exllamav2
OmniQuant
exllamav2 | OmniQuant | |
---|---|---|
17 | 4 | |
3,010 | 583 | |
- | 9.9% | |
9.8 | 7.7 | |
3 days ago | about 2 months ago | |
Python | Python | |
MIT License | MIT 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.
exllamav2
- Running Llama3 Locally
-
Mixture-of-Depths: Dynamically allocating compute in transformers
There are already some implementations out there which attempt to accomplish this!
Here's an example: https://github.com/silphendio/sliced_llama
A gist pertaining to said example: https://gist.github.com/silphendio/535cd9c1821aa1290aa10d587...
Here's a discussion about integrating this capability with ExLlama: https://github.com/turboderp/exllamav2/pull/275
And same as above but for llama.cpp: https://github.com/ggerganov/llama.cpp/issues/4718#issuecomm...
-
What do you use to run your models?
Sorry, I'm somewhat familiar with this term (I've seen it as a model loader in Oobabooga), but still not following the correlation here. Are you saying I should instead be using this project in lieu of llama.cpp? Or are you saying that there is, perhaps, an exllamav2 "extension" or similar within llama.cpp that I can use?
-
I just started having problems with the colab again. I get errors and it just stops. Help?
EDIT: I reported the bug to the exllamav2 Github. It's actually already fixed, just not on any current built release.
-
Yi-34B-200K works on a single 3090 with 47K context/4bpw
install exllamav2 from git with pip install git+https://github.com/turboderp/exllamav2.git. Make sure you have flash attention 2 as well.
-
Tested: ExllamaV2's max context on 24gb with 70B low-bpw & speculative sampling performance
Recent releases for exllamav2 brings working fp8 cache support, which I've been very excited to test. This feature doubles the maximum context length you can run with your model, without any visible downsides.
-
Show HN: Phind Model beats GPT-4 at coding, with GPT-3.5 speed and 16k context
Without batching, I was actually thinking that's kind of modest.
ExllamaV2 will get 48 tokens/s on a 4090, which is much slower/cheaper than an H100:
https://github.com/turboderp/exllamav2#performance
I didn't test codellama, but the 3090 TI figures are in the ballpark of my generation speed on a 3090.
-
Guide for Llama2 70b model merging and exllama2 quantization
First, you need the convert.py script from turboderp's Exllama2 repo. You can read all about the convert.py arguments here.
-
LLM Falcon 180B Needs 720GB RAM to Run
> brute aggressive quantization
Cutting edge quantization like ExLlama's EX2 is far from brute force: https://github.com/turboderp/exllamav2#exl2-quantization
> The format allows for mixing quantization levels within a model to achieve any average bitrate between 2 and 8 bits per weight. Moreover, it's possible to apply multiple quantization levels to each linear layer, producing something akin to sparse quantization wherein more important weights (columns) are quantized with more bits. The same remapping trick that lets ExLlama work efficiently with act-order models allows this mixing of formats to happen with little to no impact on performance. Parameter selection is done automatically by quantizing each matrix multiple times, measuring the quantization error (with respect to the chosen calibration data) for each of a number of possible settings, per layer. Finally, a combination is chosen that minimizes the maximum quantization error over the entire model while meeting a target average bitrate.
Llama.cpp is also working on a feature that let's a small model "guess" the output of a big model which then "checks" it for correctness. This is more of a performance feature, but you could also arrange it to accelerate a big model on a small GPU.
- 70B Llama 2 at 35tokens/second on 4090
OmniQuant
-
Run Mistral 7B on M1 Mac
Not on iOS. On macOS, I personally think WizardLM 13B v1.2 is a very strong model and keep hearing good things about it from users on our discord and in support emails. Now that there's OmniQuant support for Mixtral models[1], I'm plan to add support for Mixtral-8x7B-Instruct-v0.1 in the next version of the macOS app, which in my tests, looks like a very good all purpose model that's also pretty good at coding. It's pretty memory hungry (~41GB of RAM), but that's the price to pay for an uncompromising implementation. Existing quantized implementations quantize the MoE gates, leading to a significant drop in perplexity when compared with results from fp16 inference.
[1]: https://github.com/OpenGVLab/OmniQuant/commit/798467
- OmniQuant of Falcon-180B has been released!
-
70B Llama 2 at 35tokens/second on 4090
I think OmniQuant is notable because it shifts the bend of the curve to 3-bit. While < 3-bit still ramps up, it's notable in that it's usable and doesn't go asymptotic: https://github.com/OpenGVLab/OmniQuant/blob/main/imgs/weight...
What EXL2 seems to bring to the table is that you can target an arbitrary quantize bit-weight (eg, if you're a bit short on VRAM, you don't need to go from 4->3 or 3->2, but can specify say 3.75bwp). You have some control w/ other schemes by setting group size, or with k-quants, but EXL2 is definitely allows you to be finer grained. I haven't gotten a chance to sit down with EXL2 yet, but if no one else does it, it's on my todo-list to be able to do 1:1 perplexity and standard benchmark evals on all the various new quantization methods, just as a matter of curiosity.
- OmniQuant: Omnidirectionally Calibrated Quantization for Large Language Models
What are some alternatives?
llama.cpp - LLM inference in C/C++
gptq - Code for the ICLR 2023 paper "GPTQ: Accurate Post-training Quantization of Generative Pretrained Transformers".
koboldcpp - A simple one-file way to run various GGML and GGUF models with KoboldAI's UI
Cgml - GPU-targeted vendor-agnostic AI library for Windows, and Mistral model implementation.
SillyTavern - LLM Frontend for Power Users.
llamafile - Distribute and run LLMs with a single file.
ChatGPT-AutoExpert - 🚀🧠💬 Supercharged Custom Instructions for ChatGPT (non-coding) and ChatGPT Advanced Data Analysis (coding).
BlockMerge_Gradient - Merge Transformers language models by use of gradient parameters.
hkhomekit
refact - WebUI for Fine-Tuning and Self-hosting of Open-Source Large Language Models for Coding
ollama - Get up and running with Llama 3, Mistral, Gemma, and other large language models.