Comments (5)
@DarkLight1337 you were quicker by one hour, but you still have failing tests, so I win 😜
from vllm.
It's not a competition xD We can combine our solutions in your PR if need be.
from vllm.
I know! I was being (not so) funny.
from vllm.
I have updated my PR with more test cases. From my understanding, the behaviour of disabling logprobs
and specifying zero top logprobs
should be distinct. In particular:
- Disabling
logprobs
should return nologprobs
at all:- Completions API: Input
logprobs=None
should result in outputtop_logprobs==None
- Chat Completions API: Input
logprobs=False
should result in outputlen(top_logprobs)==0
- Completions API: Input
- Specifying zero top
logprobs
should return thelogprob
for the output token only for Completions API (if it exists):- Completions API: Input
logprobs=0
should result in outputlen(top_logprobs)<=1
- Chat Completions API: Input
logprobs=True,top_logprobs=0
should result in outputlen(top_logprobs)<=k
len(top_logprobs)==0
- Completions API: Input
- Specifying
k
toplogprobs
should return the topk
items, plus thelogprob
for the output token only for Completions API (if it exists):- Completions API: Input
logprobs=k
should result in outputlen(top_logprobs)<=k+1
- Chat Completions API: Input
logprobs=True,top_logprobs=k
should result in outputlen(top_logprobs)<=k+1
len(top_logprobs)==k
- Completions API: Input
Edit: Thanks @br3no for the correction!
from vllm.
My PR #5026 now passes all tests as well.
from vllm.
Related Issues (20)
- max_tokens must be at least 1, got -160
- [Performance]: About the use of flash_attn_varlen_func() HOT 1
- [Usage]: NVIDIA多型号的GPU如何利用到? HOT 1
- [Feature]: support torch 2.3.1 HOT 1
- [Misc]: 我在使用vllm启动的openai api在进行对话时出现这样的情况 HOT 2
- [Feature]: Continuous streaming of `UsageInfo` HOT 1
- [Usage]: Running Llama 3 70B on A100 GPU - Tried to allocate 160MiB. GPU HOT 2
- [Usage]: Is this an error ? "async_llm_engine.py:154] Aborted request cmpl-xxxxx"
- [Bug]: "Triton Error [CUDA]: device kernel image is invalid" when loading Mixtral-8x7B-Instruct-v0.1 in fused_moe.py HOT 1
- [RFC]: proper resource cleanup for LLM class with file-like usage HOT 10
- [New Model]: Chameleon support HOT 1
- [Feature]: Support Nemotron-4-340B HOT 1
- [RFC]: Add runtime weight update API HOT 3
- [Usage]: qwen2-1.5b-gptq-in4 single gpu multiprocessing deployment fail
- [Bug]: Two V100 server with a total of 16GPU running Distributed Inference and Serving Vllm with error HOT 7
- [Misc]: how to understand: NUM_ELEMS_PER_THREAD = HEAD_SIZE / THREAD_GROUP_SIZE
- [Bug]: asyncio.exceptions.CancelledError asyncio.exceptions.TimeoutError HOT 1
- api_server.py: error: unrecognized arguments: --tool-use-prompt-template --enable-api-tools --enable-auto-tool-choice HOT 1
- [Bug]: RuntimeError: CUDA error: CUBLAS_STATUS_EXECUTION_FAILED when calling cublasLtMatmul with transpose_mat1 t transpose_mat2 n m 9216 n 3398 k 7168 mat1_ld 7168 mat2_ld 7168 result_ld 9216 computeType 68 scaleType 0
- [Bug]: asyncio.exceptions.CancelledError asyncio.exceptions.TimeoutError 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 vllm.