Comments (3)
CharlesRiggins
commented on July 4, 2024
1
Chunked prefill is ready starting from version v0.4.2. You can upgrade to the latest version and try again.
from vllm.
simon-mo
commented on July 4, 2024
cc @rkooo567
from vllm.
rkooo567
commented on July 4, 2024
As @CharlesRiggins the feature is supported from 0.4.2. But I recommend you to use 0.5.0
I want to know more about --enable-chunked-prefill. and the constriant between args max-num-seqs, max-num-batched-tokens and max-model-len.
No constraint. I recommend you to use --max-num-batched-tokens, not --max-num-seqs when using chunked prefill though. I think it is easier to control the performance with --max-num-batched-tokens.
--max-model-len is unrelated to chunked prefill imo.
can I use --enable-chunked-prefill and --enable-prefix-caching at the same time?
It is not supported yet. There's an open issue. We should support it soon.
from vllm.
Related Issues (20)
- [Bug]: Mixtral8x7B very high spikes for Inter Token Latency (ITL) HOT 1
- [Usage]: How to specify a GPU when i use vllm by 'python -m vllm.entrypoints.openai.api_server --served-model-name..' HOT 2
- [Feature]: Decouple the benchmark script from the components of vLLM HOT 1
- [Installation]: `ModuleNotFoundError: No module named 'numpy.lib.function_base'` due to NumPy 2.0 release HOT 2
- [Usage]: 'InternVLChatConfig' object has no attribute 'num_attention_heads' HOT 2
- [Bug]: The "sorted(gpu_ids)" operation in ray_gpu_executor.py causes an incorrect order of GPU IDs When using the NVIDIA HGX A100 (16-GPU) platform for model inference. HOT 1
- [Bug]: Embedding doesn't work with `device="cpu"` HOT 1
- [bug]vllm deployement is failing on nvidia because of numpy2.0 upgrade HOT 1
- [Bug]: GPTQ-Marlin kernel illegal memory access with `group_size=32`, `desc_act=True`, `tp=4` HOT 4
- [Feature]: Add config of use_dummy_driver rather than default 'False' HOT 3
- [Feature]: support Qwen2 embedding HOT 2
- RuntimeError: No suitable kernel. h_in=16 h_out=4096 dtype=Float out_dtype=Half HOT 2
- [Feature]: Access to user information in scheduler
- [Bug]: Concurrent requests messing up GREEDY responses HOT 4
- [Feature]: Initial LLM token HOT 5
- [Bug]: Failed: /home/runner/work/vllm/vllm/csrc/custom_all_reduce.cuh:310 'invalid argument' HOT 4
- [Usage]: Qwen2-7B-Instruct got stuck in infinite loop using vllm==0.5.0 with tp = 2
- [Bug]: many models may not load the weights correctly if `tie_word_embeddings` is enabled
- Why is the GPU KV cache usage very low? HOT 1
- [Installation]: 是否支持在windows操作系统上安装vllm
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 vllm.