Comments (2)
I just ran the 1 token 100 times vs the 100 tokens 1 time test and using GPT2 on CPU there's a ~15x speed difference.
from jsonformer.
@Ryul0rd I ran a bunch of performance tests, running generate with max_new_tokens=1 100 times in a loop is not that much slower than running generate once with max_new_tokens=100
The tranformers library's sample implementation seems to be doing this in Python, inside a while loop, so I don't think the main reason for slow perf is because of multiple generate calls.
However, I did track down the source of the slowness, which was caused by the non-vectorized implementation of the number logit warper. This should be fixed in 0.11.0, with much better performance (I'm seeing a 4x speedup on the large example in the readme)
from jsonformer.
Related Issues (20)
- Add Optional and Union types
- Support for JSON-LD
- Make `Jsonformer` derive from `PreTrainedModel`
- Description of keys
- ctransformers / GGML support HOT 6
- Need to support Generation Config for model generation parameters
- allow for do_sample=True HOT 1
- High VRam Requirement HOT 3
- Json Former Causing High V-Memory Usage and Error with Increasing Token Size HOT 1
- Error in logits_processors.py - OutputNumbersTokens().__call__() for some models HOT 2
- Supporting InferenceClient
- AWQ RuntimeError HOT 2
- Documentation for returning an array of objects HOT 10
- Issue with array response HOT 1
- feature pairity with openai function calling
- Support for GPTQ models
- Support for LLaVa HOT 1
- compatibility with Gemini
- Infinite recursion bug when calling generate_number HOT 1
- How to specify a schema where an object can be any key-value combination
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 jsonformer.