Initial DeepSeek-3.1-Base changes

[createthis]

Not working properly, but inference starts at least.

[createthis]

Apparently this does work properly, it's just how base models behave. Example usage:

./build/bin/llama-cli \
    --model /data2/DeepSeek-V3.1-Base-GGUF/q4_k_m/DeepSeek-V3.1-Base-Q4_K_M-00001-of-00009.gguf \
    --numa numactl \
    --threads 32 \
    --ctx-size 163840 \
    --n-gpu-layers 62 \
    -ot "blk\.(3|4|5|6|7|8|9|10|11)\.ffn_.*=CUDA0" \
    -ot exps=CPU \
    -ub 4096 -b 4096 \
    --seed 3407 \
    --temp 0.3 \
    --min-p 0.0 \
    --log-colors \
    --flash-attn \
    --jinja \
    --mlock \
    --prio 2 \
    -p "You are a helpful assistant.<｜User｜>Hello, how are you?<｜Assistant｜>I'm doing well thanks! Yourself?<｜User｜>I'm doing great! Can you explain the laws of thermodynamics?<｜Assistant｜>" \
    -no-cnv \
    -ngl 0

Stolen from the model card at https://huggingface.co/bartowski/deepseek-ai_DeepSeek-V3.1-Base-Q4_K_M-GGUF

TIL. Closing, probably not useful until the Instruct comes out.