â„–15394280[Quote]
Static key-value cache was used whenever supported. The same prompt was used for all runs, resulting in prompt lengths of 35-36 tokens (depending on the tokenizer).
Static key-value cache was used whenever supported. The same prompt was used for all runs, resulting in prompt lengths of 35-36 tokens (depending on the tokenizer).
Static key-value cache was used whenever supported. The same prompt was used for all runs, resulting in prompt lengths of 35-36 tokens (depending on the tokenizer).
Static key-value cache was used whenever supported. The same prompt was used for all runs, resulting in prompt lengths of 35-36 tokens (depending on the tokenizer).
Static key-value cache was used whenever supported. The same prompt was used for all runs, resulting in prompt lengths of 35-36 tokens (depending on the tokenizer).