vllm
axolotl
vllm | axolotl | |
---|---|---|
31 | 29 | |
19,344 | 5,987 | |
12.6% | 12.0% | |
9.9 | 9.8 | |
2 days ago | 1 day ago | |
Python | Python | |
Apache License 2.0 | Apache License 2.0 |
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
axolotl
-
Ask HN: Most efficient way to fine-tune an LLM in 2024?
The approach I see used is axolotl with QLoRA using cloud GPUs which can be quite cheap.
https://github.com/OpenAccess-AI-Collective/axolotl
- FLaNK AI - 01 April 2024
-
LoRA from Scratch implementation for LLM finetuning
https://github.com/OpenAccess-AI-Collective/axolotl
- Optimized Triton Kernels for full fine tunes
- Axolotl
-
Let’s Collaborate to Build a High-Quality, Open-Source Dataset for LLMs!
One option is to look at what Axolotl uses. They have a list of different dataset formats that they support. They're mostly in JSON with specific field names, so you could start putting a dataset together with a text editor or a JSON editor.
- Axolotl: Streamline fine-tuning of AI models
-
Dataset Creation Tools?
You can save that overall set into a json file and load it up as training data in whatever you're using. I'm using axolotl for it at the moment. Though a GUI based option is probably best for the first couple of tries until you get a feel for the options.
-
Progress on Reproducing Phi-1/1.5
Looking forward to the results! If it turns out the dataset is reproducible, then it might be a good candidate for ReLora training on axolotl!
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.
signal-cli - signal-cli provides an unofficial commandline, JSON-RPC and dbus interface for the Signal messenger.
CTranslate2 - Fast inference engine for Transformer models
gpt-llm-trainer
lmdeploy - LMDeploy is a toolkit for compressing, deploying, and serving LLMs.
LoRA - Code for loralib, an implementation of "LoRA: Low-Rank Adaptation of Large Language Models"
Llama-2-Onnx
mlc-llm - Enable everyone to develop, optimize and deploy AI models natively on everyone's devices.
tritony - Tiny configuration for Triton Inference Server
LMFlow - An Extensible Toolkit for Finetuning and Inference of Large Foundation Models. Large Models for All.
faster-whisper - Faster Whisper transcription with CTranslate2
koboldcpp - A simple one-file way to run various GGML and GGUF models with KoboldAI's UI