Comments (5)
I am trying to fix this bug by delving into the code on my end, but I can't even figure out where the code lives. The first line is "from transformers import GPTNeoForCausalLM, GPT2Tokenizer". I can't find where "GPTNeoForCausalLM" is in transformers. When I do a text search on the whole library folder it turns up empty.
from gpt-neo.
I found out how to browse the source code, but I am now confused about how dividing all the scores by a common value will change the ranking or the final result: https://github.com/huggingface/transformers/blob/87e6e4fe5c7e65cb69e70306f22de6daf16b6e14/src/transformers/generation_logits_process.py#L141
from gpt-neo.
This is not really a bug. I found out I had to go way lower, lower than a typical "float" one would expect most programming languages to be able to handle. I specified 0.00000000000001 as the temperature and now the output is pretty consistent.
from gpt-neo.
I would like to reopen this issue because in some situations with long prompts, even 1e-18 is not small enough to create a totally deterministic response, and at such a small number, the script has a chance of throwing an exception due to "probability tensor contains either inf
, nan
or element < 0"
from gpt-neo.
The workaround is disable sampling
from gpt-neo.
Related Issues (20)
- Exception: stream did not contain valid UTF-8
- Dataset preparation HOT 1
- Inferencing HOT 2
- Freeze Transformer Weight HOT 1
- Incosistent inference TPU vs GPU (huggingface) HOT 1
- Colab: Download of pre trained dataset not possible. the-eye.eu is offline HOT 1
- GPT3_1_3B configuration for a v3-32 TPU HOT 3
- the-eye.eu is down again, is there a mirror? HOT 6
- GPT-neo 350M weights? HOT 3
- Links in the readme to the-eye.eu don't work HOT 1
- Argument not a list with same length as devices
- The locally ran gpt-neo-2.7B is using CPU instead of GPU HOT 1
- Generation should allow user to specify max length of generated portion, rather than total HOT 5
- Not able to generate predicted text after `Done with copy master to slices.` with 1.3B pre-trained model
- The model should return just the generated text, not the prompt text + generated text. HOT 2
- TPU device does not support heartbeats.
- IndexError: index out of range in self HOT 1
- FYI:Japanese pre-trained gpt-neo implementation showcase by using PyTorch, Transformers, and Rust HOT 1
- Cannot Connect To Local TPU-VM HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from gpt-neo.