LibertAIDAI/GLM-5.3-Flash-NVFP4 · Discussions

Resources

View closed (1)

Error with CUDA Graph Memory Calculations on T<4 set ups

#10 opened about 1 month ago by

The model's single-inference speed measured on two DGX systems is only 16 tokens/s, which does not meet productivity requirements.

#9 opened about 1 month ago by

Datapoint: fp4 vs fp8 GLM-5.3-Flash endpoint parity on GSM8K + HumanEval

#8 opened about 1 month ago by

Repetition

#7 opened about 1 month ago by

GB10 tile values confirmed for DFLASH verify shape + a memory-config note for spec decoding

#6 opened about 2 months ago by

GLM-5.3-Flash (NVFP4) on 4× DGX Spark — Multi-Node TP4 Benchmark (5.3–7.0 tok/s)

#5 opened about 2 months ago by

Deployment blocked: FlashInfer "Unsupported architecture" on DGX Spark (SM121)

#4 opened about 2 months ago by

Sm120/rtx6000pro supported?

#3 opened about 2 months ago by

Is sglang supported?

#1 opened about 2 months ago by