feat(server): add logits watermark (#90)

This commit is contained in:
OlivierDehaene 2023-03-02 12:30:41 +01:00 committed by GitHub
parent f874c47831
commit 9b8ea6a6c7
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
11 changed files with 141 additions and 7 deletions

View File

@ -55,6 +55,10 @@ struct Args {
otlp_endpoint: Option<String>, otlp_endpoint: Option<String>,
#[clap(long, env)] #[clap(long, env)]
cors_allow_origin: Vec<String>, cors_allow_origin: Vec<String>,
#[clap(long, env)]
watermark_gamma: Option<f32>,
#[clap(long, env)]
watermark_delta: Option<f32>,
} }
fn main() -> ExitCode { fn main() -> ExitCode {
@ -88,6 +92,8 @@ fn main() -> ExitCode {
json_output, json_output,
otlp_endpoint, otlp_endpoint,
cors_allow_origin, cors_allow_origin,
watermark_gamma,
watermark_delta,
} = args; } = args;
// Signal handler // Signal handler
@ -243,6 +249,8 @@ fn main() -> ExitCode {
huggingface_hub_cache, huggingface_hub_cache,
weights_cache_override, weights_cache_override,
disable_custom_kernels, disable_custom_kernels,
watermark_gamma,
watermark_delta,
otlp_endpoint, otlp_endpoint,
status_sender, status_sender,
shutdown, shutdown,
@ -414,6 +422,8 @@ fn shard_manager(
huggingface_hub_cache: Option<String>, huggingface_hub_cache: Option<String>,
weights_cache_override: Option<String>, weights_cache_override: Option<String>,
disable_custom_kernels: bool, disable_custom_kernels: bool,
watermark_gamma: Option<f32>,
watermark_delta: Option<f32>,
otlp_endpoint: Option<String>, otlp_endpoint: Option<String>,
status_sender: mpsc::Sender<ShardStatus>, status_sender: mpsc::Sender<ShardStatus>,
shutdown: Arc<Mutex<bool>>, shutdown: Arc<Mutex<bool>>,
@ -494,6 +504,16 @@ fn shard_manager(
env.push(("DISABLE_CUSTOM_KERNELS".into(), "True".into())) env.push(("DISABLE_CUSTOM_KERNELS".into(), "True".into()))
} }
// Watermark Gamma
if let Some(watermark_gamma) = watermark_gamma {
env.push(("WATERMARK_GAMMA".into(), watermark_gamma.to_string().into()))
}
// Watermark Delta
if let Some(watermark_delta) = watermark_delta {
env.push(("WATERMARK_DELTA".into(), watermark_delta.to_string().into()))
}
// Start process // Start process
tracing::info!("Starting shard {rank}"); tracing::info!("Starting shard {rank}");
let mut p = match Popen::create( let mut p = match Popen::create(

View File

@ -40,6 +40,8 @@ message NextTokenChooserParameters {
uint64 seed = 5; uint64 seed = 5;
/// repetition penalty /// repetition penalty
float repetition_penalty = 6; float repetition_penalty = 6;
/// token watermarking using "A Watermark for Large Language Models"
bool watermark = 7;
} }
message StoppingCriteriaParameters { message StoppingCriteriaParameters {

View File

@ -53,6 +53,9 @@ pub(crate) struct GenerateParameters {
#[schema(inline, max_items = 4, example = json ! (["photographer"]))] #[schema(inline, max_items = 4, example = json ! (["photographer"]))]
pub stop: Vec<String>, pub stop: Vec<String>,
#[serde(default)] #[serde(default)]
#[schema(default = "false", example = true)]
pub watermark: bool,
#[serde(default)]
#[schema(default = "true")] #[schema(default = "true")]
pub details: bool, pub details: bool,
#[serde(default)] #[serde(default)]
@ -72,7 +75,8 @@ fn default_parameters() -> GenerateParameters {
do_sample: false, do_sample: false,
max_new_tokens: default_max_new_tokens(), max_new_tokens: default_max_new_tokens(),
return_full_text: None, return_full_text: None,
stop: vec![], stop: Vec::new(),
watermark: false,
details: false, details: false,
seed: None, seed: None,
} }

View File

@ -234,6 +234,7 @@ mod tests {
do_sample: false, do_sample: false,
seed: 0, seed: 0,
repetition_penalty: 0.0, repetition_penalty: 0.0,
watermark: false
}, },
stopping_parameters: StoppingCriteriaParameters { stopping_parameters: StoppingCriteriaParameters {
max_new_tokens: 0, max_new_tokens: 0,

