(max_model_len)
| 82 | |
| 83 | |
| 84 | def _create_fd_config(max_model_len): |
| 85 | model_config: Mock = Mock() |
| 86 | model_config.max_model_len = max_model_len |
| 87 | model_config.architectures = ["test_model"] |
| 88 | model_config.mm_max_tokens_per_item = None |
| 89 | speculative_config = SpeculativeConfig({}) |
| 90 | graph_opt_config = GraphOptimizationConfig({}) |
| 91 | scheduler_config = SchedulerConfig({}) |
| 92 | parallel_config = ParallelConfig({}) |
| 93 | cache_config = CacheConfig({}) |
| 94 | cache_config.cache_transfer_protocol = "rdma,ipc" |
| 95 | cache_config.pd_comm_port = "2334" |
| 96 | fd_config = FDConfig( |
| 97 | model_config=model_config, |
| 98 | speculative_config=speculative_config, |
| 99 | graph_opt_config=graph_opt_config, |
| 100 | scheduler_config=scheduler_config, |
| 101 | parallel_config=parallel_config, |
| 102 | cache_config=cache_config, |
| 103 | ) |
| 104 | |
| 105 | return fd_config |
| 106 | |
| 107 | |
| 108 | def _create_share_inputs(max_num_seqs, max_draft_token_num, max_model_len, vocab_size): |
no test coverage detected