4 ms·
The model card has the benchmark results relative to other Llama models including Llama 2: https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md#base-pre
by minimaxir 3y ago
The model card has the benchmark results relative to other Llama models including Llama 2: https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md#base-pretrained-models https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md...
The dramatic performance increase of Llama 3 relative to Llama 2 (even Llama 2 13B!) is very impressive. Doubling the context window to 8k will open a lot of new oppertunities too.
- oersted 3y agoFor the instruction tuned models, Llama 3 8B is even significantly better than Llama 2 70B!
- rileyphone 3y agoTo be fair, the Llama 2 instruction tuning was notably bad.
- oersted 3y agoI see it more as an indirect signal for how good Llama 3 8B can get after proper fine-tuning by the community.
- seydor 3y agohow much vram does the 8B model use?
- cjbprime 3y agoIn general you can swap B for GB (and use the q8 quantization), so 8GB VRAM can probably just about work.
- derac 3y agoYou can use 5 bits per parameter with negligible loss of capability as a general rule. 4 bits for a tiny bit worse results. This is subject to changes in how good quantization is in general and on the specific model.
- loudmax 3y agoDisappointed to note that the 8k context length is far short of Mixtral 8x22B's 64k context length. Still, the published performance metrics are impressive. Kudos to Meta for putting these models out there.
- rising-sky 3y agoThey’re going to increase the context window https://www.threads.net/@zuck/post/C56MOZ3xdHI/?xmt=AQGzjzazsycd7hUW10eWagOXTUp4Tq351z9JKlDvMU0JKA https://www.threads.net/@zuck/post/C56MOZ3xdHI/?xmt=AQGzjzaz...