View File

@ -72,6 +72,7 @@ async fn health(infer: Extension<Infer>) -> Result<(), (StatusCode, Json<ErrorRe
max_new_tokens: 1, max_new_tokens: 1,
return_full_text: None, return_full_text: None,
stop: Vec::new(), stop: Vec::new(),
watermark: false,
details: false, details: false,
seed: None, seed: None,
}, },

View File

@ -157,6 +157,7 @@ fn validate(
max_new_tokens, max_new_tokens,
stop: stop_sequences, stop: stop_sequences,
seed, seed,
watermark,
.. ..
} = request.parameters; } = request.parameters;
@ -232,6 +233,7 @@ fn validate(
top_p, top_p,
do_sample, do_sample,
seed, seed,
watermark,
}; };
let stopping_parameters = StoppingCriteriaParameters { let stopping_parameters = StoppingCriteriaParameters {
max_new_tokens, max_new_tokens,

View File

@ -67,7 +67,9 @@ class CausalLMBatch(Batch):
for r in pb.requests: for r in pb.requests:
inputs.append(r.inputs) inputs.append(r.inputs)
input_lengths.append(r.input_length) input_lengths.append(r.input_length)
next_token_choosers.append(NextTokenChooser.from_pb(r.parameters, device)) next_token_choosers.append(
NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
)
stopping_criteria = StoppingCriteria.from_pb( stopping_criteria = StoppingCriteria.from_pb(
r.stopping_parameters, tokenizer r.stopping_parameters, tokenizer
) )

View File

@ -100,7 +100,9 @@ class GalacticaCausalLMBatch(CausalLMBatch):
# Add escape_custom_split_sequence to the CausalLMBatch logic # Add escape_custom_split_sequence to the CausalLMBatch logic
inputs.append(escape_custom_split_sequence(r.inputs)) inputs.append(escape_custom_split_sequence(r.inputs))
input_lengths.append(r.input_length) input_lengths.append(r.input_length)
next_token_choosers.append(NextTokenChooser.from_pb(r.parameters, device)) next_token_choosers.append(
NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
)
stopping_criterias.append( stopping_criterias.append(
StoppingCriteria.from_pb(r.stopping_parameters, tokenizer) StoppingCriteria.from_pb(r.stopping_parameters, tokenizer)
) )

View File

@ -77,7 +77,9 @@ class Seq2SeqLMBatch(Batch):
# Decoder sequence only contains the bos_token # Decoder sequence only contains the bos_token
decoder_input_ids.append(tokenizer.bos_token_id) decoder_input_ids.append(tokenizer.bos_token_id)
decoder_input_lengths.append(1) decoder_input_lengths.append(1)
next_token_choosers.append(NextTokenChooser.from_pb(r.parameters, device)) next_token_choosers.append(
NextTokenChooser.from_pb(r.parameters, len(tokenizer), device)
)
stopping_criteria = StoppingCriteria.from_pb( stopping_criteria = StoppingCriteria.from_pb(
r.stopping_parameters, tokenizer r.stopping_parameters, tokenizer
) )

