Method admit_request

examples/server/server.py:12912–12948

(self, request: CompletionRequest)

        return False

    def admit_request(self, request: CompletionRequest) -> None:
        match_seq_id = request.match_sequence_id
        match_length = request.match_length
        claimable = match_seq_id in self.scheduler.free_sequences
        if request.sequence_cache_match is not None:
            base_seq_id = self.scheduler.claim_unused_sequence()
            match_length, request.prompt_logits = self.scheduler.hydrate_sequence_cache_match(
                request,
                base_seq_id,
            )
        elif claimable:
            base_seq_id = self.scheduler.claim_free_sequence(match_seq_id)
            request.prompt_logits = self.scheduler.checkpoint_logits.get(base_seq_id)
            self.scheduler.metrics.checkpoint_hits_total += 1
        else:
            base_seq_id = self.scheduler.claim_unused_sequence()
            request.prompt_logits = None
        sibling_count = request.internal_completion_count - 1
        sibling_seq_ids: List[int] = []
        for _ in range(sibling_count):
            seq_id = self.scheduler.claim_unused_sequence()
            sibling_seq_ids.append(seq_id)
        self.scheduler.activate_request(
            request,
            base_seq_id=base_seq_id,
            sibling_seq_ids=sibling_seq_ids,
        )
        request.prompt_cursor = match_length
        if request.prompt_cursor == len(request.prompt_tokens):
            request.prompt_done = True
            self.scheduler.maybe_save_prompt_checkpoint(request)
            self.scheduler.maybe_save_sequence_cache(request)
            self.scheduler.start_completions(
                request,
                prompt_output_index=None,
                prompt_logits=request.prompt_logits,
            )

    def copy_prompt_state(
        self,