mirror of
https://github.com/KoboldAI/KoboldAI-Client.git
synced 2025-06-05 21:59:24 +02:00
83 lines
2.9 KiB
Python
83 lines
2.9 KiB
Python
import os
|
|
from typing import Optional
|
|
from transformers import AutoConfig
|
|
|
|
import utils
|
|
from logger import logger
|
|
from modeling.inference_model import InferenceModel
|
|
|
|
|
|
class HFInferenceModel(InferenceModel):
|
|
def __init__(self, model_name: str) -> None:
|
|
super().__init__()
|
|
self.model_config = None
|
|
self.model_name = model_name
|
|
|
|
self.model = None
|
|
self.tokenizer = None
|
|
|
|
def get_local_model_path(
|
|
self, legacy: bool = False, ignore_existance: bool = False
|
|
) -> Optional[str]:
|
|
"""
|
|
Returns a string of the model's path locally, or None if it is not downloaded.
|
|
If ignore_existance is true, it will always return a path.
|
|
"""
|
|
|
|
if self.model_name in ["NeoCustom", "GPT2Custom", "TPUMeshTransformerGPTJ", "TPUMeshTransformerGPTNeoX"]:
|
|
model_path = utils.koboldai_vars.custmodpth
|
|
assert model_path
|
|
|
|
# Path can be absolute or relative to models directory
|
|
if os.path.exists(model_path):
|
|
return model_path
|
|
|
|
model_path = os.path.join("models", model_path)
|
|
|
|
try:
|
|
assert os.path.exists(model_path)
|
|
except AssertionError:
|
|
logger.error(f"Custom model does not exist at '{utils.koboldai_vars.custmodpth}' or '{model_path}'.")
|
|
raise
|
|
|
|
return model_path
|
|
|
|
basename = utils.koboldai_vars.model.replace("/", "_")
|
|
if legacy:
|
|
ret = basename
|
|
else:
|
|
ret = os.path.join("models", basename)
|
|
|
|
if os.path.isdir(ret) or ignore_existance:
|
|
return ret
|
|
return None
|
|
|
|
def init_model_config(self) -> None:
|
|
# Get the model_type from the config or assume a model type if it isn't present
|
|
try:
|
|
self.model_config = AutoConfig.from_pretrained(
|
|
self.get_local_model_path() or self.model_name,
|
|
revision=utils.koboldai_vars.revision,
|
|
cache_dir="cache",
|
|
)
|
|
utils.koboldai_vars.model_type = self.model_config.model_type
|
|
|
|
if "gptq_bits" in dir(self.model_config):
|
|
utils.koboldai_vars.gptq_model = True
|
|
utils.koboldai_vars.gptq_bits = self.model_config.gptq_bits
|
|
utils.koboldai_vars.gptq_groupsize = self.model_config.gptq_groupsize
|
|
utils.koboldai_vars.gptq_file = None
|
|
else:
|
|
utils.koboldai_vars.gptq_model = False
|
|
except ValueError:
|
|
utils.koboldai_vars.model_type = {
|
|
"NeoCustom": "gpt_neo",
|
|
"GPT2Custom": "gpt2",
|
|
}.get(utils.koboldai_vars.model)
|
|
|
|
if not utils.koboldai_vars.model_type:
|
|
logger.warning(
|
|
"No model type detected, assuming Neo (If this is a GPT2 model use the other menu option or --model GPT2Custom)"
|
|
)
|
|
utils.koboldai_vars.model_type = "gpt_neo"
|