$ ./ds4-agent \
 --backend rocm \
 --threads 30 \
 --model ~/models/antirez/deepseek-v4-gguf/DeepSeek-V4-Flash-IQ2XXS-w2Q2K-AProjQ8-SExpQ8-OutQ8-chat-v2-imatrix.gguf \
 --mtp ~/models/antirez/deepseek-v4-gguf/DeepSeek-V4-Flash-MTP-Q4K-Q8_0-F32.gguf \
 --mtp-draft 1 \
 --ctx 16000 \
ds4: MTP support model loaded: /home/sam/models/antirez/deepseek-v4-gguf/DeepSeek-V4-Flash-MTP-Q4K-Q8_0-F32.gguf (draft=1)
ds4: ROCm backend initialized on AMD Radeon Graphics (sm_115)
ds4: ROCm chunk-copying 80.76 GiB model image
ds4: ROCm loading model tensors into device cache: 80.06 GiBds4: ROCm model chunk copy complete in 18.271s (80.76 GiB tensors)
ds4: ROCm chunk-copying 3.55 GiB model image
ds4: ROCm loading model tensors into device cache: 2.06 GiBds4: ROCm model chunk copy complete in 0.782s (3.55 GiB tensors)
ds4: ROCm preparing model tensor mappings: 80.24 GiB
ds4: ROCm startup model preparation covered 80.76 GiB of tensor spans in 0.694s
ds4: ROCm preparing model tensor mappings: 2.38 GiB
ds4: ROCm startup model preparation covered 3.55 GiB of tensor spans in 0.012s
ds4: rocm backend initialized for graph diagnostics
DwarfStar 🐋 Agent, context 16k tokens
ds4-agent>