doc-to-lora/tmp/test_emb.py
2025-09-28 15:46:47 +00:00

51 lines
1.6 KiB
Python

import os
import torch
from ctx_to_lora.model_loading import get_model, get_tokenizer
# tokenizer = AutoTokenizer.from_pretrained("google/gemma-2-2b-it")
# model = AutoModelForCausalLM.from_pretrained(
# "google/gemma-2-2b-it",
# device_map="auto",
# torch_dtype=torch.bfloat16,
# attn_implementation="sdpa",
# )
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
os.environ["TRANSFORMERS_NO_ADVISORY_WARNINGS"] = "true"
os.environ["FLASH_ATTENTION_DETERMINISTIC"] = "1"
# torch.use_deterministic_algorithms(True, warn_only=True)
torch.backends.cuda.matmul.allow_fp16_reduced_precision_reduction = False
torch.backends.cuda.matmul.allow_bf16_reduced_precision_reduction = False
torch.backends.cudnn.benchmark = False
torch.backends.cuda.matmul.allow_tf32 = False
torch.backends.cudnn.allow_tf32 = False
# model, tokenizer = get_model_and_tokenizer(
# "google/gemma-3-1b-it",
# train=False,
# requires_grad=False,
# use_flash_attn=True,
# peft_config=None,
# model_kwargs={"attn_implementation": "flash_attention_2"},
# tokenizer_kwargs=None,
# device="cuda",
# dtype=torch.bfloat16,
# )
device = "cuda" if torch.cuda.is_available() else "cpu"
model_kwargs = {"attn_implementation": "flash_attention_2"}
# model_name_or_path = "google/gemma-2-2b-it"
model_name_or_path = "google/gemma-2-2b-it"
tokenizer = get_tokenizer(model_name_or_path, train=False)
model = get_model(
model_name_or_path,
train=False,
requires_grad=False,
model_kwargs=model_kwargs,
use_flash_attn=True,
dtype=torch.bfloat16,
)
breakpoint()