Skip to main content

vLLM Engine Invariants

Library version: 0.7.3
Anchor frozen at: 2026-04-27

Summary: 49 proposed invariants, 0 validation-confirmed cases.

Invariants by extraction source​

Static miner (AST analysis) (15)​

  • vllm_compilationconfig_dormant_backend_eq [dormant] - CompilationConfig.post_init: marks dormant when backend ==
  • vllm_eplbconfig_raises_log_balancedness_interval_le_0 [error] - EPLBConfig.validate_eplb_config: raises when log_balancedness present True AND log_balancedness...
  • vllm_loraconfig_dormant_max_cpu_loras_unset_true [dormant] - LoRAConfig._validate_lora_config: marks dormant when max_cpu_loras absent True
  • vllm_parallelconfig_dormant_all2all_backend_eq_pplx [dormant] - ParallelConfig._validate_parallel_config: marks dormant when all2all_backend == pplx
  • vllm_parallelconfig_dormant_data_parallel_rank_set_true [dormant] - ParallelConfig.post_init: marks dormant when distributed_executor_backend == external_launche...
  • vllm_parallelconfig_raises_api_process_rank_ge_ref_api_process_count [error] - ParallelConfig._validate_parallel_config: raises when _api_process_rank >= @_api_process_count
  • vllm_parallelconfig_raises_data_parallel_external_lb_set_true [error] - ParallelConfig._validate_parallel_config: raises when data_parallel_size <= 1 AND data_parallel_e...
  • vllm_samplingparams_dormant_bad_words_unset_true [dormant] - SamplingParams.post_init: marks dormant when bad_words absent True
  • vllm_samplingparams_dormant_seed_eq_neg1 [dormant] - SamplingParams.post_init: marks dormant when seed == -1
  • vllm_samplingparams_dormant_skip_reading_prefix_cache_unset_true [dormant] - SamplingParams.post_init: marks dormant when skip_reading_prefix_cache absent True
  • vllm_samplingparams_raises_min_tokens_gt_ref_max_tokens [error] - SamplingParams._verify_args: raises when max_tokens present True AND min_tokens > @max_tokens
  • vllm_samplingparams_raises_min_tokens_lt_0 [error] - SamplingParams._verify_args: raises when min_tokens < 0
  • vllm_samplingparams_raises_n_lt_1 [error] - SamplingParams._verify_args: raises when n < 1
  • vllm_samplingparams_raises_n_not_type_int [error] - SamplingParams._verify_args: raises when n type_is_not int
  • vllm_samplingparams_raises_top_k_lt_neg1 [error] - SamplingParams._verify_args: raises when top_k < -1

Dynamic miner (constructor probing) (1)​

  • vllm_sampling_token_counts_max_tokens_lt_min_tokens [error] - sampling_token_counts: error when max_tokens < min_tokens

