maxtext.integration.vllm.maxtext_vllm_adapter package

MaxText vLLM adapter package.

maxtext.integration.vllm.maxtext_vllm_adapter.register()

Register MaxTextForCausalLM model with tpu_inference and vllm.

Note, this function is invoked directly by the vLLM engine during startup. As such, it leverages vLLM logging to report its status.

Submodules

• maxtext.integration.vllm.maxtext_vllm_adapter.adapter module
  • next_power_of_two()
  • generate_maxtext_config()
  • MaxTextForCausalLM
    • MaxTextForCausalLM.forward()
    • MaxTextForCausalLM.get_input_embeddings()
    • MaxTextForCausalLM.embed_input_ids()
    • MaxTextForCausalLM.compute_logits()
    • MaxTextForCausalLM.load_weights()