch.float16, low_cpu_mem_usage=True "ykilcher/gpt-4chan", revision="float16", torch_dtype=torch.float16, low_cpu_mem_usage=True
) )
model.cuda()
tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B") tokenizer = AutoTokenizer.from_pretrained("EleutherAI/gpt-j-6B")
prompt = ( prompt = (
@ -78,6 +79,7 @@ prompt = (
) )
input_ids = tokenizer(prompt, return_tensors="pt").input_ids input_ids = tokenizer(prompt, return_tensors="pt").input_ids
input_ids = input_ids.cuda()
gen_tokens = model.generate( gen_tokens = model.generate(
input_ids, input_ids,
@ -91,16 +93,15 @@ gen_text = tokenizer.batch_decode(gen_tokens)[0]
### Limitations and Biases ### Limitations and Biases
This is a statistical model This is a statistical model. As such, it continues text as is likely under the distribution the model has learned from the training data. Outputs should not be interpreted as "correct", "truthful", or otherwise as anything more than a statistical function of the input. That being said, GPT-4chan does significantly outperform GPT-J (and GPT-3) on the [TruthfulQA Benchmark](https://arxiv.org/abs/2109.07958) that measures whether a language model is truthful in generating answers to questions.
Dataset from 2016 to 2019 and biased. The dataset is time- and domain-limited. It was collected from 2016 to 2019 on 4chan's _politically incorrect_ board. As such, political topics from that area will be overrepresented in the model's distribution, compared to other models (e.g. GPT-J 6B). Also, due to the very lax rules and anonymity of posters, a large part of the dataset contains offensive material. Thus, it is **very likely that the model will produce offensive outputs**, including but not limited to: toxicity, hate speech, racism, sexism, homo- and transphobia, xenophobia, and anti-semitism.
Will be offensive. Due to the above limitations, it is strongly recommend to not deploy this model into a real-world environment unless its behavior is well-understood and explicit and strict limitations on the scope, impact, and duration of the deployment are enforced.
Do not deploy without appropriate measures.
## Evaluation results ## Evaluation results
### Language Model Evaluation Harness ### Language Model Evaluation Harness
The following table compares GPT-J 6B to GPT-4chan on a subset of the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness). The following table compares GPT-J 6B to GPT-4chan on a subset of the [Language Model Evaluation Harness](https://github.com/EleutherAI/lm-evaluation-harness).
@ -447,3 +448,4 @@ Differences exceeding standard errors are marked in the "Significant" column wit
| reversed_words | acc | 0 | 0 | 0.0003 | 0.000173188 | + | | reversed_words | acc | 0 | 0 | 0.0003 | 0.000173188 | + |
<figcaption><p>Some results are missing due to errors or computational constraints.</p> <figcaption><p>Some results are missing due to errors or computational constraints.</p>
</figcaption></figure> </figcaption></figure>