no thinking process?
the model works fine in politic issues, but compare to v1, v2 hasn't the thinking process, just throw out the final result.
launch command: ./llama.cpp/llama-server --model /dockerdata/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-i1-GGUF/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased.i1-Q4_K_M.gguf --port 8080 --ctx-size 10240 --n-gpu-layers 90 --host 0.0.0.0 --threads 30
system prompt had added.
Thanks a lot for the feedback. I'm aware of it and am currently working on reasoner finetuning this model to bring back the original models reasoning capabilities. Expect this to take around another week as in a first try I used a bad reasoning dataset that did not lead to the desired results and the one I plan on using in my next try is relatively big. If you don't want to wait feel free to finetune it by your own on dolphin-r1-reasoning-deepseek.jsonl of https://huggingface.co/datasets/cognitivecomputations/dolphin-r1
You are doing the great job, thanks a lot
@aaa1991 The uncensored unbiased reasoning finetune of DeepSeek-R1-Distill-Llama-70B is now available under. https://huggingface.co/nicoboss/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-Reasoner - it turned out very well. Make sure to use the system prompt specified in its model card when testing it.
Hi guys whats up, still no ideas how to deploy decentralized blockchain ASI?