# SPDX-FileCopyrightText: Copyright (c) 2022-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# 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 math
from collections import OrderedDict
import numpy as np
import tensorrt as trt
from ..._common import default_net
from ..._utils import pad_vocab_size, str_dtype_to_trt
from ...functional import (PositionEmbeddingType, Tensor, assertion, concat,
constant, gather_last_token_logits, gpt_attention,
shape, split)
from ...layers import (MLP, AttentionMaskType, AttentionParams, ColumnLinear,
Embedding, KeyValueCacheParams, LayerNorm, RowLinear)
from ...mapping import Mapping
from ...module import Module, ModuleList
from ...parameter import Parameter
from ...quantization import QuantMode
class ChatGLMAttention(Module):
def __init__(self,
hidden_size,
num_attention_heads,
max_position_embeddings,
num_layers=1,
apply_query_key_layer_scaling=False,
bias=True,
dtype=None,
position_embedding_type:
PositionEmbeddingType = PositionEmbeddingType.learned_absolute,
use_int8_kv_cache=False,
tp_group=None,
tp_size=1,
multi_block_mode=False,
multi_query_mode=False):
super().__init__()
self.attention_mask_type = AttentionMaskType.bidirectional
self.attention_head_size = hidden_size // num_attention_heads
self.num_attention_heads = num_attention_heads // tp_size
self.num_attention_kv_heads = 1 if multi_query_mode else self.num_attention_heads
self.hidden_size = hidden_size // tp_size
self.max_position_embeddings = max_position_embeddings
self.num_layers = num_layers
self.apply_query_key_layer_scaling = apply_query_key_layer_scaling
self.norm_factor = math.sqrt(self.attention_head_size)
self.q_scaling = 1
if self.apply_query_key_layer_scaling:
self.norm_factor *= self.num_layers
self.q_scaling *= self.num_layers
self.multi_block_mode = multi_block_mode
self.multi_query_mode = multi_query_mode
self.rotary_embedding_dim = 0
self.position_embedding_type = position_embedding_type
self.dtype = dtype
self.use_int8_kv_cache = use_int8_kv_cache
if self.use_int8_kv_cache:
self.kv_orig_quant_scale = Parameter(shape=(1, ), dtype='float32')
self.kv_quant_orig_scale = Parameter(shape=(1, ), dtype='float32')
else:
self.register_parameter('kv_orig_quant_scale', None)
self.register_parameter('kv_quant_orig_scale', None)
# Note: in multi_query_mode, only query heads are split between multiple GPUs,
# while key/value head are not split as there is only one head per key/value.
# The output feature size is therefore (h/tp + 2) * d, where h is num_heads,
# d is head_size, and tp is tensor_parallel_size.
# In ColumnLinear op, the output dim is calculated by (h + 2*tp) * d / tp,
# which matches the desired output size (h/tp + 2) * d after splitting
self.qkv = ColumnLinear(hidden_size,
hidden_size *
3 if not multi_query_mode else hidden_size +
2 * tp_size * self.attention_head_size,
bias=bias,
dtype=dtype,
tp_group=tp_group,
tp_size=tp_size,
gather_output=False)
self.dense = RowLinear(hidden_size,
hidden_size,
bias=bias,
dtype=dtype,
tp_group=tp_group,
tp_size=tp_size)
def forward(self,
hidden_states: Tensor,
position_embedding,
use_cache=False,
kv_cache_params=None,
attention_params=None):
if not default_net().plugin_config.gpt_attention_plugin:
raise ValueError(
'ChatGLM is only supported with GPTAttention plugin')
assert isinstance(hidden_states, Tensor)
qkv = self.qkv(hidden_states)
# attention
qkv = qkv.view(
concat([
shape(qkv, 0),
shape(qkv, 1), self.num_attention_heads, 3,
self.attention_head_size
]))
query, key, value = split(qkv, 1, dim=3)
query = query.view(
concat([
shape(qkv, 0),
shape(qkv, 1), self.num_attention_heads,
self.attention_head_size
]))
key = key.view(
concat([
shape(qkv, 0),
shape(qkv, 1), self.num_attention_heads,
self.attention_head_size
]))
value = value.view(
concat([
shape(qkv, 0),
shape(qkv, 1), self.num_attention_heads,
self.attention_head_size
]))
zero = constant(
np.ascontiguousarray(
np.zeros([1, 1, 1, 1],
dtype=np.float16
if self.dtype == trt.float16 else np.float32)))
def rotate(x64):
x32_part0, x32_part1 = x64.split(32, dim=-1)
x32_part1_negtive = zero - x32_part1
y64 = concat([x32_part1_negtive, x32_part0], dim=3)
return y64
def rotate_embedding(x, position_embedding_value):
cos0, cos1, sin0, sin1 = position_embedding_value
x128 = x
x64_part0, x64_part1 = x128.split(64, dim=-1)
x64_part0_rotate = rotate(x64_part0)
y64_part0 = x64_part0 * cos0 + x64_part0_rotate * sin0
x64_part1_rotate = rotate(x64_part1)
y64_part1 = x64_part1 * cos1 + x64_part1_rotate * sin1
y128 = concat([y64_part0, y64_part1], dim=3)
y128 = y128.view(shape(x))
return y128
query = rotate_embedding(query, position_embedding)
key = rotate_embedding(key, position_embedding)
kv_orig_quant_scale = self.kv_orig_quant_scale.value if self.use_int8_kv_cache else None
kv_quant_orig_scale = self.kv_quant_orig_scale.value if self.use_int8_kv_cache else None
qkv = concat([query, key, value], dim=2)
qkv = qkv.view(
concat([shape(qkv, 0),
shape(qkv, 1), self.hidden_size * 3]))
context, past_key_value = gpt_attention(
tensor=qkv,
past_key_value=kv_cache_params.get_first_past_key_value(),
sequence_length=attention_params.sequence_length,
host_past_key_value_lengths=kv_cache_params.
