aiden lanes: drop manual HF mounts (launcher auto-mounts ~/.cache/huggingface = shared NAS on all 4 nodes); GLM 4x -> NVFP4 (in shared cache, +23% vs FP8)

recipes/dsv4-vision-aiden-2x/recipe.toml

@@ -88,7 +88,3 @@ HF_HUB_OFFLINE = "1"
 TRANSFORMERS_OFFLINE = "1"
 VLLM_NO_USAGE_STATS = "1"
 
-[[extra_mounts]]
-host = "/mnt/nas/ai/huggingface"
-container = "/opt/local-hf"
-env = "HF_HOME"

recipes/dsv4-vision-aiden-4x/recipe.toml

@@ -101,7 +101,3 @@ HF_HUB_OFFLINE = "1"
 TRANSFORMERS_OFFLINE = "1"
 VLLM_NO_USAGE_STATS = "1"
 
-[[extra_mounts]]
-host = "/mnt/nas/ai/huggingface"
-container = "/opt/local-hf"
-env = "HF_HOME"

recipes/glm53-flash-aiden-4x/recipe.toml

@@ -25,7 +25,7 @@
 engine = "vllm"
 name = "GLM-5.3-Flash aiden image (vLLM, 4x Spark, DFlash2-7, FP8) [DEFERRED]"
 docker_image = "aidendle94/sparkrun-vllm-glm53-gb10:production-1.1"
-model = "unsloth/GLM-5.3-Flash-FP8"
+model = "LibertAIDAI/GLM-5.3-Flash-NVFP4"
 required_nodes = 4
 tensor_parallel_size = 4
 pipeline_parallel_size = 1