Pydantic field lift (33)​

  • vllm_cacheconfig_cache_dtype_in_7_values [error] - CacheConfig.cache_dtype must be one of ['auto', 'bfloat16', 'fp8', 'fp8_e4m3', 'fp8_e5m2', 'fp8_i...
  • vllm_cacheconfig_cpu_offload_gb_ge_0 [error] - CacheConfig.cpu_offload_gb requires >= 0
  • vllm_cacheconfig_gpu_memory_utilization_gt_0 [error] - CacheConfig.gpu_memory_utilization requires > 0
  • vllm_cacheconfig_gpu_memory_utilization_le_1 [error] - CacheConfig.gpu_memory_utilization requires <= 1
  • vllm_cacheconfig_kv_offloading_backend_in_2_values [error] - CacheConfig.kv_offloading_backend must be one of ['native', 'lmcache']
  • vllm_cacheconfig_mamba_block_size_gt_0 [error] - CacheConfig.mamba_block_size requires > 0
  • vllm_cacheconfig_mamba_cache_dtype_in_3_values [error] - CacheConfig.mamba_cache_dtype must be one of ['auto', 'float32', 'float16']
  • vllm_cacheconfig_mamba_cache_mode_in_3_values [error] - CacheConfig.mamba_cache_mode must be one of ['all', 'align', 'none']
  • vllm_cacheconfig_mamba_ssm_cache_dtype_in_3_values [error] - CacheConfig.mamba_ssm_cache_dtype must be one of ['auto', 'float32', 'float16']
  • vllm_cacheconfig_prefix_caching_hash_algo_in_4_values [error] - CacheConfig.prefix_caching_hash_algo must be one of ['sha256', 'sha256_cbor', 'xxhash', 'xxhash_c...
  • vllm_cacheconfig_swap_space_ge_0 [error] - CacheConfig.swap_space requires >= 0
  • vllm_compilationconfig_compile_cache_save_format_in_2_values [error] - CompilationConfig.compile_cache_save_format must be one of ['binary', 'unpacked']
  • vllm_eplbconfig_num_redundant_experts_ge_0 [error] - EPLBConfig.num_redundant_experts requires >= 0
  • vllm_eplbconfig_policy_in_1_values [error] - EPLBConfig.policy must be one of ['default']
  • vllm_loraconfig_lora_dtype_in_3_values [error] - LoRAConfig.lora_dtype must be one of ['auto', 'float16', 'bfloat16']
  • vllm_loraconfig_max_lora_rank_in_9_values [error] - LoRAConfig.max_lora_rank must be one of [1, 8, 16, 32, 64, 128, 256, 320, 512]
  • vllm_loraconfig_max_loras_ge_1 [error] - LoRAConfig.max_loras requires >= 1
  • vllm_modelconfig_convert_in_4_values [error] - ModelConfig.convert must be one of ['auto', 'none', 'embed', 'classify']
  • vllm_modelconfig_logprobs_mode_in_4_values [error] - ModelConfig.logprobs_mode must be one of ['raw_logits', 'raw_logprobs', 'processed_logits', 'proc...
  • vllm_modelconfig_mm_encoder_tp_mode_in_2_values [error] - ModelConfig.mm_encoder_tp_mode must be one of ['weights', 'data']
  • vllm_modelconfig_mm_processor_cache_type_in_2_values [error] - ModelConfig.mm_processor_cache_type must be one of ['shm', 'lru']
  • vllm_modelconfig_runner_in_4_values [error] - ModelConfig.runner must be one of ['auto', 'generate', 'pooling', 'draft']
  • vllm_multimodalconfig_mm_processor_cache_gb_ge_0 [error] - MultiModalConfig.mm_processor_cache_gb requires >= 0
  • vllm_multimodalconfig_video_pruning_rate_ge_0p0 [error] - MultiModalConfig.video_pruning_rate requires >= 0.0
  • vllm_multimodalconfig_video_pruning_rate_lt_1p0 [error] - MultiModalConfig.video_pruning_rate requires < 1.0
  • vllm_parallelconfig__api_process_count_gt_0 [error] - ParallelConfig._api_process_count requires > 0
  • vllm_parallelconfig__api_process_rank_ge_neg1 [error] - ParallelConfig._api_process_rank requires >= -1
  • vllm_parallelconfig_all2all_backend_in_7_values [error] - ParallelConfig.all2all_backend must be one of ['naive', 'pplx', 'deepep_high_throughput', 'deepep...
  • vllm_parallelconfig_data_parallel_backend_in_2_values [error] - ParallelConfig.data_parallel_backend must be one of ['ray', 'mp']
  • vllm_parallelconfig_expert_placement_strategy_in_2_values [error] - ParallelConfig.expert_placement_strategy must be one of ['linear', 'round_robin']
  • vllm_poolerconfig_seq_pooling_type_in_3_values [error] - PoolerConfig.seq_pooling_type must be one of ['CLS', 'LAST', 'MEAN']
  • vllm_poolerconfig_tok_pooling_type_in_2_values [error] - PoolerConfig.tok_pooling_type must be one of ['ALL', 'STEP']
  • vllm_structuredoutputsconfig_backend_in_5_values [error] - StructuredOutputsConfig.backend must be one of ['auto', 'xgrammar', 'guidance', 'outlines', 'lm-f...