host_past_key_value_lengths,
context_lengths=attention_params.context_lengths,
cache_indirection=kv_cache_params.cache_indirection,
host_request_types=attention_params.host_request_types,
num_heads=self.num_attention_heads,
num_kv_heads=self.num_attention_kv_heads,
hidden_size_per_head=self.attention_head_size,
q_scaling=self.q_scaling,
rotary_embedding_dim=self.rotary_embedding_dim,
position_embedding_type=self.position_embedding_type,
multi_block_mode=self.multi_block_mode,
kv_orig_quant_scale=kv_orig_quant_scale,
kv_quant_orig_scale=kv_quant_orig_scale,
kv_cache_quant_mode=QuantMode.from_description(
use_int8_kv_cache=self.use_int8_kv_cache),
max_context_length=attention_params.max_context_length,
mask_type=self.attention_mask_type.value,
host_context_lengths=attention_params.host_context_lengths)
context = self.dense(context)
if use_cache:
return (context, past_key_value)
else:
return context
class ChatGLM6BDecoderLayer(Module):
def __init__(self,
hidden_size,
num_attention_heads,
max_position_embeddings,
num_layers,
dtype=None,
apply_query_key_layer_scaling=False,
hidden_act='relu',
quant_mode=QuantMode(0),
inter_size=None,
bias=True,
tp_group=None,
tp_size=1):
super().__init__()
self.hidden_size = hidden_size
self.num_attention_heads = num_attention_heads
self.max_position_embeddings = max_position_embeddings
self.num_layers = num_layers
self.dtype = dtype
self.apply_query_key_layer_scaling = apply_query_key_layer_scaling
self.hidden_act = hidden_act
self.tp_group = tp_group
self.tp_size = tp_size
self.input_layernorm = LayerNorm(normalized_shape=hidden_size,
dtype=dtype)
self.attention = ChatGLMAttention(
hidden_size,
num_attention_heads,
max_position_embeddings,
num_layers,
apply_query_key_layer_scaling,
dtype=dtype,
position_embedding_type=PositionEmbeddingType.learned_absolute,
bias=bias,
tp_group=tp_group,
tp_size=tp_size,
use_int8_kv_cache=quant_mode.has_int8_kv_cache())
if inter_size is None:
inter_size = hidden_size * 4
self.mlp = MLP(hidden_size=hidden_size,
ffn_hidden_size=inter_size,
hidden_act=hidden_act,
dtype=dtype,
bias=bias,
tp_group=tp_group,
tp_size=tp_size)
self.post_layernorm = LayerNorm(normalized_shape=hidden_size,
dtype=dtype)
def forward(self,
hidden_states: Tensor,
position_embedding,
use_cache=False,
kv_cache_params=None,
attention_params=None):
assert isinstance(hidden_states, Tensor)
hidden_states = self.input_layernorm(hidden_states)
attention_output = self.attention(hidden_states,
position_embedding,
use_cache=use_cache,
kv_cache_params=kv_cache_params,
attention_params=attention_params)
if use_cache:
attention_output, presents = attention_output
hidden_states = hidden_states * 7.484375 + attention_output
hidden_states = self.post_layernorm(hidden_states)
mlp_output = self.mlp(hidden_states)
hidden_states = hidden_states * 7.484375 + mlp_output
if use_cache:
return (hidden_states, presents)
return hidden_states
[docs]class ChatGLM6BModel(Module):
def __init__(self,
num_layers,
num_heads,
hidden_size,
vocab_size,
hidden_act,
max_position_embeddings,
dtype=None,
mapping=Mapping(),
apply_query_key_layer_scaling=False,
inter_size=None,
bias=True,
quant_mode=QuantMode(0)):
super().__init__()
self.half_head_size = hidden_size // num_heads // 2
self.embedding = Embedding(vocab_size, hidden_size, dtype=dtype)
self.position_embedding_cos = Embedding(max_position_embeddings,
self.half_head_size,
dtype=dtype)
self.position_embedding_sin = Embedding(max_position_embeddings,
self.half_head_size,
dtype=dtype)
self.layers = ModuleList([
ChatGLM6BDecoderLayer(
hidden_size=hidden_size,
num_attention_heads=num_heads,
max_position_embeddings=max_position_embeddings,
num_layers=num_layers,
dtype=dtype,
apply_query_key_layer_scaling=apply_query_key_layer_scaling,
hidden_act=hidden_act,
tp_group=mapping.tp_group,
tp_size=mapping.tp_size,
inter_size=inter_size,
bias=bias,
quant_mode=quant_mode) for _ in range(num_layers)
