Comments (7)
@niparis I just pushed an image to run on GPU using GPTQ-for-llama, please test it out and let me know your thoughts! here is the image
I also added a section for more details in the README.md
from llm-api.
I have pushed a new image with the requested fix, you can find it in the latest README.me file
from llm-api.
I'm currently trying to work out a docker image that can run GPTQ for Llama, let me know if you have some tips
from llm-api.
It is working great.
One suggestion tho, can we get to adjust max_length and max_new_tokens as parameters please ? Default values are at 50 for both.
I edited gptq_llama.py in the meantime:
l154: max_length = params.get("max_length", 2048)
l157: max_new_tokens = params.get("max_new_tokens", 2048)
l168: max_new_tokens=max_new_tokens,
edit: I tried this before but it did not work, that's why I'm bothering you.
llm = LLMAPI(
... host_name="http://localhost:8000",
... params={"max_length": 2048, "max_new_tokens": 2048, "temp": 0.2},
... verbose=True
... )
from llm-api.
Sorry, I'm actually dumb I just had to edit the config.yaml.
edit: Well, I was talking too fast
Input length of input_ids is 240, but
max_lengthis set to 50. This can lead to unexpected behavior. You should consider increasing
max_new_tokens.
I'll put the values in my script like this
llm = LLMAPI(
host_name="http://localhost:8000",
params = {"n_predict": 300, "temp": 0.2, "max_length": 2048, "max_new_tokens": 2048},
verbose=True
)
Also it'd be great if the models were stored in separate folders instead of gptq_llama so tokenizer.model config.json and tokenizer_config.json don't get overwritten each time we change modelss.
Now on the bright side, this implementation is way better than using the textgen webui's api, it's simpler and faster. I'm using a french finetuned model and use a script that'll output on tty, Since I barely know what I'm doing, that saves me the headache of having to deal with unicode, utf8 and special characters for the time being.
from llm-api.
Thanks for the feedback and for being patient, I usually have very little time to work on this project. But I will adjust the directory thing for sure!
from llm-api.
Closing this for being outdated after supporting different GPU options
from llm-api.
Related Issues (12)
- any advice for getting this running with `gpt-x-alpaca` models? HOT 20
- Question and stop sequence are included in LLM response HOT 5
- No module named 'torch' HOT 4
- How to train model with my data - can i use api? HOT 1
- Illegal instruction (core dumped) HOT 3
- Is this quantized 7b ggml model too old? HOT 9
- Request: Lora support HOT 1
- Consider refactor for model types
- I need a tutorial or real example. HOT 2
- GPTQ Models with safetensors? - 'Missing key(s) in state_dict' HOT 2
- Example config file HOT 2
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 llm-api.