ros/LLaMA-Factory
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

add unittest

Browse Source 
Former-commit-id: 8a1f0c5f922989e08a19c65de0b2c4afd2a5771f
This commit is contained in:
hiyouga
2024-07-19 01:06:27 +08:00
parent 4c1513a845
commit 994b9089e9
16 changed files with 436 additions and 260 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/model/model_utils/test_attention.py
View File

@@ -16,8 +16,7 @@ import os
from transformers.utils import is_flash_attn_2_available, is_torch_sdpa_available
from llamafactory.hparams import get_infer_args
from llamafactory.model import load_model, load_tokenizer
from llamafactory.train.test_utils import load_infer_model
TINY_LLAMA = os.environ.get("TINY_LLAMA", "llamafactory/tiny-random-Llama-3")
@@ -42,9 +41,7 @@ def test_attention():
"fa2": "LlamaFlashAttention2",
}
for requested_attention in attention_available:
model_args, _, finetuning_args, _ = get_infer_args({"flash_attn": requested_attention, **INFER_ARGS})
tokenizer_module = load_tokenizer(model_args)
model = load_model(tokenizer_module["tokenizer"], model_args, finetuning_args)
model = load_infer_model(flash_attn=requested_attention, **INFER_ARGS)
for module in model.modules():
if "Attention" in module.__class__.__name__:
assert module.__class__.__name__ == llama_attention_classes[requested_attention]