paul_mk1

I don't think there is anything conceptually new in this work, other than it is applied to LLMs.

But in fairness, getting these techniques to work at scale is no small feat. In my experience quantization aware training at these low bit depths was always finicky and required a very careful hand. I'd be interested to know if it has become easier to do, now that there are so many more parameters in LLMs.

In any case full kudos to the authors and I'm glad to see people continuing this work.

Nice to know there is a trail to relevant citations. I missed the BitNet paper and need to catch up.

Btw TrueNorth project evolved into "NorthPole" chip by the same group, and was recently in the press. From afar NorthPole looks like an interesting design point and leverages on-chip memory (SRAM)--so it's targeting speed and efficiency at the expense of memory density (so perhaps like Groq in some respects). Tbh I haven't followed the field closely after leaving the group.

Comparisons between different chip architectures are imperfect. In our opinion the most fair thing to do is 1) match the TFLOPs (since these workloads are compute bound), and 2) find a similar card that can run the same size models. Since MI210 has 64GB, the closest one is the A6000 with 48GB. Many of the less expensive ADA cards can't even fit a 13B model in VRAM, so a comparison could not even be made.

This is a good observation, the cards do have different memory bandwidth with the MI210 having more than double the bandwidth via HBM2e.

Note that the comparisons between the two cards (MI210 and A6000) are being made for high throughput workloads, and in this regime the performance is compute bound. So as long as the memory bandwidth is decent (as it is for GDDR6 with 768.0 GB/s) the lower memory bandwidth is not the main bottleneck. There are also other architectural differences so any comparison will ultimately be imperfect, but we found A6000 to be the closest match for the workloads that we cared about (ie high throughput workloads at reasonable latencies).

Also worth noting that there are still more stack optimizations on the table, which again can shift the bottleneck between compute and memory. In those cases it might makes sense to compare with another card with matched memory bandwidth.

paul @ mk1

>The "-ngl 32" means that only 32 out of 35 layers are being run on the GPU, and this results in a huge slow down as the GPU syncs with the CPU, and then computes the last 3 layers on the CPU.

Thanks for the updated run configuration. It was a misunderstanding on our part about what llama.cpp considers “layers”, since layers are traditionally understood as learned parameter decoder layers (as they do in Hugging Face models). And, in this case the llama 7B model has 32 layers.

>On my XTX 7900, I get a 55% speed up on llama.cpp (to 132.61 tok/sec) when running all layers on the GPU, rather than only 32 as in your measurements.

On my 4090 I now get 128 t/s for Q5_1, and 116 t/s for Q6_k. So these are ballpark to mk600's 125 t/s for batch=1. Not surprising that different inference runtimes approach the same speed as they become memory bound for similar model sizes.

Hi, one of the founders here.

Attempting to address some of the comments in a single message.

To help understand why we decided not to compare to existing methods: I think it would be difficult to do so fairly, since there are many tradeoffs and different use cases. It's not always the case that one technique is bad and the other is good, it's more about the targeted design point (say, cloud vs local). We are openly offering our numbers / benchmarks and looking for early partners that are aligned with our current value proposition (hence the closed beta).

A good example is that llama.cpp is a fantastic framework to run models locally for the single-user case (batch=1). While llama.cpp supports different backends (RPi, CPU, GPU), I don't think it would be particularly fair to compare and show that MKML is better at a given perplexity, compression ratio, and speed on GPU for a multi-user case (batch >> 1), when that is not llama.cpp’s targeted use case (afaik). For example MKML achieves ~2700 tok/sec at batch 32 (i.e. 32 prompts in parallel) on a 4090 for a Llama-2 7B, with a ~4̶.̶2̶G̶B 5.2GB memory footprint, and perplexity that is ~fp16.

Also, we're not currently wrapping any open source tools or techniques for quantization. Everything is our own and there’s more news to come soon.

If anyone has specific technical questions I'd be happy to answer as best I can.

Cheers, Paul Merolla