TP/quantization/weight loading refactor part 2 - Refactor quantized linear logic and extend quantization support to all models (#1622)

Refactor the tensor parallelism, quantization, and weight-loading codes.

Summary of the new features enabled by this PR:
- **All models** are able to be quantized with AWQ and SqueezeLLM, and [soon GPTQ](https://github.com/vllm-project/vllm/pull/1580).
- Model loading code became much simpler.
- Support model parallelism for all MQA/GQA models when the number of key/value heads is smaller than the tensor parallel size.
This commit is contained in:
Zhuohan Li
2023-11-15 22:50:41 -08:00
committed by GitHub
parent 660a7fcfa4
commit 7076fa1c9f
36 changed files with 2159 additions and 2508 deletions
+92 -149
View File
@@ -30,17 +30,19 @@ from vllm.model_executor.input_metadata import InputMetadata
from vllm.model_executor.layers.attention import (PagedAttention,
PagedAttentionWithALiBi,
PagedAttentionWithRoPE)
from vllm.model_executor.layers.linear import (ColumnParallelLinear,
LinearMethodBase,
QKVParallelLinear,
RowParallelLinear)
from vllm.model_executor.layers.sampler import Sampler
from vllm.model_executor.weight_utils import (convert_pyslice_to_tensor,
hf_model_weights_iterator,
load_tensor_parallel_weights)
from vllm.model_executor.parallel_utils.parallel_state import (
get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size)
from vllm.model_executor.parallel_utils.layers import (VocabParallelEmbedding,
ColumnParallelLinear,
RowParallelLinear)
from vllm.model_executor.layers.vocab_parallel_embedding import (
VocabParallelEmbedding, ParallelLMHead)
from vllm.model_executor.parallel_utils.communication_op import (
tensor_model_parallel_all_reduce)
from vllm.model_executor.parallel_utils.parallel_state import (
get_tensor_model_parallel_rank, get_tensor_model_parallel_world_size)
from vllm.model_executor.weight_utils import (default_weight_loader,
hf_model_weights_iterator)
from vllm.sequence import SamplerOutput
from vllm.transformers_utils.configs import RWConfig
@@ -48,19 +50,6 @@ KVCache = Tuple[torch.Tensor, torch.Tensor]
FalconConfig = Union[HF_FalconConfig, RWConfig]
# NOTE(Hesslow): Unfortunately we did not fuse matmul and bias during
# training, this means that there's one additional quantization to bfloat16
# between the operations. In order not to degrade the quality of our HF-port,
# we keep these characteristics in the final model.
class FalconLinear(nn.Linear):
def forward(self, x: torch.Tensor) -> torch.Tensor:
hidden_states = x @ self.weight.T
if self.bias is None:
return hidden_states
return hidden_states + self.bias
def _get_alibi_slopes(total_num_heads: int) -> torch.Tensor:
closest_power_of_2 = 2**math.floor(math.log2(total_num_heads))
base = torch.tensor(2**(-(2**-(math.log2(closest_power_of_2) - 3))),
@@ -86,7 +75,11 @@ def _get_alibi_slopes(total_num_heads: int) -> torch.Tensor:
class FalconAttention(nn.Module):
def __init__(self, config: FalconConfig):
def __init__(
self,
config: FalconConfig,
linear_method: Optional[LinearMethodBase] = None,
):
super().__init__()
self.hidden_size = config.hidden_size
@@ -103,41 +96,29 @@ class FalconAttention(nn.Module):
if self.new_decoder_architecture:
self.total_num_kv_heads = config.num_kv_heads
assert self.total_num_heads % tp_size == 0
self.num_kv_heads = self.total_num_kv_heads // tp_size
self.query_key_value = ColumnParallelLinear(
self.hidden_size,
(self.total_num_heads + 2 * self.total_num_kv_heads) *
self.head_dim,
bias=config.bias,
gather_output=False,
skip_bias_add=True,
)
elif self.multi_query:
self.total_num_kv_heads = 1
self.num_kv_heads = 1
self.query = ColumnParallelLinear(
self.hidden_size,
self.total_num_heads * self.head_dim,
bias=config.bias,
gather_output=False,
skip_bias_add=True,
)
self.key_value = FalconLinear(self.hidden_size,
2 * self.head_dim,
bias=config.bias)
else:
self.total_num_kv_heads = self.total_num_heads
self.num_kv_heads = self.num_heads
self.query_key_value = ColumnParallelLinear(
self.hidden_size,
(self.total_num_heads + 2 * self.total_num_kv_heads) *
self.head_dim,
bias=config.bias,
gather_output=False,
skip_bias_add=True,
)
if self.total_num_kv_heads >= tp_size:
