8e192ff967
Co-authored-by: beagleski <yunanzhang@microsoft.com> Co-authored-by: bapatra <bapatra@microsoft.com> Co-authored-by: Barun Patra <codedecde@users.noreply.github.com> Co-authored-by: Michael Goin <michael@neuralmagic.com>
66 lines
2.6 KiB
Python
66 lines
2.6 KiB
Python
from typing import Dict, Optional
|
|
|
|
from transformers import AutoConfig, PretrainedConfig
|
|
|
|
from vllm.logger import init_logger
|
|
from vllm.transformers_utils.configs import (ChatGLMConfig, DbrxConfig,
|
|
JAISConfig, MPTConfig, RWConfig)
|
|
|
|
logger = init_logger(__name__)
|
|
|
|
_CONFIG_REGISTRY: Dict[str, PretrainedConfig] = {
|
|
"chatglm": ChatGLMConfig,
|
|
"dbrx": DbrxConfig,
|
|
"mpt": MPTConfig,
|
|
"RefinedWeb": RWConfig, # For tiiuae/falcon-40b(-instruct)
|
|
"RefinedWebModel": RWConfig, # For tiiuae/falcon-7b(-instruct)
|
|
"jais": JAISConfig,
|
|
}
|
|
|
|
|
|
def get_config(model: str,
|
|
trust_remote_code: bool,
|
|
revision: Optional[str] = None,
|
|
code_revision: Optional[str] = None,
|
|
rope_scaling: Optional[dict] = None) -> PretrainedConfig:
|
|
try:
|
|
config = AutoConfig.from_pretrained(
|
|
model,
|
|
trust_remote_code=trust_remote_code,
|
|
revision=revision,
|
|
code_revision=code_revision)
|
|
except ValueError as e:
|
|
if (not trust_remote_code and
|
|
"requires you to execute the configuration file" in str(e)):
|
|
err_msg = (
|
|
"Failed to load the model config. If the model is a custom "
|
|
"model not yet available in the HuggingFace transformers "
|
|
"library, consider setting `trust_remote_code=True` in LLM "
|
|
"or using the `--trust-remote-code` flag in the CLI.")
|
|
raise RuntimeError(err_msg) from e
|
|
else:
|
|
raise e
|
|
if config.model_type in _CONFIG_REGISTRY:
|
|
config_class = _CONFIG_REGISTRY[config.model_type]
|
|
config = config_class.from_pretrained(model,
|
|
revision=revision,
|
|
code_revision=code_revision)
|
|
if rope_scaling is not None:
|
|
logger.info("Updating rope_scaling from %r to %r",
|
|
getattr(config, "rope_scaling", None), rope_scaling)
|
|
config.update({"rope_scaling": rope_scaling})
|
|
return config
|
|
|
|
|
|
def get_hf_text_config(config: PretrainedConfig):
|
|
"""Get the "sub" config relevant to llm for multi modal models.
|
|
No op for pure text models.
|
|
"""
|
|
if hasattr(config, "text_config"):
|
|
# The code operates under the assumption that text_config should have
|
|
# `num_attention_heads` (among others). Assert here to fail early
|
|
# if transformers config doesn't align with this assumption.
|
|
assert hasattr(config.text_config, "num_attention_heads")
|
|
return config.text_config
|
|
else:
|
|
return config |