vllm
LAVIS
vllm | LAVIS | |
---|---|---|
31 | 18 | |
19,344 | 8,838 | |
12.6% | 3.5% | |
9.9 | 6.3 | |
1 day ago | 25 days ago | |
Python | Jupyter Notebook | |
Apache License 2.0 | BSD 3-clause "New" or "Revised" License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
vllm
-
AI leaderboards are no longer useful. It's time to switch to Pareto curves
I guess the root cause of my claim is that OpenAI won't tell us whether or not GPT-3.5 is an MoE model, and I assumed it wasn't. Since GPT-3.5 is clearly nondeterministic at temp=0, I believed the nondeterminism was due to FPU stuff, and this effect was amplified with GPT-4's MoE. But if GPT-3.5 is also MoE then that's just wrong.
What makes this especially tricky is that small models are truly 100% deterministic at temp=0 because the relative likelihoods are too coarse for FPU issues to be a factor. I had thought 3.5 was big enough that some of its token probabilities were too fine-grained for the FPU. But that's probably wrong.
On the other hand, it's not just GPT, there are currently floating-point difficulties in vllm which significantly affect the determinism of any model run on it: https://github.com/vllm-project/vllm/issues/966 Note that a suggested fix is upcasting to float32. So it's possible that GPT-3.5 is using an especially low-precision float and introducing nondeterminism by saving money on compute costs.
Sadly I do not have the money[1] to actually run a test to falsify any of this. It seems like this would be a good little research project.
[1] Or the time, or the motivation :) But this stuff is expensive.
-
Mistral AI Launches New 8x22B Moe Model
The easiest is to use vllm (https://github.com/vllm-project/vllm) to run it on a Couple of A100's, and you can benchmark this using this library (https://github.com/EleutherAI/lm-evaluation-harness)
- FLaNK AI for 11 March 2024
-
Show HN: We got fine-tuning Mistral-7B to not suck
Great question! scheduling workloads onto GPUs in a way where VRAM is being utilised efficiently was quite the challenge.
What we found was the IO latency for loading model weights into VRAM will kill responsiveness if you don't "re-use" sessions (i.e. where the model weights remain loaded and you run multiple inference sessions over the same loaded weights).
Obviously projects like https://github.com/vllm-project/vllm exist but we needed to build out a scheduler that can run a fleet of GPUs for a matrix of text/image vs inference/finetune sessions.
disclaimer: I work on Helix
-
Mistral CEO confirms 'leak' of new open source AI model nearing GPT4 performance
FYI, vLLM also just added experiment multi-lora support: https://github.com/vllm-project/vllm/releases/tag/v0.3.0
Also check out the new prefix caching, I see huge potential for batch processing purposes there!
- VLLM Sacrifices Accuracy for Speed
- Easy, fast, and cheap LLM serving for everyone
- vllm
- Mixtral Expert Parallelism
- Mixtral 8x7B Support
LAVIS
- FLaNK AI for 11 March 2024
- FLaNK 04 March 2024
-
[D] Why is most Open Source AI happening outside the USA?
For multimodal, there's China (*many), then Salesforce.
-
Need help for a colab notebook running Lavis blip2_instruct_vicuna13b?
Been trying for all day to get a working inference for this example: https://github.com/salesforce/LAVIS/tree/main/projects/instructblip
-
most sane web3 job listing
There's also been big breakthroughs in computer vision. Not that long ago it was hard to recognize if a photo contained a bird; that's solved now by models like CLIP, Yolo, or Segment Anything. Now research has moved on to generating 3D scenes from images or interactively answering questions about images.
-
I work at a non-tech company and have been asked to make software that is impossible. How do I explain it to my boss?
The new hotness is multimodal vision-language models like InstructBLIP that can interactively answer questions about images. Check out the examples in the github repo, I would not have thought this was possible a few years ago.
-
Two-minute Daily AI Update (Date: 5/15/2023)
Salesforce’s BLIP family has a new member– InstructBLIP, a vision-language instruction-tuning framework using BLIP-2 models. It has achieved state-of-the-art zero-shot generalization performance on a wide range of vision-language tasks, substantially outperforming BLIP-2 and Flamingo. (Source)
-
InstructBLIP: Towards General-purpose Vision-Language Models with Instruction Tuning
Github
-
Can I use my own art as a training set?
Most of my workflows are self-made. For captioning I used Blip-2 in a custom script I made that automates the process by going into directories and their sub-directories and creates a .txt file beside each image. This way I can keep my images organized in their proper directories, without having to put dump them all in a single place.
- FLiP Stack Weekly for 13-Feb-2023
What are some alternatives?
TensorRT - NVIDIA® TensorRT™ is an SDK for high-performance deep learning inference on NVIDIA GPUs. This repository contains the open source components of TensorRT.
pytorch-widedeep - A flexible package for multimodal-deep-learning to combine tabular data with text and images using Wide and Deep models in Pytorch
CTranslate2 - Fast inference engine for Transformer models
CLIP-Caption-Reward - PyTorch code for "Fine-grained Image Captioning with CLIP Reward" (Findings of NAACL 2022)
lmdeploy - LMDeploy is a toolkit for compressing, deploying, and serving LLMs.
sparseml - Libraries for applying sparsification recipes to neural networks with a few lines of code, enabling faster and smaller models
Llama-2-Onnx
robo-vln - Pytorch code for ICRA'21 paper: "Hierarchical Cross-Modal Agent for Robotics Vision-and-Language Navigation"
tritony - Tiny configuration for Triton Inference Server
DeepViewAgg - [CVPR'22 Best Paper Finalist] Official PyTorch implementation of the method presented in "Learning Multi-View Aggregation In the Wild for Large-Scale 3D Semantic Segmentation"
faster-whisper - Faster Whisper transcription with CTranslate2
linkis - Apache Linkis builds a computation middleware layer to facilitate connection, governance and orchestration between the upper applications and the underlying data engines.