|
|
@@ -44,7 +44,7 @@
|
|
|
# hf download local-inference-lab/GLM-5.3-Flash-NVFP4
|
|
|
|
|
|
engine = "vllm"
|
|
|
-name = "GLM-5.3-Flash NVFP4-lab (vLLM, 2x Spark, MTP-3)"
|
|
|
+name = "GLM-5.3-Flash NVFP4-lab (vLLM, 2x Spark, LAB DFlash2-7)"
|
|
|
docker_image = "glm53-flash-nvfp4-lab:local"
|
|
|
docker_image_dir = "glm53-flash-nvfp4-lab"
|
|
|
model = "local-inference-lab/GLM-5.3-Flash-NVFP4"
|
|
|
@@ -84,7 +84,7 @@ args = [
|
|
|
'{"enable_cutedsl_warmup":false,"enable_flashinfer_autotune":false}'""",
|
|
|
# MTP-3 draft from the MXFP8 layer-45 shard (needs modelopt.patch in image)
|
|
|
"--speculative-config", """
|
|
|
-'{"method":"mtp","num_speculative_tokens":3}'""",
|
|
|
+'{"method":"dflash","model":"local-inference-lab/GLM-5.3-Flash-DFlash2","num_speculative_tokens":7}'""",
|
|
|
"--language-model-only",
|
|
|
# MTP + CUDA graphs degrades OUTPUT QUALITY on the glm53 day-0 dev build
|
|
|
# (forum #381350/#44, CONFIRMED by A/B bench 2026-08-28: same seed/flags,
|
