[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:
co-authored by
OpenAI Codex
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
8c34a37232
commit
ea40bb9e90
@@ -307,7 +307,26 @@ class Sampler:
|
||||
or self.sampling_states.any_explicit_seed(idx_mapping_np)
|
||||
)
|
||||
|
||||
# Sample the next token.
|
||||
return self._sample_random(
|
||||
processed_logits,
|
||||
expanded_idx_mapping,
|
||||
idx_mapping_np,
|
||||
pos,
|
||||
top_k,
|
||||
top_p,
|
||||
use_flashinfer,
|
||||
)
|
||||
|
||||
def _sample_random(
|
||||
self,
|
||||
processed_logits: torch.Tensor,
|
||||
expanded_idx_mapping: torch.Tensor,
|
||||
idx_mapping_np: np.ndarray,
|
||||
pos: torch.Tensor,
|
||||
top_k: torch.Tensor | None,
|
||||
top_p: torch.Tensor | None,
|
||||
use_flashinfer: bool,
|
||||
) -> tuple[torch.Tensor, torch.Tensor]:
|
||||
if use_flashinfer:
|
||||
sampled = flashinfer_sample(processed_logits, top_k, top_p).to(torch.int64)
|
||||
else:
|
||||
|
||||
Reference in New Issue
Block a user