vllm
virtualagc
vllm | virtualagc | |
---|---|---|
31 | 13 | |
19,344 | 2,494 | |
12.6% | 1.1% | |
9.9 | 8.9 | |
1 day ago | 4 days ago | |
Python | Assembly | |
Apache License 2.0 | GNU General Public License v3.0 or later |
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
virtualagc
-
Mistral CEO confirms 'leak' of new open source AI model nearing GPT4 performance
I don't think that's a great example.
For instance, I can step through and even modify that code using tooling like AGC emulators like this one http://www.ibiblio.org/apollo/#gsc.tab=0
What makes it open source is access to the same level of source access that the original developers worked in.
That's what's missing here. Mistral's engineers do not simply open this binary in their editor to do their job.
- Exploring the Internals of Linux v0.01
-
Apollo 14 LMAE specs/exit velocity calculation
Glad you're enjoying the videos! I'm Mike from them. :) The Apollo 14 code I linked above was reconstructed in the same way as Luminary 69/2, but the process was far, far too involved for a video. Like I said above, I probably won't be much help with math, but I'd be more than happy to assist with finding documentation or navigating AGC code whenever you want!
-
Apollo 11 was FAKE. What is the best evidence to support this?
The fact that all the information about the computer and software is public and you can verify it yourself like thousands of people have. https://www.ibiblio.org/apollo/
-
TIL NASA landed on the moon using a computer with just 4KB of RAM
If you’re curious about the the technical specs of the AGC check out Ron Burkey’s website, it’s even got the source code for you to see for yourself: https://www.ibiblio.org/apollo/
- Ask HN: What weird technical scene are you fond/part of?
- The Apollo On-Board Computers
-
What is the evidence for the moon landing being fake?
Of course you should be able to tell us because all the hardware architecture and software can be found here. So you must have fully reviewed it and located the issues. Right? Than please point out where the error is.
- How did a Space Rocket take off again from the surface of Moon?
-
Exploring the software that flies SpaceX rockets and starships
It had manually encoded ROM in the form of "core rope memory", which is pretty wacky, but it was a digital computer. In fact, it was the first IC computer.
You can learn way too much about it and even operate (a simulation of) one here: http://www.ibiblio.org/apollo/
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.
ArduinoCore-avr - The Official Arduino AVR core
CTranslate2 - Fast inference engine for Transformer models
rvc - A 32-bit RISC-V emulator in a shader (and C)
lmdeploy - LMDeploy is a toolkit for compressing, deploying, and serving LLMs.
new-wave - Stack Computer Bytecode Interpreters: The New Wave
Llama-2-Onnx
CAM6 - Cellular Automata Machine (CAM6) Simulator
tritony - Tiny configuration for Triton Inference Server
CygnusX1 - A thrust-vectoring model rocket flight computer. Comes with all you need to keep your rocket pointing up.
faster-whisper - Faster Whisper transcription with CTranslate2
SVM-Face-and-Object-Detection-Shader - SVM using HOG descriptors implemented in fragment shaders