Hugging Face's logo

0xTank
/
SuperGLM-5.2-v2-W4W8-Overlay

Text Generation
Safetensors
vllm
glm-5.2
superglm
w4a16
w8a16
compressed-tensors
dgx-spark
speculative-decoding
overlay
conversational
Model card Files
xet
 Community
SuperGLM-5.2-v2-W4W8-Overlay / AGENT_INSTALL.md
0xTank's picture
0xTank
Add agent installation guide
5ef1362 verified
preview code
|
Raw
History Blame Contribute Delete
997 Bytes

Agent-directed installation

Give an infrastructure agent this Hugging Face URL and instruct it to read this file and the linked GitHub AGENTS.md completely before changing the cluster.

The overlay is not a standalone model. The supported autonomous path is:

git clone --branch v1.0.0 https://github.com/0xTank/SuperGLM-5.2-v2-W4W8-200K-4x-DGX-Spark
cd SuperGLM-5.2-v2-W4W8-200K-4x-DGX-Spark
cat AGENTS.md
./scripts/agent-install.sh --prepare
# Review the dry-run evidence and confirm disruption is authorized.
./scripts/agent-install.sh --launch

If config/cluster.env does not exist, the installer creates a template and stops. The agent must discover the actual four-node addresses, users, HCA, interface, and per-node GID indices. It must never copy another cluster’s private values.

Success requires a passing checkpoint receipt, four healthy RoCE-connected ranks, the exact SuperGLM v2 served name, 200K configured context, and a passing natural streaming smoke test.