4 ms·
We saw this before. Anyone know what the VRAM requirements are, if it’s been quantized, and if/when llama.cpp might support?
by azinman2 3y ago
We saw this before. Anyone know what the VRAM requirements are, if it’s been quantized, and if/when llama.cpp might support?
- visarga 3y agoProbably never gonna run on normal boxes, it uses RingAttention which only makes sense when you have many GPUs. It's a datacenter-only model.
- azinman2 3y agoI’d still like to know the requirements. You can rent A100’s or have access thru work, etc.
- ramoz 3y ago> Scaling Inference We additionally scale our inference code to support million-length sequences by implementing RingAttention for decoding. Inference for such long sequences requires a minimum of v4-128 with a TPU mesh sharding of 32 tensor parallelism, and 4 sequence parallelism (ring dimension). We perform inference in pure single precision, where additional improvements can be made through techniques in scalability such as quantization.
- Filligree 3y agoIt might even fit in a max-sized macbook pro.
- machinekob 3y agoMaybe in few years with 4bit quantisation. (around 1TiB is needed for that)
- 5kg 3y agoFrom https://arxiv.org/pdf/2402.08268.pdf https://arxiv.org/pdf/2402.08268.pdf > We trained our models using TPUv4-1024, which is approximately equivalent to 450 A100s > Inference for such long sequences requires a minimum of v4-128 So you'll need ~60 A100 for inference.
- K0balt 3y agoThere are models with shorter attention sizes that probably are much smaller in vram needs. The model itself is only about 32g, so in 5K-M quant it wouldn’t be that bad, and the smaller attention versions might actually be workable in higher end workstations with cleverness like llama.cpp. Back of the napkin guess tells me maybe around 60g for the 32k context version
- woadwarrior01 3y agoFrom the paper[1]: > We additionally scale our inference code to support million-length sequences by implementing RingAttention for decoding. Inference for such long sequences requires a minimum of v4-128 with a TPU mesh sharding of 32 tensor parallelism, and 4 sequence parallelism (ring dimension). Each TPU-v4 has 32 GiB of HBM memory, so about 4 TiB (128 x 32GiB) of memory in fp16, without quantization. [1]: https://arxiv.org/abs/2402.08268 https://arxiv.org/abs/2402.08268
- Mathnerd314 3y agoThe model itself is only 13.6GB though: https://huggingface.co/LargeWorldModel/LWM-Chat-1M-Jax/tree/main https://huggingface.co/LargeWorldModel/LWM-Chat-1M-Jax/tree/...
- woadwarrior01 3y agoIndeed. Almost all of the inference memory overhead comes from attention matrices and the KV cache.
- moffkalast 3y agoSo uh, they thought "we have this datacenter full of GPUs laying around, you know what would be best, if we made a model that needs the entire thing just for inference". Brute forcing a quadratic complexity problem seems like wastefulness at its worst.
- cyanydeez 3y agoaren't most LLMs just a brute force probability engine?
- 3abiton 3y agoSounds like they're loading the internet into those vrams.