HF_TOKEN=hf_your_token_here

VLLM_IMAGE=vllm/vllm-openai:gemma4-0505-arm64-cu130
APPLY_VLLM_PR_PATCH=1
VLLM_PR_NUMBER=41745
VLLM_PATCH_REPO=https://github.com/vllm-project/vllm.git

PATCH_TRANSFORMERS_FOR_MTP=1
HF_HUB_VERSION=1.13.0
REGEX_VERSION=2025.10.23
TOKENIZERS_VERSION=0.22.1
TRANSFORMERS_SOURCE=git

HOST=0.0.0.0
MAX_MODEL_LEN=8192
MAX_NUM_BATCHED_TOKENS=8192
MAX_NUM_SEQS=1
GPU_MEMORY_UTILIZATION=0.82
KV_CACHE_DTYPE=fp8
ATTENTION_BACKEND=auto
LOAD_FORMAT=safetensors
LANGUAGE_MODEL_ONLY=1
REASONING_PARSER=gemma4
ENABLE_AUTO_TOOL_CHOICE=1
TOOL_CALL_PARSER=gemma4
PRESERVE_THINKING=0

OMP_NUM_THREADS=16
VLLM_CPU_OMP_THREADS_BIND=1
TORCHINDUCTOR_MAX_AUTOTUNE=0
TRITON_MAX_AUTOTUNE=0

GEMMA26_MODEL=google/gemma-4-26B-A4B-it
GEMMA31_MODEL=google/gemma-4-31B-it
GEMMA26_ASSISTANT_MODEL=google/gemma-4-26B-A4B-it-assistant
GEMMA31_ASSISTANT_MODEL=google/gemma-4-31B-it-assistant

GEMMA26_PORT=8100
GEMMA26_MODEL_NAME=gemma4-26b-a4b
GEMMA26_MTP_PORT=8101
GEMMA26_MTP_MODEL_NAME=gemma4-26b-a4b-mtp
GEMMA31_PORT=8102
GEMMA31_MODEL_NAME=gemma4-31b
GEMMA31_MTP_PORT=8103
GEMMA31_MTP_MODEL_NAME=gemma4-31b-mtp

GEMMA26_MTP_NUM_SPECULATIVE_TOKENS=4
GEMMA31_MTP_NUM_SPECULATIVE_TOKENS=4

GEMMA31_MTP_MAX_MODEL_LEN=8192
GEMMA31_MTP_MAX_NUM_BATCHED_TOKENS=12288
GEMMA31_MTP_MAX_NUM_SEQS=1
GEMMA31_MTP_GPU_MEMORY_UTILIZATION=0.72
GEMMA31_MTP_LOAD_FORMAT=safetensors
GEMMA31_MTP_ATTENTION_BACKEND=auto
GEMMA31_MTP_MEM_LIMIT=96g
GEMMA31_MTP_MEMSWAP_LIMIT=96g
GEMMA31_MTP_RESTART=no

VLLM_EXTRA_ARGS=
GEMMA26_EXTRA_ARGS=
GEMMA26_MTP_EXTRA_ARGS=
GEMMA31_EXTRA_ARGS=
GEMMA31_MTP_EXTRA_ARGS=