Files
MOVZXandHinata 61fbe0b865 [Mini-VLLM] Remove unreferenced definitions found by static scan
A symbol scan over every tracked file found 123 top-level definitions in
vllm/ with zero references, including references inside strings. All of
them came from features this fork deleted earlier.

- Delete dead attention helpers: MLADCPManager, split_prefill_chunks, the
  reshape helpers, compute_varlen_chunk_metadata, the LINEAR backend's
  unused BailingLinearAttentionBackend, and the sparse flash-attn entry
  points.
- Delete the CPU and AMX op wrappers from _custom_ops.py: cpu_mla_decode,
  cpu_mla_extend, cpu_mla_get_scheduler_metadata,
  gather_mla_context_cache_cpu.
- Delete dead quantization helpers: awq_pack, gptq_pack,
  gptq_quantize_weights, get_and_maybe_dequant_weights,
  marlin_quant_fp8_torch, the rand_marlin_weight_*_like pair,
  ocp_mx_weight_dtype_and_rows, get_fp8_block_weight_scale,
  input_to_float8.
- Delete dead model helpers: resolve_visual_encoder_outputs,
  run_dp_sharded_vision_model, fast_topk, get_layer_index,
  is_pp_missing_parameter, skip_spec_layers, KimiK3MergedQKVGateLinear,
  selective_scan_fn, causal_conv1d_update_torch.
- Delete dead utils: format_kib, format_mib, chunk_list, prev_power_of_2,
  split_host_port, join_host_port, make_zmq_path, make_async_with_semaphore,
  is_meta_module, is_non_overlapping_and_dense, supports_xpu_graph,
  create_kv_caches_with_random*, and the unused has_* capability probes.
- Delete the unreachable deep-gemm mega-MoE warmup path, which imported
  vllm.models.deepseek_v4.nvidia.model and vllm.models.kimi_k3.nvidia.model,
  neither of which exists anymore.
- Drop the imports and the two now-empty guards left behind.

Verified statically: every changed file compiles, no unused import
remains, and no deleted name is referenced in vllm, tests or csrc. The
runtime gates still need the Python environment rebuilt.

Co-authored-by: Hinata <hinata@idihore.id>
Signed-off-by: MOVZX <movzx@yahoo.com>
2026-10-19 21:39:56 +07:00

105 lines
3.5 KiB
Python

# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
"""EC connector helper utilities."""
from typing import TYPE_CHECKING
from vllm.logger import init_logger
from vllm.v1.outputs import ECConnectorOutput, ModelRunnerOutput
if TYPE_CHECKING:
from vllm.config import ModelConfig
logger = init_logger(__name__)
class PlaceholderMetadataResolver:
"""Resolves which processed keys a model needs published per modality.
Reads `MultiModalDataParser.embedding_fields`, the same declaration the
consumer's parser requires, so the two cannot drift. An empty set means
the modality cannot be delivered out of band, and the consumer will
process the media itself.
"""
def __init__(self, model_config: "ModelConfig") -> None:
self._model_config = model_config
self._cache: dict[str, set[str]] = {}
def fields_for(self, modality: str) -> set[str]:
if modality in self._cache:
return self._cache[modality]
fields: set[str] = set()
try:
from vllm.multimodal import MULTIMODAL_REGISTRY
info = MULTIMODAL_REGISTRY.create_processor(self._model_config).info
fields = info.data_parser.placeholder_metadata_fields(modality)
except Exception:
logger.warning(
"Could not determine the placeholder metadata fields for "
"modality %s; the consumer will preprocess the media itself.",
modality,
exc_info=True,
)
self._cache[modality] = fields
return fields
class ECOutputAggregator:
"""Merge every worker's EC connector output onto the single
ModelRunnerOutput that reaches the scheduler.
Mirrors KVOutputAggregator: only `output_rank`'s output is returned to the
scheduler, but the EC connector may have run on any rank.
"""
def aggregate(
self, outputs: list[ModelRunnerOutput | None], output_rank: int = 0
) -> ModelRunnerOutput | None:
output = outputs[output_rank]
if not output:
return None
finished_sending = set[str]()
finished_recving = set[str]()
worker_meta = None
connector_stats = None
for model_runner_output in outputs:
assert model_runner_output is not None
ec_output = model_runner_output.ec_connector_output
if not ec_output:
continue
finished_sending |= ec_output.finished_sending or set()
finished_recving |= ec_output.finished_recving or set()
if meta := ec_output.ec_connector_worker_meta:
worker_meta = (
meta if worker_meta is None else worker_meta.aggregate(meta)
)
if stats := ec_output.ec_connector_stats:
connector_stats = (
stats
if connector_stats is None
else connector_stats.aggregate(stats)
)
aggregated = ECConnectorOutput(
finished_sending=finished_sending or None,
finished_recving=finished_recving or None,
ec_connector_stats=connector_stats,
ec_connector_worker_meta=worker_meta,
)
if aggregated.is_empty():
output.ec_connector_output = None
return output
# `output` is the shared empty output whenever `output_rank` had no work,
# so attach through the copy-on-write helper.
return ModelRunnerOutput.with_ec_conn_output(output, aggregated)