])
self.ln_f = LayerNorm(normalized_shape=hidden_size, dtype=dtype)
[docs] def forward(self,
input_ids=None,
position_ids=None,
use_cache=False,
kv_cache_params=None,
attention_params=None):
batch_size = shape(input_ids, 0)
input_len = shape(input_ids, 1)
hidden_states = self.embedding(input_ids)
position_embedding_cos = self.position_embedding_cos(position_ids)
position_embedding_sin = self.position_embedding_sin(position_ids)
position_embedding_cos0, position_embedding_cos1 = position_embedding_cos.split(
1, dim=1)
position_embedding_sin0, position_embedding_sin1 = position_embedding_sin.split(
1, dim=1)
position_embedding_cos0 = position_embedding_cos0.view(
concat([batch_size, input_len, 1, self.half_head_size]))
position_embedding_cos1 = position_embedding_cos1.view(
concat([batch_size, input_len, 1, self.half_head_size]))
position_embedding_sin0 = position_embedding_sin0.view(
concat([batch_size, input_len, 1, self.half_head_size]))
position_embedding_sin1 = position_embedding_sin1.view(
concat([batch_size, input_len, 1, self.half_head_size]))
position_embedding = [
position_embedding_cos0, position_embedding_cos1,
position_embedding_sin0, position_embedding_sin1
]
if kv_cache_params.past_key_value is None:
kv_cache_params.past_key_value = tuple([None] * len(self.layers))
if use_cache:
presents = []
for layer, past in zip(self.layers, kv_cache_params.past_key_value):
hidden_states = layer(
hidden_states,
position_embedding,
use_cache=use_cache,
kv_cache_params=KeyValueCacheParams(
past_key_value=[past],
host_past_key_value_lengths=kv_cache_params.
host_past_key_value_lengths,
cache_indirection=kv_cache_params.cache_indirection),
attention_params=attention_params)
if use_cache:
presents.append(hidden_states[1])
hidden_states = hidden_states[0]
hidden_states = self.ln_f(hidden_states)
if use_cache:
return (hidden_states, tuple(presents))
return hidden_states
[docs]class ChatGLM6BHeadModel(ChatGLM6BModel):
def __init__(self,
num_layers,
num_heads,
hidden_size,
vocab_size,
hidden_act,
max_position_embeddings,
dtype,
mapping=Mapping(),
apply_query_key_layer_scaling=False,
inter_size=None,
bias=True,
quant_mode=QuantMode(0)):
if isinstance(dtype, str):
self._kv_dtype = str_dtype_to_trt(dtype)
else:
assert isinstance(dtype, trt.DataType)
self._kv_dtype = dtype
self._dtype = self._kv_dtype
if quant_mode.has_int8_kv_cache():
self._kv_dtype = str_dtype_to_trt('int8')
elif quant_mode.has_fp8_kv_cache():
self._kv_dtype = str_dtype_to_trt('fp8')
self.quant_mode = quant_mode
self._num_layers = num_layers
self._num_heads = num_heads
self._hidden_size = hidden_size
self._vocab_size = vocab_size
self._tp_size = mapping.tp_size
super().__init__(num_layers, num_heads, hidden_size, vocab_size,
hidden_act, max_position_embeddings, dtype, mapping,
apply_query_key_layer_scaling, inter_size, bias,
quant_mode)
vocab_size_padded = pad_vocab_size(vocab_size, mapping.tp_size)
self.lm_head = ColumnLinear(hidden_size,
vocab_size_padded,
bias=False,
dtype=dtype,
tp_group=mapping.tp_group,
tp_size=mapping.tp_size,
gather_output=True)
[docs] def forward(self,
input_ids=None,
position_ids=None,
use_cache=False,
last_token_ids=None,
kv_cache_params=None,
attention_params=None):
hidden_states = super().forward(input_ids, position_ids, use_cache,
kv_cache_params, attention_params)
if use_cache:
hidden_states, presents = hidden_states
hidden_states = gather_last_token_logits(
hidden_states, last_token_ids,
default_net().plugin_config.remove_input_padding)
# [batch_size, hidden_size] -> [batch_size, vocab_size]
lm_logits = self.lm_head(hidden_states)
lm_logits.mark_output('logits', self._dtype)
# out_inter.mark_output('inter', str_dtype_to_trt('float32'))
if use_cache and default_net().plugin_config.paged_kv_cache == False:
for i, present in enumerate(presents):
present.mark_output(f'present_key_value_{i}', self._kv_dtype)
return (lm_logits, presents)
return lm_logits