Makefile

# ---------- Config ----------
NS        ?= ai
GATEWAY   ?= http://gateway.example.internal
ADMIN_TOKEN ?= changeme
# switch to kubectl if you don't have 'kc' alias
KUBECTL   ?= kubectl

# ---- simple short aliases ----
# canonical strings (already defined)
GLM1P5B       := local:zai-org/glm-edge-1.5b-chat
GLM4B         := local:zai-org/glm-edge-4b-chat
QWEN7B        := local:Qwen/Qwen2.5-Coder-7B-Instruct
QWEN7B_TGI    := local:Qwen/Qwen2.5-Coder-7B-Instruct@tgi
QWEN14B       := local:Qwen/Qwen2.5-Coder-14B-Instruct
QWEN14B_TGI   := local:Qwen/Qwen2.5-Coder-14B-Instruct@tgi
SC2_15B       := local:bigcode/starcoder2-15b-instruct
SC2_15B_TGI   := local:bigcode/starcoder2-15b-instruct@tgi
DSC2_16B      := local:deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct
DSC2_16B_TGI  := local:deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct@tgi

# pick a short alias and we’ll map it to MODEL
# accepted values: glm1p5b, glm4b, q7b, q7b@tgi, q14b, q14b@tgi, sc2, sc2@tgi, dsc2, dsc2@tgi
MODEL_SHORT ?= glm1p5b

# map MODEL_SHORT -> MODEL
ifeq ($(MODEL_SHORT),glm1p5b)
  MODEL := $(GLM1P5B)
endif
ifeq ($(MODEL_SHORT),glm4b)
  MODEL := $(GLM4B)
endif
ifeq ($(MODEL_SHORT),q7b)
  MODEL := $(QWEN7B)
endif
ifeq ($(MODEL_SHORT),q7b@tgi)
  MODEL := $(QWEN7B_TGI)
endif
ifeq ($(MODEL_SHORT),q14b)
  MODEL := $(QWEN14B)
endif
ifeq ($(MODEL_SHORT),q14b@tgi)
  MODEL := $(QWEN14B_TGI)
endif
ifeq ($(MODEL_SHORT),sc2)
  MODEL := $(SC2_15B)
endif
ifeq ($(MODEL_SHORT),sc2@tgi)
  MODEL := $(SC2_15B_TGI)
endif
ifeq ($(MODEL_SHORT),dsc2)
  MODEL := $(DSC2_16B)
endif
ifeq ($(MODEL_SHORT),dsc2@tgi)
  MODEL := $(DSC2_16B_TGI)
endif

# convenience wrappers: hotload-%, unload-%, chat-% using the short alias
.PHONY: hotload-% unload-% chat-%
hotload-%:
	@$(MAKE) --no-print-directory hotload MODEL_SHORT=$*

unload-%:
	@$(MAKE) --no-print-directory unload MODEL_SHORT=$*

# PROMPT still honored; usage: make chat-q7b PROMPT='…'
chat-%:
	@$(MAKE) --no-print-directory chat MODEL_SHORT=$* PROMPT="$(PROMPT)"

health:
	@curl -s $(GATEWAY)/healthz | jq .

models:
	@curl -s $(GATEWAY)/v1/models | jq .

hotload:
	@test -n "$(MODEL)" || (echo "MODEL is required"; exit 1)
	@curl -sS --fail -X POST $(GATEWAY)/admin/hotload \
	  -H "X-Admin-Token: $(ADMIN_TOKEN)" -H 'Content-Type: application/json' \
	  -d '{"model":"$(MODEL)"}' | jq .

unload:
	@test -n "$(MODEL)" || (echo "MODEL is required"; exit 1)
	@curl -sS --fail -X POST $(GATEWAY)/admin/unload \
	  -H "X-Admin-Token: $(ADMIN_TOKEN)" -H 'Content-Type: application/json' \
	  -d '{"model":"$(MODEL)"}' | jq .

purge:
	@test -n "$(MODEL)" || (echo "MODEL is required"; exit 1)
	@curl -sS --fail -X POST $(GATEWAY)/admin/purge-cache \
	  -H "X-Admin-Token: $(ADMIN_TOKEN)" -H 'Content-Type: application/json' \
	  -d '{"model":"$(MODEL)"}' | jq .

# Chat with a one-line prompt. Ex: make chat MODEL='$(QWEN14B_TGI)' PROMPT='Implement quicksort in Rust'
PROMPT ?= Say hello from $(MODEL).
chat:
	@test -n "$(MODEL)" || (echo "MODEL is required"; exit 1)
	@jq -n --arg m "$(MODEL)" --arg p "$(PROMPT)" \
	  '{model:$$m, messages:[{role:"user", content:$$p}], temperature:0.2, max_tokens:256}' \
	| curl -sS --fail $(GATEWAY)/v1/chat/completions -H 'Content-Type: application/json' -d @- | jq .

# ---------- K8s quick look ----------
.PHONY: ns-info ns-events pods gw-logs vllm-logs tgi-logs svc-eps

ns-info:
	@$(KUBECTL) get ns $(NS) || true
	@echo
	@$(KUBECTL) -n $(NS) get deploy,rs,po,svc | sed 's/^/  /'

ns-events:
	@$(KUBECTL) -n $(NS) get events --sort-by=.lastTimestamp | tail -n 100

pods:
	@$(KUBECTL) -n $(NS) get pods -o wide

svc-eps:
	@$(KUBECTL) -n $(NS) get svc,endpoints

# Gateway logs (all pods)
gw-logs:
	@$(KUBECTL) -n $(NS) logs -l app.kubernetes.io/name=gateway --all-containers=true --tail=200

# vLLM logs (all vLLM pods)
vllm-logs:
	@$(KUBECTL) -n $(NS) logs -l ai.zlabs.dev/component=vllm --all-containers=true --tail=200

# TGI logs (all TGI pods)
tgi-logs:
	@$(KUBECTL) -n $(NS) logs -l ai.zlabs.dev/component=tgi --all-containers=true --tail=200

# Follow logs (Ctrl+C to stop)
follow-gw:
	@$(KUBECTL) -n $(NS) logs -l app.kubernetes.io/name=gateway --all-containers=true -f

follow-vllm:
	@$(KUBECTL) -n $(NS) logs -l ai.zlabs.dev/component=vllm --all-containers=true -f

follow-tgi:
	@$(KUBECTL) -n $(NS) logs -l ai.zlabs.dev/component=tgi --all-containers=true -f