refactor(build): rename docker-images/glm52-int4-int8 to glm52

Shorter, matches the sparkrun-vllm-ds4-gb10:glm52 tag it produces and the recipes/glm52 family. The dir holds the int4-int8 overlay for the glm52 image, not a single recipe; the old name implied a 1:1 recipe mapping that never held. No behavior change — same Dockerfile, same build.sh, same output tag; only the docker_image_dir references (2 active, 1 commented) and the Dockerfile build-path comment move with it.

docker-images/glm52-int4-int8/Dockerfile → docker-images/glm52/Dockerfile

@@ -5,7 +5,7 @@
 # nvidia_entrypoint.sh first, then hands off with exec after the unset.
 #
 # Build:
-#   cd docker-images/glm52-int4-int8 && bash build.sh
+#   cd docker-images/glm52 && bash build.sh
 #   Tags: sparkrun-vllm-ds4-gb10:glm52
 
 FROM aidendle94/sparkrun-vllm-ds4-gb10:production-hybrid-1.3

recipes/glm52-hybrid-fp8-mxfp4/recipe.toml

@@ -33,7 +33,7 @@ tensor_parallel_size = 4
 pipeline_parallel_size = 1
 distributed_backend = "mp"
 shm_size = "10gb"
-docker_image_dir = "glm52-int4-int8"
+docker_image_dir = "glm52"
 
 entrypoint = ["vllm", "serve"]
 

recipes/glm52-int4-int8/recipe.toml

@@ -25,7 +25,7 @@ tensor_parallel_size = 4
 pipeline_parallel_size = 1
 distributed_backend = "mp"
 shm_size = "64gb"
-# docker_image_dir = "glm52-int4-int8"             # removed for experiment
+# docker_image_dir = "glm52"             # removed for experiment
 
 entrypoint = ["vllm", "serve"]
 

recipes/glm52-mxfp4-experts-gptq/recipe.toml

@@ -30,7 +30,7 @@ tensor_parallel_size = 4
 pipeline_parallel_size = 1
 distributed_backend = "mp"
 shm_size = "10gb"
-docker_image_dir = "glm52-int4-int8"
+docker_image_dir = "glm52"
 
 entrypoint = ["vllm", "serve"]