# Number of KV heads is greater than TP size, so we partition
# the KV heads across multiple tensor parallel GPUs.
assert self.total_num_kv_heads % tp_size == 0
else:
# Number of KV heads is less than TP size, so we replicate
# the KV heads across multiple tensor parallel GPUs.
assert tp_size % self.total_num_kv_heads == 0
self.num_kv_heads = max(1, self.total_num_kv_heads // tp_size)
self.query_key_value = QKVParallelLinear(
self.hidden_size,
self.head_dim,
self.total_num_heads,
self.total_num_kv_heads,
bias=config.bias,
skip_bias_add=True,
linear_method=linear_method,
)
self.q_size = self.num_heads * self.head_dim
self.kv_size = self.num_kv_heads * self.head_dim
@@ -149,7 +130,6 @@ class FalconAttention(nn.Module):
self.hidden_size,
self.hidden_size,
bias=config.bias,
input_is_parallel=True,
skip_bias_add=True,
reduce_results=self.reduce_row_parallel_results)
@@ -196,18 +176,10 @@ class FalconAttention(nn.Module):
input_metadata: InputMetadata,
cache_event: Optional[torch.cuda.Event],
) -> torch.Tensor:
if not self.new_decoder_architecture and self.multi_query:
q, bias = self.query(hidden_states)
if bias is not None:
q += bias
kv = self.key_value(hidden_states)
k, v = kv.split([self.kv_size, self.kv_size], dim=-1)
else:
qkv, bias = self.query_key_value(hidden_states)
if bias is not None:
qkv += bias
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size],
dim=-1)
qkv, bias = self.query_key_value(hidden_states)
if bias is not None:
qkv += bias
q, k, v = qkv.split([self.q_size, self.kv_size, self.kv_size], dim=-1)
k_cache, v_cache = kv_cache
if self.use_rotary:
attn_output = self.attn(positions, q, k, v, k_cache, v_cache,
@@ -221,15 +193,19 @@ class FalconAttention(nn.Module):
class FalconMLP(nn.Module):
def __init__(self, config: FalconConfig):
def __init__(
self,
config: FalconConfig,
linear_method: Optional[LinearMethodBase] = None,
):
super().__init__()
hidden_size = config.hidden_size
self.dense_h_to_4h = ColumnParallelLinear(hidden_size,
4 * hidden_size,
bias=config.bias,
gather_output=False,
skip_bias_add=True)
skip_bias_add=True,
linear_method=linear_method)
self.act = nn.GELU()
self.reduce_row_parallel_results = not (config.new_decoder_architecture
or config.parallel_attn)
@@ -237,9 +213,9 @@ class FalconMLP(nn.Module):
4 * hidden_size,
hidden_size,
bias=config.bias,
input_is_parallel=True,
skip_bias_add=True,
reduce_results=self.reduce_row_parallel_results)
reduce_results=self.reduce_row_parallel_results,
linear_method=linear_method)
def forward(self, x: torch.Tensor) -> torch.Tensor:
# NOTE(zhuohan): Following huggingface, we do not fuse bias add here.
@@ -253,12 +229,16 @@ class FalconMLP(nn.Module):
class FalconDecoderLayer(nn.Module):
def __init__(self, config: FalconConfig):
def __init__(
self,
config: FalconConfig,
linear_method: Optional[LinearMethodBase] = None,
):
super().__init__()
hidden_size = config.hidden_size
self.num_heads = config.num_attention_heads
self.self_attention = FalconAttention(config)
self.mlp = FalconMLP(config)
self.self_attention = FalconAttention(config, linear_method)
self.mlp = FalconMLP(config, linear_method)
self.config = config
if config.new_decoder_architecture:
@@ -334,7 +314,11 @@ class FalconDecoderLayer(nn.Module):
class FalconModel(nn.Module):
def __init__(self, config: FalconConfig):
def __init__(
self,
config: FalconConfig,
linear_method: Optional[LinearMethodBase] = None,
):
super().__init__()
self.config = config
self.embed_dim = config.hidden_size
@@ -349,7 +333,8 @@ class FalconModel(nn.Module):
# Transformer blocks
self.h = nn.ModuleList([
FalconDecoderLayer(config) for _ in range(config.num_hidden_layers)
FalconDecoderLayer(config, linear_method)
for _ in range(config.num_hidden_layers)
])
# Final Layer Norm
@@ -383,15 +368,18 @@ class FalconModel(nn.Module):
class FalconForCausalLM(nn.Module):
def __init__(self, config: FalconConfig):
def __init__(
self,
config: FalconConfig,
linear_method: Optional[LinearMethodBase] = None,
):
super().__init__()
self.config = config
self.transformer = FalconModel(config)
self.lm_head = ColumnParallelLinear(
