The latest version (2026-07-26) drops a segmentation fault if I try to run:
~/ds4/ds4-server -m ~/models/DeepSeek-V4-Flash-Q4KExperts-F16HC-F16Compressor-F16Indexer-Q8Attn-Q8Shared-Q8Out-chat-v2-imatrix.gguf --role coordinator --layers 0:14 --listen 0.0.0.0 10000 -c 1048576 --host 0.0.0.0 --port 8000
The environment is a 128Gb Strix Halo box with Debian and Rocm 7.2
The binary is built with make rocm
The last commit I can build and start is: 519c4d8fe71a67bc80bba6e7b034d0dadc2114b5 (Fix PRO streaming and sampling correctness)
My guess is, the GLM 5.2 support broke something.
The latest version (2026-07-26) drops a segmentation fault if I try to run:
~/ds4/ds4-server -m ~/models/DeepSeek-V4-Flash-Q4KExperts-F16HC-F16Compressor-F16Indexer-Q8Attn-Q8Shared-Q8Out-chat-v2-imatrix.gguf --role coordinator --layers 0:14 --listen 0.0.0.0 10000 -c 1048576 --host 0.0.0.0 --port 8000The environment is a 128Gb Strix Halo box with Debian and Rocm 7.2
The binary is built with
make rocmThe last commit I can build and start is:
519c4d8fe71a67bc80bba6e7b034d0dadc2114b5(Fix PRO streaming and sampling correctness)My guess is, the GLM 5.2 support broke something.