mirror of
				https://github.com/ggml-org/llama.cpp.git
				synced 2025-10-31 08:51:55 +00:00 
			
		
		
		
	| @@ -2142,7 +2142,8 @@ struct server_context { | ||||
|  | ||||
|         // find the slot that has been least recently used | ||||
|         if (ret == nullptr) { | ||||
|             int64_t t_last = ggml_time_us(); | ||||
|             int64_t t_last = -1; | ||||
|  | ||||
|             for (server_slot & slot : slots) { | ||||
|                 // skip the slot if it is not available | ||||
|                 if (slot.is_processing()) { | ||||
| @@ -2150,7 +2151,7 @@ struct server_context { | ||||
|                 } | ||||
|  | ||||
|                 // select the current slot if the criteria match | ||||
|                 if (slot.t_last_used < t_last) { | ||||
|                 if (!ret || slot.t_last_used <= t_last) { | ||||
|                     t_last = slot.t_last_used; | ||||
|                     ret = &slot; | ||||
|                 } | ||||
|   | ||||
		Reference in New Issue
	
	Block a user
	 Georgi Gerganov
					Georgi Gerganov