|
@@ -39,9 +39,9 @@
|
|
|
|
|
|
|
|
engine = "vllm"
|
|
engine = "vllm"
|
|
|
name = "GLM-5.3-Flash NVFP4-Spark (vLLM, 4x Spark, MTP-3, r38 tip)"
|
|
name = "GLM-5.3-Flash NVFP4-Spark (vLLM, 4x Spark, MTP-3, r38 tip)"
|
|
|
-docker_image = "glm53-jovian-r38:local"
|
|
|
|
|
|
|
+docker_image = "glm53-jovian-r35:local" # r35 loads the Spark checkpoint (r38 pool-deadlock, see bug report)
|
|
|
docker_image_dir = "glm53-jovian-r27"
|
|
docker_image_dir = "glm53-jovian-r27"
|
|
|
-build_profile = "jovian-r38-glm53"
|
|
|
|
|
|
|
+build_profile = "glm53-r35"
|
|
|
entrypoint = ["vllm", "serve"]
|
|
entrypoint = ["vllm", "serve"]
|
|
|
model = "local-inference-lab/GLM-5.3-Flash-NVFP4-Spark"
|
|
model = "local-inference-lab/GLM-5.3-Flash-NVFP4-Spark"
|
|
|
parent_model = "zai-org/GLM-5.3-Flash"
|
|
parent_model = "zai-org/GLM-5.3-Flash"
|
|
@@ -95,6 +95,8 @@ HF_HUB_OFFLINE = "1"
|
|
|
# r27 lanes since the 2026-09-08 A/B (4x: decode +31..+90%, quality 160 vs
|
|
# r27 lanes since the 2026-09-08 A/B (4x: decode +31..+90%, quality 160 vs
|
|
|
# 155). Needs the RoCE fabric; single-node serves skip it silently.
|
|
# 155). Needs the RoCE fabric; single-node serves skip it silently.
|
|
|
VLLM_ENABLE_ROCE_ALLREDUCE = "1"
|
|
VLLM_ENABLE_ROCE_ALLREDUCE = "1"
|
|
|
|
|
+# Route mxfp8 linears away from flashinfer cutlass (rejects small GDN GEMMs: min n/k=128)
|
|
|
|
|
+VLLM_DISABLED_KERNELS = "FlashInferCutlassMxfp8LinearKernel"
|
|
|
# FlashInfer JIT compiles at first boot with N parallel nvcc jobs ->
|
|
# FlashInfer JIT compiles at first boot with N parallel nvcc jobs ->
|
|
|
# OOM-killed on GB10 while weights+KV are resident; cap the ninja jobs.
|
|
# OOM-killed on GB10 while weights+KV are resident; cap the ninja jobs.
|
|
|
MAX_JOBS = "2"
|
|
MAX_JOBS = "2"
|
