Add SGLang serving instructions
#5
by sh-25 - opened
Adds SGLang as a supported runtime and documents the validated high-throughput aggregated TP4 serving command from the GLM-5.3-Flash cookbook.
Validation:
- SGLang 0.0.0.dev1+g008403017 on 4x NVIDIA GB300 GPUs
- High Throughput: full GSM8K test split, 1319/1319 examples, 97.19% accuracy, 0 request errors
- Low Latency adaptive EAGLE: full GSM8K test split, 1319/1319 examples, 96.66% accuracy, 0 request errors
- Cookbook-protocol random performance: all High Throughput metrics within 2.89% of the cookbook reference
This PR intentionally documents the High Throughput command without MTP/speculative decoding. Addresses discussion #2.
zhiyucheng changed pull request status to merged