[LoRA][2/2]Remove LoRA extra vocab (#28545)
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
@@ -51,7 +51,6 @@ from vllm.model_executor.layers.logits_processor import LogitsProcessor
|
||||
from vllm.model_executor.layers.quantization import QuantizationConfig
|
||||
from vllm.model_executor.layers.rotary_embedding import get_rope
|
||||
from vllm.model_executor.layers.vocab_parallel_embedding import (
|
||||
DEFAULT_VOCAB_PADDING_SIZE,
|
||||
ParallelLMHead,
|
||||
VocabParallelEmbedding,
|
||||
)
|
||||
@@ -301,23 +300,18 @@ class MixtralModel(nn.Module):
|
||||
config = vllm_config.model_config.hf_config
|
||||
cache_config = vllm_config.cache_config
|
||||
quant_config = vllm_config.quant_config
|
||||
lora_config = vllm_config.lora_config
|
||||
|
||||
parallel_config = vllm_config.parallel_config
|
||||
|
||||
self.config = config
|
||||
self.quant_config = quant_config
|
||||
lora_vocab = (
|
||||
(lora_config.lora_extra_vocab_size * (lora_config.max_loras or 1))
|
||||
if lora_config
|
||||
else 0
|
||||
)
|
||||
self.vocab_size = config.vocab_size + lora_vocab
|
||||
|
||||
self.vocab_size = config.vocab_size
|
||||
self.org_vocab_size = config.vocab_size
|
||||
|
||||
self.embed_tokens = VocabParallelEmbedding(
|
||||
self.vocab_size,
|
||||
config.hidden_size,
|
||||
org_num_embeddings=config.vocab_size,
|
||||
)
|
||||
|
||||
self.enable_eplb = parallel_config.enable_eplb
|
||||
@@ -508,34 +502,24 @@ class MixtralForCausalLM(nn.Module, SupportsLoRA, SupportsPP, MixtureOfExperts):
|
||||
super().__init__()
|
||||
config = vllm_config.model_config.hf_config
|
||||
quant_config = vllm_config.quant_config
|
||||
lora_config = vllm_config.lora_config
|
||||
|
||||
self.config = config
|
||||
self.lora_config = lora_config
|
||||
|
||||
self.quant_config = quant_config
|
||||
|
||||
self.model = MixtralModel(
|
||||
vllm_config=vllm_config, prefix=maybe_prefix(prefix, "model")
|
||||
)
|
||||
self.unpadded_vocab_size = config.vocab_size
|
||||
if lora_config:
|
||||
self.unpadded_vocab_size += lora_config.lora_extra_vocab_size
|
||||
|
||||
self.lm_head = ParallelLMHead(
|
||||
self.unpadded_vocab_size,
|
||||
config.vocab_size,
|
||||
config.hidden_size,
|
||||
org_num_embeddings=config.vocab_size,
|
||||
padding_size=DEFAULT_VOCAB_PADDING_SIZE
|
||||
# We need bigger padding if using lora for kernel
|
||||
# compatibility
|
||||
if not lora_config
|
||||
else lora_config.lora_vocab_padding_size,
|
||||
quant_config=quant_config,
|
||||
prefix=maybe_prefix(prefix, "lm_head"),
|
||||
)
|
||||
if self.config.tie_word_embeddings:
|
||||
self.lm_head.weight = self.model.embed_tokens.weight
|
||||
self.logits_processor = LogitsProcessor(
|
||||
self.unpadded_vocab_size, config.vocab_size
|
||||
)
|
||||
self.logits_processor = LogitsProcessor(config.vocab_size)
|
||||
self.make_empty_intermediate_tensors = (
|
||||
self.model.make_empty_intermediate_tensors
|
||||
)
|
||||
|
||||
Reference in New Issue
Block a user