Morph MORPH
SIGN IN SIGN UP
dhruvhead / glm-5.2-sm120 UNCLAIMED
0 0 0 Shell

COMMITS

main
June 19, 2026
0
Default reasoning effort to high (was max) — fixes runaway thinking, 500s, empty content
0xSero committed
579d78b
0
Default to reasoning ON (it is a reasoning model): native template + --reasoning-parser glm45
0xSero committed
af3af3d
0
launch.sh: restart-tolerant boot wait (no false failure when --restart bounces the container mid-boot)
0xSero committed
82b3972
0
Serve as model id glm-5.2 (keep GLM-5.2-NVFP4-REAP-469B as alias)
0xSero committed
8035026
0
Forward serve logs + persist peak prefill/decode (monitor.sh)
0xSero committed
c04529f
0
Docs: lead with the DCP=4 flagship spec — 250k ctx, ~3x concurrency, ~60 tok/s
0xSero committed
9cd6634
June 18, 2026
0
Resilience: add restart policy (unless-stopped) so the server self-heals after a kill
0xSero committed
df3d60d
0
Make it just work: thinking OFF by default (direct answers), self-verifying launch.sh + chat.sh
0xSero committed
8fe3e26
0
Document DCP context-vs-speed tradeoff (measured): DCP=1 for fast <=128k, DCP=4 for 250k
0xSero committed
3aca172
0
Fix outsider failures: document reasoning-model max_tokens gotcha, portable MODEL mount, align port 8000, compose spec-config model field
0xSero committed
90369bb
0
README: link HF model card + add weights download step
0xSero committed
b60282d
0
GLM-5.2-NVFP4-REAP-469B vLLM serving for 4x RTX PRO 6000 (SM120)
0xSero committed
9222d02