View File

@ -13,6 +13,7 @@ from typing import List, Tuple, Optional
from text_generation.pb import generate_pb2 from text_generation.pb import generate_pb2
from text_generation.pb.generate_pb2 import FinishReason from text_generation.pb.generate_pb2 import FinishReason
from text_generation.utils.watermark import WatermarkLogitsProcessor
class Sampling: class Sampling:
@ -35,6 +36,8 @@ class Greedy:
class NextTokenChooser: class NextTokenChooser:
def __init__( def __init__(
self, self,
vocab_size,
watermark=False,
temperature=1.0, temperature=1.0,
repetition_penalty=1.0, repetition_penalty=1.0,
top_k=None, top_k=None,
@ -47,6 +50,11 @@ class NextTokenChooser:
# the following idea is largely copied from this PR: https://github.com/huggingface/transformers/pull/5420/files # the following idea is largely copied from this PR: https://github.com/huggingface/transformers/pull/5420/files
# all samplers can be found in `generation_utils_samplers.py` # all samplers can be found in `generation_utils_samplers.py`
sampling = do_sample sampling = do_sample
if watermark:
warpers.append(WatermarkLogitsProcessor(vocab_size, device=device))
if repetition_penalty is not None and repetition_penalty != 1.0:
warpers.append(RepetitionPenaltyLogitsProcessor(penalty=repetition_penalty))
if temperature is not None and temperature != 1.0: if temperature is not None and temperature != 1.0:
temperature = float(temperature) temperature = float(temperature)
warpers.append(TemperatureLogitsWarper(temperature)) warpers.append(TemperatureLogitsWarper(temperature))
@ -57,8 +65,6 @@ class NextTokenChooser:
if top_p is not None and top_p < 1.0: if top_p is not None and top_p < 1.0:
warpers.append(TopPLogitsWarper(top_p=top_p)) warpers.append(TopPLogitsWarper(top_p=top_p))
sampling = True sampling = True
if repetition_penalty is not None and repetition_penalty != 1.0:
warpers.append(RepetitionPenaltyLogitsProcessor(penalty=repetition_penalty))
self.warpers = warpers self.warpers = warpers
self.choice = Sampling(seed, device) if sampling else Greedy() self.choice = Sampling(seed, device) if sampling else Greedy()
@ -77,9 +83,14 @@ class NextTokenChooser:
@classmethod @classmethod
def from_pb( def from_pb(
cls, pb: generate_pb2.NextTokenChooserParameters, device: torch.device cls,
pb: generate_pb2.NextTokenChooserParameters,
vocab_size: int,
device: torch.device,
) -> "NextTokenChooser": ) -> "NextTokenChooser":
return NextTokenChooser( return NextTokenChooser(
vocab_size=vocab_size,
watermark=pb.watermark,
temperature=pb.temperature, temperature=pb.temperature,
repetition_penalty=pb.repetition_penalty, repetition_penalty=pb.repetition_penalty,
top_k=pb.top_k, top_k=pb.top_k,

View File

@ -0,0 +1,87 @@
# coding=utf-8
# Copyright 2023 Authors of "A Watermark for Large Language Models"
# available at https://arxiv.org/abs/2301.10226
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import os
import torch
from transformers import LogitsProcessor
GAMMA = os.getenv("WATERMARK_GAMMA", 0.5)
DELTA = os.getenv("WATERMARK_DELTA", 2.0)
class WatermarkLogitsProcessor(LogitsProcessor):
def __init__(
self,
vocab_size: int,
gamma: float = GAMMA,
delta: float = DELTA,
hash_key: int = 15485863, # just a large prime number to create a rng seed with sufficient bit width
device: str = "cpu",
):
# watermarking parameters
self.vocab_size = vocab_size
self.gamma = gamma
self.delta = delta
self.rng = torch.Generator(device=device)
self.hash_key = hash_key
def _seed_rng(self, input_ids: torch.LongTensor) -> None:
assert (
input_ids.shape[-1] >= 1
), "requires at least a 1 token prefix sequence to seed rng"
prev_token = input_ids[-1].item()
self.rng.manual_seed(self.hash_key * prev_token)
def _get_greenlist_ids(self, input_ids: torch.LongTensor) -> list[int]:
# seed the rng using the previous tokens/prefix
self._seed_rng(input_ids)
greenlist_size = int(self.vocab_size * self.gamma)
vocab_permutation = torch.randperm(
self.vocab_size, device=input_ids.device, generator=self.rng
)
greenlist_ids = vocab_permutation[:greenlist_size]
return greenlist_ids
@staticmethod
def _calc_greenlist_mask(
scores: torch.FloatTensor, greenlist_token_ids
) -> torch.BoolTensor:
green_tokens_mask = torch.zeros_like(scores)
green_tokens_mask[-1, greenlist_token_ids] = 1
final_mask = green_tokens_mask.bool()
return final_mask
@staticmethod
def _bias_greenlist_logits(
scores: torch.Tensor, greenlist_mask: torch.Tensor, greenlist_bias: float
) -> torch.Tensor:
scores[greenlist_mask] = scores[greenlist_mask] + greenlist_bias
return scores
def __call__(
self, input_ids: torch.LongTensor, scores: torch.FloatTensor
) -> torch.FloatTensor:
assert len(input_ids) == 1
greenlist_ids = self._get_greenlist_ids(input_ids[0])
green_tokens_mask = self._calc_greenlist_mask(
scores=scores, greenlist_token_ids=greenlist_ids
)
scores = self._bias_greenlist_logits(
scores=scores, greenlist_mask=green_tokens_mask, greenlist_bias=self.delta
)
return scores