File size: 11,175 Bytes
aea8975 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 |
# Copyright 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# Redistribution and use in source and binary forms, with or without
# modification, are permitted provided that the following conditions
# are met:
# * Redistributions of source code must retain the above copyright
# notice, this list of conditions and the following disclaimer.
# * Redistributions in binary form must reproduce the above copyright
# notice, this list of conditions and the following disclaimer in the
# documentation and/or other materials provided with the distribution.
# * Neither the name of NVIDIA CORPORATION nor the names of its
# contributors may be used to endorse or promote products derived
# from this software without specific prior written permission.
#
# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS ``AS IS'' AND ANY
# EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
# IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
# PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
# CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
# OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
# (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
import json
import numpy as np
import triton_python_backend_utils as pb_utils
from transformers import AutoTokenizer
class TritonPythonModel:
"""Your Python model must use the same class name. Every Python model
that is created must have "TritonPythonModel" as the class name.
"""
def initialize(self, args):
"""`initialize` is called only once when the model is being loaded.
Implementing `initialize` function is optional. This function allows
the model to initialize any state associated with this model.
Parameters
----------
args : dict
Both keys and values are strings. The dictionary keys and values are:
* model_config: A JSON string containing the model configuration
* model_instance_kind: A string containing model instance kind
* model_instance_device_id: A string containing model instance device ID
* model_repository: Model repository path
* model_version: Model version
* model_name: Model name
"""
# Parse model configs
model_config = json.loads(args['model_config'])
tokenizer_dir = model_config['parameters']['tokenizer_dir'][
'string_value']
skip_special_tokens = model_config['parameters'].get(
'skip_special_tokens')
if skip_special_tokens is not None:
skip_special_tokens_str = skip_special_tokens[
'string_value'].lower()
if skip_special_tokens_str in [
'true', 'false', '1', '0', 't', 'f', 'y', 'n', 'yes', 'no'
]:
self.skip_special_tokens = skip_special_tokens_str in [
'true', '1', 't', 'y', 'yes'
]
else:
print(
f"[TensorRT-LLM][WARNING] Don't setup 'skip_special_tokens' correctly (set value is {skip_special_tokens['string_value']}). Set it as True by default."
)
self.skip_special_tokens = True
else:
print(
f"[TensorRT-LLM][WARNING] Don't setup 'skip_special_tokens'. Set it as True by default."
)
self.skip_special_tokens = True
self.tokenizer = AutoTokenizer.from_pretrained(tokenizer_dir,
legacy=False,
padding_side='left',
trust_remote_code=True)
if not self.tokenizer.pad_token:
self.tokenizer.pad_token = self.tokenizer.eos_token
# Parse model output configs
output_config = pb_utils.get_output_config_by_name(
model_config, "OUTPUT")
# Convert Triton types to numpy types
self.output_dtype = pb_utils.triton_string_to_numpy(
output_config['data_type'])
def execute(self, requests):
"""`execute` must be implemented in every Python model. `execute`
function receives a list of pb_utils.InferenceRequest as the only
argument. This function is called when an inference is requested
for this model. Depending on the batching configuration (e.g. Dynamic
Batching) used, `requests` may contain multiple requests. Every
Python model, must create one pb_utils.InferenceResponse for every
pb_utils.InferenceRequest in `requests`. If there is an error, you can
set the error argument when creating a pb_utils.InferenceResponse.
Parameters
----------
requests : list
A list of pb_utils.InferenceRequest
Returns
-------
list
A list of pb_utils.InferenceResponse. The length of this list must
be the same as `requests`
"""
responses = []
