psydok
psydok
This error if I don't manually switch the checkpoint to dreamshaper_v8.safetensors. I had juggernautXL_v8.safetensors for the xl. ``` 2024-10-08T08:20:50: [Unload] Trying to free all memory for cuda:0 with 0 models...
Also, can you tell me if this is a normal speed for the flux model? It seems to take a very long time to generate at 40-90 seconds. ``` “forge_preset":...
Figured out that to change the model you need to `POST /sdapi/v1/options`. But what will happen when several requests for different models come to the service at the same time,...
It worked! Thank you! https://github.com/lllyasviel/stable-diffusion-webui-forge/pull/2054
Подскажите, пожалуйста, удалось ли решить проблему?
Я заметил, что если отправлять большой файл, но просить распарсить только 2 страницы, то OOM не падает. Возможно есть какая-нибудь фича, чтобы включить парсинг файлов от 20-30мб чанками (по странично)...
At same time, response time increases for all tools. therefore, decrease in rpm from llmperf seems to justify it. but vllm's rpm does not change (if you do not look...
Sorry, more indicative RPM chart turned out to be 300/200 = input/output tokens. 