Open-Orca/Mistral-7B-OpenOrca · Hugging Face
huggingface.co
We’re on a journey to advance and democratize artificial intelligence through open source and open science.

This release is trained on a curated filtered subset of most of our GPT-4 augmented data.

HF Leaderboard evals place this model as #2 for all models smaller than 30B at release time, outperforming all but one 13B model.

GGUF files:

Mistral-7B-OpenOrca-GGUF

Warning (if I’m not mistaken):

Llama.cpp hasn’t assigned high priority tag to the sliding window. Axolotl replaced Mistral’s attention block by a “simple” flash attention.

That implies, in my opinion, that the new releases do not capitalize on the speedup claimed by Mistral developers.

We can’t expect the new versions to be faster than Llama, because there is no sliding attention to speed up inference.

  • justynasty@lemmy.kya.moeOPEnglish
    5·
    1 year ago

    Mistral 7B uses a sliding window attention (SWA) mechanism (Child et al., Beltagy et al.), in which each layer attends to the previous 4,096 hidden states. The main improvement, and reason for which this was initially investigated, is a linear compute cost of O(sliding_window.seq_len). In practice, changes made to FlashAttention and xFormers yield a 2x speed improvement for sequence length of 16k with a window of 4k. Source: Mistral 7B news For longer prompts.

    Talk about merging changes