Context window of this model?
#4
by
yiouyou
- opened
4096 as original llama2, or less?
Thanks!
It is originally the same 4096 as llama2. If you refer to our model card, you can use a sequence length even longer. The reason this model's config is 2048 is because meta-llama/Llama-2-70b-hf had it recorded as 2048 in the config when we were training. (They later modified the config)