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>
88 lines
3.1 KiB
Python
88 lines
3.1 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
"""Skip Triton autotuning under VLLM_TRITON_FORCE_FIRST_CONFIG."""
|
|
|
|
from vllm.logger import init_logger
|
|
from vllm.triton_utils.importing import HAS_TRITON
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
_installed: bool = False
|
|
|
|
|
|
def install() -> None:
|
|
"""Install the Autotuner.run replacement."""
|
|
global _installed
|
|
if _installed:
|
|
return
|
|
if not HAS_TRITON:
|
|
return
|
|
|
|
import importlib
|
|
|
|
autotuner_mod = importlib.import_module("triton.runtime.autotuner")
|
|
Autotuner = autotuner_mod.Autotuner
|
|
from triton.compiler.errors import CompileTimeAssertionFailure
|
|
from triton.runtime.errors import OutOfResources, PTXASError
|
|
|
|
_invalid_config_errors = (OutOfResources, CompileTimeAssertionFailure, PTXASError)
|
|
_picked_cache: dict[tuple, int] = {}
|
|
seen_kernels: set[str] = set()
|
|
|
|
def _run_first_valid_config(self, *args, **kwargs):
|
|
if not self.configs:
|
|
return self.fn(*args, **kwargs)
|
|
|
|
key_vals = tuple(kwargs[name] for name in self.keys if name in kwargs)
|
|
cache_key = (id(self), key_vals)
|
|
kernel_name = getattr(self.base_fn, "__name__", repr(self.fn))
|
|
|
|
cached_idx = _picked_cache.get(cache_key)
|
|
candidate_indices = (
|
|
[cached_idx] if cached_idx is not None else list(range(len(self.configs)))
|
|
)
|
|
|
|
last_exc: Exception | None = None
|
|
for idx in candidate_indices:
|
|
config = self.configs[idx]
|
|
if config.pre_hook is not None:
|
|
full_nargs = {
|
|
**dict(zip(self.arg_names, args)),
|
|
**kwargs,
|
|
**config.all_kwargs(),
|
|
}
|
|
config.pre_hook(full_nargs)
|
|
# Prefer self.fn.run(...) — the kernel-launch entrypoint for both
|
|
# JITFunction and Heuristics. Calling JITFunction(...) directly
|
|
# raises "Cannot call @triton.jit'd outside of the scope of a
|
|
# kernel". Fall back to plain call only if .run is missing.
|
|
launch = getattr(self.fn, "run", self.fn)
|
|
try:
|
|
result = launch(*args, **kwargs, **config.all_kwargs())
|
|
except _invalid_config_errors as e:
|
|
last_exc = e
|
|
continue
|
|
|
|
if cached_idx is None:
|
|
_picked_cache[cache_key] = idx
|
|
self.best_config = config
|
|
if kernel_name not in seen_kernels:
|
|
seen_kernels.add(kernel_name)
|
|
logger.info(
|
|
"[triton-autotune-disabled] kernel=%s configs=%d "
|
|
"picked_index=%d picked=%s",
|
|
kernel_name,
|
|
len(self.configs),
|
|
idx,
|
|
config,
|
|
)
|
|
return result
|
|
|
|
raise RuntimeError(
|
|
f"No valid config for kernel "
|
|
f"{kernel_name} key={key_vals} (tried {len(self.configs)} configs)"
|
|
) from last_exc
|
|
|
|
Autotuner.run = _run_first_valid_config
|
|
_installed = True
|