Comments (7)
I have the same problem, does anyone know how to solve it? 🌹
from lm-evaluation-harness.
Batching is handled internally by vllm so it shouldn't matter much. Have you tried reducing the gpu_memory_utilization? If you aren't using data parallel, then you could also try using the latest release of vllm if still on 0.3.2.
from lm-evaluation-harness.
Could you explain how data_parallel works? I don't seem to see it in vLLM's documentation.
from lm-evaluation-harness.
Could you explain how data_parallel works? I don't seem to see it in vLLM's documentation.
We create multiple vllm models across different devices so each can process a subset of the data concurrently. But we need to update it for vllm > 0.3.2. Latest vllm versions should work without it though.
from lm-evaluation-harness.
Thanks for the clarification. And also why do we want to lower the gpu_memory_utilization? I thought if we make it close to 1 it won't reserve any memory which means it uses more memory to do the inference. Is it not true?
from lm-evaluation-harness.
Thanks for the clarification. And also why do we want to lower the gpu_memory_utilization? I thought if we make it close to 1 it won't reserve any memory which means it uses more memory to do the inference. Is it not true?
I think it's the opposite 😅. Setting it lower helps with OOM most of the time.
from lm-evaluation-harness.
My bad you are absolutely right 🧎♀️. Let me try these new settings and see if that helps. Thanks a lot!
from lm-evaluation-harness.
Related Issues (20)
- The input format for XNLI seems wired? HOT 2
- Getting error on lm-evaluation for merged models deployed on HF HOT 2
- Avoid slow testing due to network issues. HOT 2
- I get this error whenever I try to run an eval: ImportError: cannot import name 'HfApi' from 'huggingface_hub' HOT 8
- Add More Tests
- eval gsm8k from local dataset folder with the bug info "ValueError: BuilderConfig 'main' not found."
- Errors when loading exact_match.py HOT 2
- Using Language Models as Evaluators HOT 3
- Evaluation results of llama2 with lm-evaluation-harness using wikitext-2 HOT 1
- MPS backend out of memory evaluating fine-tuned Mixtral-8x7B-Instruct-v0.1 on a machine with 100+ GB HOT 2
- sha256 for datasets or samples
- Bug: wrong `until` default value for chat based model HOT 2
- AssertionError: aggregation named 'mean' conflicts with existing registered aggregation!
- Evaluate encoder-decoder-models HOT 1
- Inconsistent evaluation results with Chat Template
- How to use Zeno
- how to select the --model parameter for the meta format checkpoints HOT 3
- --device cuda:3 not honored when using --model vllm HOT 3
- `--tasks list` does not work HOT 4
- Is there something wrong with 'google/gemma-1.1-2b-it' ? HOT 4
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 lm-evaluation-harness.