[Bugfix] Fix unable to load some models (#10312)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-11-14 16:55:54 -08:00
committed by GitHub
parent 11cd1ae6ad
commit 972112d82f
13 changed files with 340 additions and 59 deletions
+5 -2
View File
@@ -41,7 +41,8 @@ from vllm.sequence import (VLLM_TOKEN_ID_ARRAY_TYPE, IntermediateTensors,
from vllm.utils import is_list_of
from .interfaces import SupportsMultiModal, SupportsPP
from .utils import AutoWeightsLoader, flatten_bn, merge_multimodal_embeddings
from .utils import (AutoWeightsLoader, flatten_bn, maybe_prefix,
merge_multimodal_embeddings)
# Cannot find the following 2 numbers from hf config.
_IMAGE_TOKEN_ID = 71011
@@ -245,7 +246,9 @@ class FuyuForCausalLM(nn.Module, SupportsMultiModal, SupportsPP):
gather_output=True,
)
self.language_model = PersimmonForCausalLM(
vllm_config.with_hf_config(config.text_config))
vllm_config=vllm_config.with_hf_config(config.text_config),
prefix=maybe_prefix(prefix, "language_model"),
)
self.make_empty_intermediate_tensors = (
self.language_model.make_empty_intermediate_tensors)
+1 -7
View File
@@ -161,11 +161,5 @@ class InternLM2VEForCausalLM(InternLM2ForCausalLM):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
config = vllm_config.model_config.hf_config
cache_config = vllm_config.cache_config
quant_config = vllm_config.quant_config
self.model = InternLM2VEModel(config,
cache_config,
quant_config,
self.model = InternLM2VEModel(vllm_config=vllm_config,
prefix=maybe_prefix(prefix, "model"))
+8 -24
View File
@@ -382,11 +382,7 @@ class MiniCPMVBaseModel(nn.Module, SupportsMultiModal, SupportsPP):
instantiated.
"""
def __init__(
self,
vllm_config: VllmConfig,
prefix: str = "",
):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
config = vllm_config.model_config.hf_config
multimodal_config = vllm_config.model_config.multimodal_config
quant_config = vllm_config.quant_config
@@ -699,12 +695,8 @@ class MiniCPMVBaseModel(nn.Module, SupportsMultiModal, SupportsPP):
class MiniCPMV2_0(MiniCPMVBaseModel):
def __init__(
self,
vllm_config: VllmConfig,
prefix: str = "",
):
super().__init__(vllm_config)
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
assert self.version == (2, 0)
def init_llm(
@@ -857,12 +849,8 @@ class MiniCPMV2_5(MiniCPMVBaseModel, SupportsLoRA):
embedding_modules = {}
embedding_padding_modules = []
def __init__(
self,
vllm_config: VllmConfig,
prefix: str = "",
):
super().__init__(vllm_config)
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
assert self.version == (2, 5)
def init_llm(
@@ -999,12 +987,8 @@ class MiniCPMV2_6(MiniCPMVBaseModel, SupportsLoRA):
embedding_modules = {}
embedding_padding_modules = []
def __init__(
self,
vllm_config: VllmConfig,
prefix: str = "",
):
super().__init__(vllm_config)
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
assert self.version == (2, 6)
def init_llm(
@@ -1117,7 +1101,7 @@ class MiniCPMV(MiniCPMVBaseModel, SupportsLoRA):
embedding_modules = {}
embedding_padding_modules = []
def __new__(cls, vllm_config: VllmConfig, prefix: str = ""):
def __new__(cls, *, vllm_config: VllmConfig, prefix: str = ""):
config = vllm_config.model_config.hf_config
if not hasattr(config, "version"):
if config.hidden_size == 2304 and config.query_num == 64:
+1 -1
View File
@@ -65,7 +65,7 @@ class MLPSpeculator(nn.Module):
https://huggingface.co/ibm-fms and https://huggingface.co/ibm-granite
"""
def __init__(self, vllm_config: VllmConfig, prefix: str = "") -> None:
def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
super().__init__()
config = vllm_config.model_config.hf_config
self.n_predict = config.n_predict
+6 -2
View File
@@ -1,3 +1,7 @@
"""
Whenever you add an architecture to this page, please also update
`tests/models/registry.py` with example HuggingFace models for it.
"""
import importlib
import os
import pickle
@@ -58,14 +62,14 @@ _TEXT_GENERATION_MODELS = {
"LLaMAForCausalLM": ("llama", "LlamaForCausalLM"),
"MambaForCausalLM": ("mamba", "MambaForCausalLM"),
"FalconMambaForCausalLM": ("mamba", "MambaForCausalLM"),
"MiniCPMForCausalLM": ("minicpm", "MiniCPMForCausalLM"),
"MiniCPM3ForCausalLM": ("minicpm3", "MiniCPM3ForCausalLM"),
"MistralForCausalLM": ("llama", "LlamaForCausalLM"),
"MixtralForCausalLM": ("mixtral", "MixtralForCausalLM"),
"QuantMixtralForCausalLM": ("mixtral_quant", "MixtralForCausalLM"),
# transformers's mpt class has lower case
"MptForCausalLM": ("mpt", "MPTForCausalLM"),
"MPTForCausalLM": ("mpt", "MPTForCausalLM"),
"MiniCPMForCausalLM": ("minicpm", "MiniCPMForCausalLM"),
"MiniCPM3ForCausalLM": ("minicpm3", "MiniCPM3ForCausalLM"),
"NemotronForCausalLM": ("nemotron", "NemotronForCausalLM"),
"OlmoForCausalLM": ("olmo", "OlmoForCausalLM"),
"OlmoeForCausalLM": ("olmoe", "OlmoeForCausalLM"),