[feature] Watermarked generation and detection (Gumbel-max algorithm) (#54053)

Signed-off-by: Raphael Rialland <raphael.rialland@mistral.ai>
Co-authored-by: OpenAI Codex <codex@openai.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Raphaël Rialland
2026-09-09 21:00:15 -07:00
committed by GitHub
co-authored by OpenAI Codex mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 8c34a37232
commit ea40bb9e90
32 changed files with 1693 additions and 15 deletions
+20 -12
View File
@@ -78,6 +78,8 @@ from vllm.v1.outputs import (
ModelRunnerOutput,
RoutedExpertsTensors,
)
from vllm.v1.watermarking import create_watermarker
from vllm.v1.watermarking.gpu_sampler import GPUWatermarkSampler
from vllm.v1.worker.block_table import get_block_table_width
from vllm.v1.worker.cp_utils import check_attention_cp_compatibility
from vllm.v1.worker.gpu import pcp_manager as pcp
@@ -452,21 +454,27 @@ class GPUModelRunner(LoRAModelRunnerMixin):
# Initialize samplers. Model states may override via custom_sampler().
if self.is_last_pp_rank and not self.is_pooling_model:
self.sampler = Sampler(
max_num_reqs=self.max_num_reqs,
vocab_size=self.vocab_size,
device=self.device,
req_states=self.req_states,
logprobs_mode=self.model_config.logprobs_mode,
num_speculative_tokens=self.decode_query_len,
use_fp64_gumbel=self.model_config.use_fp64_gumbel,
enable_trace_replay=self.model_config.enable_trace_replay,
reasoning_config=self.vllm_config.reasoning_config,
return_sampling_mask=self.model_config.return_sampling_mask,
)
sampler_kwargs: dict[str, Any] = {
"max_num_reqs": self.max_num_reqs,
"vocab_size": self.vocab_size,
"device": self.device,
"req_states": self.req_states,
"logprobs_mode": self.model_config.logprobs_mode,
"num_speculative_tokens": self.decode_query_len,
"use_fp64_gumbel": self.model_config.use_fp64_gumbel,
"enable_trace_replay": self.model_config.enable_trace_replay,
"reasoning_config": self.vllm_config.reasoning_config,
"return_sampling_mask": self.model_config.return_sampling_mask,
}
if self.vllm_config.watermark_config is None:
self.sampler = Sampler(**sampler_kwargs)
else:
watermarker = create_watermarker(self.vllm_config.watermark_config)
self.sampler = GPUWatermarkSampler(watermarker, **sampler_kwargs)
custom = self.model_state.custom_sampler(self.sampler)
if custom:
self.vllm_config._check_watermarking_unsupported(custom_sampler=True)
self.sampler, self.rejection_sampler = custom
elif self.speculative_config is not None:
self.rejection_sampler = RejectionSampler(