feat: Implement changes in deps and refactor llama config

Author: jhwohlgemuth

Dockerfile.agentic

@@ -37,7 +37,7 @@ ENV MODEL_DIRECTORY="${MODEL_DIRECTORY}"
 # %files
 #
 COPY --chmod=0755 --chown=${USER_NAME}:${USER_NAME} ./config/llama.cpp/service/* /etc/services.d/llama.cpp/
-COPY --chmod=0755 --chown=${USER_NAME}:${USER_NAME} ./config/llama.cpp/llama-swap.yaml "${HOME}/.config/llama-swap/config.yaml"
+COPY --chmod=0755 --chown=${USER_NAME}:${USER_NAME} ./config/llama-swap/config.yaml "${HOME}/.config/llama-swap/config.yaml"
 COPY --chmod=0755 --chown=${USER_NAME}:${USER_NAME} ./provision/agentic/* /tmp/scripts/
 COPY --chmod=0755 --chown=${USER_NAME}:${USER_NAME} ./provision/healthcheck /healthcheck
 #

Dockerfile.terminal

@@ -32,6 +32,7 @@ ENV VERSION="${VERSION}"
 ENV TZ=America/New_York
 ENV SHELL=/usr/bin/zsh
 ENV TERM=xterm-256color
+ENV PATH="/home/linuxbrew/.linuxbrew/bin:/home/linuxbrew/.linuxbrew/sbin:${PATH}"
 ENV PATH="${PATH}:${HOME}/.local/bin"
 ENV PATH="${PATH}:${HOME}/.nix-profile/bin"
 ENV PATH="${PATH}:${HOME}/.pixi/bin"
@@ -68,7 +69,6 @@ RUN curl -fsSL https://shell.jasonwohlgemuth.com/install.sh | bash \
         && brew trust oven-sh/bun \
         && brew bundle --file /tmp/scripts/Brewfile \
         && brew cleanup --prune=all \
-        && eval "$(pkgx integrate)" \
     && install_nix \
     && install_ohmyzsh && /tmp/scripts/configure_ohmyzsh.sh \
     && mkdir -p "${HOME}/.ssh" && chmod 700 "${HOME}/.ssh"
@@ -91,4 +91,4 @@ HEALTHCHECK NONE
 USER "${USER_NAME}"
 WORKDIR "${HOME}"
 ENTRYPOINT [ "/init" ]
-CMD ["/bin/zsh"]
+CMD ["/bin/zsh"]

Makefile

@@ -36,8 +36,8 @@ lint:
 		fi; \
     done
 download:
-	@pwsh -NoProfile -File ./scripts/Get-Models.ps1 -Model all -Token "${HF_API_TOKEN}"
-# 	@yamllint .
+	@pwsh -NoProfile -File ./scripts/Get-Models.ps1 -Model "$(MODELS)" -Token "${HF_API_TOKEN}"
+
 #
 # Build tasks
 #
@@ -119,6 +119,7 @@ TARGETS = \
 	changelog \
 	check \
 	convert \
+	download \
 	lint \
 	update-mamba-version \
 	gold-image \
@@ -129,3 +130,6 @@ TARGETS = \
 	push-notebook \
 	build-image \
 	push-image
+MODELS = \
+	openai/gpt-oss-20b \
+# 	openai/gpt-oss-120b

WARP.md

@@ -8,30 +8,15 @@ macros:
   "ssl_key": "${env.HOME}/certs/my.key"
   "ssl_cert": "${env.HOME}/certs/my.pem"
 models:
-  "smollm2":
+  "gpt-oss-20b":
     proxy: "http://127.0.0.1:${PORT}"
-    description: "When should the model be used?"
-    cmd: |
-      llama-server
-      --offline
-      --batch-size 2048
-      --host 0.0.0.0
-      --jinja
-      --model ${models_dir}/SmolLM2-135M-Instruct-Q4_K_M.gguf
-      --port ${PORT}
-      --sleep-idle-seconds 600
-      --tools all
-      --ubatch-size 2048
-  "qwen2.5":
-    proxy: "http://127.0.0.1:${PORT}"
-    description: "When should the model be used?"
     cmd: |
       llama-server
       --offline
       --batch-size 2048
       --host 0.0.0.0
       --jinja
-      --model ${models_dir}/Qwen2.5-0.5B-Instruct-Q4_K_M.gguf
+      --model ${models_dir}/gpt-oss-20b
       --port ${PORT}
       --sleep-idle-seconds 600
       --tools all
@@ -53,4 +38,4 @@ models:
       --serving-name vllm-coder
     cmdStop: podman stop ${MODEL_ID}
     checkEndpoint: "none"
-    ttl: 0
+    ttl: 0

config/llama.cpp/llama-swap.yaml config/llama-swap/config.yamlconfig/llama.cpp/llama-swap.yaml renamed to config/llama-swap/config.yaml

@@ -0,0 +1,29 @@
+healthCheckTimeout: 500
+proxy:
+  listen: "http://0.0.0.0:10732"
+macros:
+  "context_size": "${env.LLAMA_ARG_CTX_SIZE}"
+  "parallel": "${env.LLAMA_ARG_N_PARALLEL}"
+  "models_dir": "${env.HOME}/models"
+  "ssl_key": "${env.HOME}/certs/my.key"
+  "ssl_cert": "${env.HOME}/certs/my.pem"
+models:
+{{ generated_models }}
+  "vllm-coder":
+    proxy: "http://127.0.0.1:${PORT}"
+    description: "When should the model be used?"
+    cmd: |
+      podman run
+      --name ${MODEL_ID}
+      --init
+      --rm
+      --gpus all
+      --publish ${PORT}:8000
+      -v "${env.HOME}/models:/root/models"
+      docker.io/vllm/vllm-openai:latest
+      --model ${models_dir}/Qwen2.5-0.5B-Instruct-Q4_K_M.gguf
+      --port 8000
+      --serving-name vllm-coder
+    cmdStop: podman stop ${MODEL_ID}
+    checkEndpoint: "none"
+    ttl: 0