DeepSeek-R1-UD-IQ1_S merge
https://unsloth.ai/blog/deepseekr1-dynamic
https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/DeepSeek-R1-UD-IQ1_S
Will it be possible to merge this 1.58 bit dynamic quant model with SkyT1-Flash?
https://unsloth.ai/blog/deepseekr1-dynamic
https://huggingface.co/unsloth/DeepSeek-R1-GGUF/tree/main/DeepSeek-R1-UD-IQ1_SWill it be possible to merge this 1.58 bit dynamic quant model with SkyT1-Flash?
gguf is not model
Yes, these GGUF-Version of DeepSeek-R1 still has different architectures and scales with SkyT1-Flash. Thus we can not direct merge these models. However, we have new techniques for heterogeneous model fusion with different architectures and scales. We try to release a new version of FuseO1 with heterogeneous model fusion and reinforcement learning approaches. Stay tuned!
Yes, these GGUF-Version of DeepSeek-R1 still has different architectures and scales with SkyT1-Flash. Thus we can not direct merge these models. However, we have new techniques for heterogeneous model fusion with different architectures and scales. We try to release a new version of FuseO1 with heterogeneous model fusion and reinforcement learning approaches. Stay tuned!
Very interesting, looking forward! Thank you