config.hidden_size,
self.linear_method = linear_method
self.transformer = FalconModel(config, linear_method)
self.lm_head = ParallelLMHead(
config.vocab_size,
bias=False,
gather_output=False,
config.hidden_size,
)
self.sampler = Sampler(config.vocab_size)
@@ -415,89 +403,44 @@ class FalconForCausalLM(nn.Module):
return next_tokens
_column_parallel_weights = [
"word_embeddings.weight", "lm_head.weight", "dense_h_to_4h.weight",
"dense_h_to_4h.bias"
]
_row_parallel_weights = ["dense.weight", "dense_4h_to_h.weight"]
def load_weights(self,
model_name_or_path: str,
cache_dir: Optional[str] = None,
load_format: str = "auto",
revision: Optional[str] = None):
tp_size = (get_tensor_model_parallel_world_size())
tp_rank = get_tensor_model_parallel_rank()
hidden_size = self.config.hidden_size
total_num_heads = self.config.num_attention_heads
num_heads = total_num_heads // tp_size
head_size = hidden_size // total_num_heads
head_start = tp_rank * num_heads
head_end = (tp_rank + 1) * num_heads
if self.config.new_decoder_architecture:
total_num_kv_heads = self.config.num_kv_heads
num_kv_heads = total_num_kv_heads // tp_size
separated_q_kv = False
kv_head_start = tp_rank * num_kv_heads
kv_head_end = (tp_rank + 1) * num_kv_heads
elif self.config.multi_query:
total_num_kv_heads = 1
num_kv_heads = 1
separated_q_kv = True
kv_head_start = 0
kv_head_end = 1
else:
total_num_kv_heads = total_num_heads
num_kv_heads = total_num_kv_heads // tp_size
separated_q_kv = False
kv_head_start = tp_rank * num_kv_heads
kv_head_end = (tp_rank + 1) * num_kv_heads
num_query_heads_per_kv_head = total_num_heads // total_num_kv_heads
state_dict = self.state_dict()
params_dict = dict(self.named_parameters())
for name, loaded_weight in hf_model_weights_iterator(
model_name_or_path, cache_dir, load_format, revision):
param = params_dict[name]
if "query_key_value" in name:
loaded_weight = convert_pyslice_to_tensor(loaded_weight)
loaded_weight_size = loaded_weight.size()
output_dim = getattr(param, "output_dim", None)
loaded_weight_shape = loaded_weight.shape
loaded_weight = loaded_weight.view(
total_num_kv_heads, num_query_heads_per_kv_head + 2,
head_size, *loaded_weight_size[1:])
loaded_weight_shape[:output_dim] +
(total_num_kv_heads, num_query_heads_per_kv_head + 2, -1) +
loaded_weight_shape[output_dim + 1:])
wq = loaded_weight.narrow(
output_dim + 1, 0, num_query_heads_per_kv_head).reshape(
*loaded_weight_shape[:output_dim], -1,
*loaded_weight_shape[output_dim + 1:])
wk = loaded_weight.narrow(
output_dim + 1, num_query_heads_per_kv_head,
1).reshape(*loaded_weight_shape[:output_dim], -1,
*loaded_weight_shape[output_dim + 1:])
wv = loaded_weight.narrow(
output_dim + 1, num_query_heads_per_kv_head + 1,
1).reshape(*loaded_weight_shape[:output_dim], -1,
*loaded_weight_shape[output_dim + 1:])
loaded_weight = torch.cat([wq, wk, wv], dim=output_dim)
wq = loaded_weight[:, :-2].reshape(-1, *loaded_weight_size[1:])
wk = loaded_weight[:, [-2]].reshape(-1,
*loaded_weight_size[1:])
wv = loaded_weight[:, [-1]].reshape(-1,
*loaded_weight_size[1:])
wq = wq[head_size * head_start:head_size * head_end]
wk = wk[head_size * kv_head_start:head_size * kv_head_end]
wv = wv[head_size * kv_head_start:head_size * kv_head_end]
if separated_q_kv:
loaded_weight_q = wq
loaded_weight_kv = torch.cat([wk, wv], dim=0)
q_weight_name = name.replace("query_key_value", "query")
kv_weight_name = name.replace("query_key_value",
"key_value")
load_tensor_parallel_weights(state_dict[q_weight_name],
loaded_weight_q,
q_weight_name,
self._column_parallel_weights,
self._row_parallel_weights,
tp_rank)
load_tensor_parallel_weights(state_dict[kv_weight_name],
loaded_weight_kv,
kv_weight_name,
self._column_parallel_weights,
self._row_parallel_weights,
tp_rank)
continue
else:
loaded_weight = torch.cat([wq, wk, wv], dim=0)
param = state_dict[name]
load_tensor_parallel_weights(param, loaded_weight, name,
self._column_parallel_weights,
self._row_parallel_weights, tp_rank)
weight_loader = getattr(param, "weight_loader",
default_weight_loader)
weight_loader(param, loaded_weight)