# Every Python backend must iterate over everyone of the requests
# and create a pb_utils.InferenceResponse for each of them.
for idx, request in enumerate(requests):
# Get input tensors
tokens_batch = pb_utils.get_input_tensor_by_name(
request, 'TOKENS_BATCH').as_numpy()
# Get sequence length
sequence_lengths = pb_utils.get_input_tensor_by_name(
request, 'SEQUENCE_LENGTH').as_numpy()
# Get cum log probs
cum_log_probs = pb_utils.get_input_tensor_by_name(
request, 'CUM_LOG_PROBS')
# Get sequence length
output_log_probs = pb_utils.get_input_tensor_by_name(
request, 'OUTPUT_LOG_PROBS')
# Get context logits
context_logits = pb_utils.get_input_tensor_by_name(
request, 'CONTEXT_LOGITS')
# Get generation logits
generation_logits = pb_utils.get_input_tensor_by_name(
request, 'GENERATION_LOGITS')
# Get the batch index
batch_index = pb_utils.get_input_tensor_by_name(
request, 'BATCH_INDEX')
# Reshape Input
# tokens_batch = tokens_batch.reshape([-1, tokens_batch.shape[0]])
# tokens_batch = tokens_batch.T
# Postprocessing output data.
outputs = self._postprocessing(tokens_batch, sequence_lengths)
# Create output tensors. You need pb_utils.Tensor
# objects to create pb_utils.InferenceResponse.
output_tensor = pb_utils.Tensor(
'OUTPUT',
np.array(outputs).astype(self.output_dtype))
outputs = []
outputs.append(output_tensor)
if cum_log_probs:
out_cum_log_probs = pb_utils.Tensor('OUT_CUM_LOG_PROBS',
cum_log_probs.as_numpy())
outputs.append(out_cum_log_probs)
else:
out_cum_log_probs = pb_utils.Tensor(
'OUT_CUM_LOG_PROBS', np.array([[0.0]], dtype=np.float32))
outputs.append(out_cum_log_probs)
if output_log_probs:
out_output_log_probs = pb_utils.Tensor(
'OUT_OUTPUT_LOG_PROBS', output_log_probs.as_numpy())
outputs.append(out_output_log_probs)
else:
out_output_log_probs = pb_utils.Tensor(
'OUT_OUTPUT_LOG_PROBS',
np.array([[[0.0]]], dtype=np.float32))
outputs.append(out_output_log_probs)
if context_logits:
out_context_logits = pb_utils.Tensor('OUT_CONTEXT_LOGITS',
context_logits.as_numpy())
outputs.append(out_context_logits)
else:
out_context_logits = pb_utils.Tensor(
'OUT_CONTEXT_LOGITS', np.array([[[0.0]]],
dtype=np.float32))
outputs.append(out_context_logits)
if generation_logits:
out_generation_logits = pb_utils.Tensor(
'OUT_GENERATION_LOGITS', generation_logits.as_numpy())
outputs.append(out_generation_logits)
else:
out_generation_logits = pb_utils.Tensor(
'OUT_GENERATION_LOGITS',
np.array([[[[0.0]]]], dtype=np.float32))
outputs.append(out_generation_logits)
if batch_index:
out_batch_index = pb_utils.Tensor('OUT_BATCH_INDEX',
batch_index.as_numpy())
outputs.append(out_batch_index)
else:
out_batch_index = pb_utils.Tensor(
'OUT_BATCH_INDEX', np.array([[0]], dtype=np.int32))
outputs.append(out_batch_index)
# Create InferenceResponse. You can set an error here in case
# there was a problem with handling this inference request.
# Below is an example of how you can set errors in inference
# response:
#
# pb_utils.InferenceResponse(
# output_tensors=..., TritonError("An error occurred"))
inference_response = pb_utils.InferenceResponse(
output_tensors=outputs)
responses.append(inference_response)
# You should return a list of pb_utils.InferenceResponse. Length
# of this list must match the length of `requests` list.
return responses
def finalize(self):
"""`finalize` is called only once when the model is being unloaded.
Implementing `finalize` function is optional. This function allows
the model to perform any necessary clean ups before exit.
"""
print('Cleaning up...')
def _postprocessing(self, tokens_batch, sequence_lengths):
outputs = []
for batch_idx, beam_tokens in enumerate(tokens_batch):
for beam_idx, tokens in enumerate(beam_tokens):
seq_len = sequence_lengths[batch_idx][beam_idx]
# Exclude fake ids in multimodal models
fake_id_len = 0
for i in range(seq_len):
if tokens[i] < self.tokenizer.vocab_size:
fake_id_len = i
break
output = self.tokenizer.decode(
tokens[fake_id_len:seq_len],
skip_special_tokens=self.skip_special_tokens)
outputs.append(output.encode('utf8'))
return outputs
|