R0CKSTAR

icon indicating an email [email protected]

icon company @MooreThreads icon location Beijing, China icon location Building GPU Clusters @MooreThreads AI | Cloud Native | Virtualization | macOS/iOS - ex @sangfor / @vmware

Results 67 comments of R0CKSTAR

feat: Support Moore Threads GPU

> ## Error Message > initializing /usr/local/musa/lib/libmusa.so.1.0.0 dlsym: muInit - 0x7f87cc513910 dlsym: muDriverGetVersion - 0x7f87cc51a3f0 dlsym: muDeviceGetCount - 0x7f87cc51a2b0 dlsym: muDeviceGet - 0x7f87cc51a140 dlsym: muDeviceGetAttribute - 0x7f87cc51a600 dlsym: muDeviceGetUuid_v2 -...

feat: Support Moore Threads GPU

Yes, the issue you identified is exactly what we have resolved in MUSA SDK version rc3.1.1. Feel free to add me through WeChat: yeahdongcn for more info.

Eval bug: Loading fail on Gemma 3:12b > llama_model_load: error loading model: error loading model hyperparameters: key not found in model: gemma3.attention.layer_norm_rms_epsilon

The same issue occurs here when using Ollama gemma3:4b.

Eval bug: Loading fail on Gemma 3:12b > llama_model_load: error loading model: error loading model hyperparameters: key not found in model: gemma3.attention.layer_norm_rms_epsilon

```bash modelscope download --local_dir $PWD lmstudio-community/gemma-3-4b-it-GGUF --include *Q4_K_M* ./llama-cli -m /data/.yeahdongcn/models/gemma-3-4b-it-Q4_K_M.gguf -ngl 999 ``` This works for me.

[Feature Request] Pull model through Ollama API instead of invoking ollama binary

I sent a PR to query device information through Ollama API: https://github.com/ollama/ollama/pull/5479 It could be used to replace GPUtil to check the available VRAM.

Support addtional filter for extended resource pod

This looks super useful for GPU clusters.

fix: Stop reconciling if K8sGPT CR is not in the operator namespace

> > created deployment can't be scaled to the minimum replicas (sa k8sgpt not found) > > so the real issue here is the Service Account is installed in the...