423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-08-23 22:32:54 +08:00
Code Issues Packages Projects Releases Wiki Activity

remove tests

Browse Source 
Former-commit-id: a019cece8009b0ba8a6b5a309ed5abfe6cb88a75
This commit is contained in:
fzc8578 2025-01-13 15:08:35 +08:00
parent ee87d318b8
commit 313ce9a576
2 changed files with 0 additions and 37 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/llamafactory/data/mm_plugin.py
View File

@ -383,7 +383,6 @@ class CpmOPlugin(BasePlugin):
self._validate_input(images, videos) self._validate_input(images, videos)
image_bounds_list = [] image_bounds_list = []
valid_image_nums_ls = [] valid_image_nums_ls = []
flag = False
for input_ids in batch_ids: for input_ids in batch_ids:
input_ids_ = torch.tensor(input_ids) input_ids_ = torch.tensor(input_ids)
@ -395,8 +394,6 @@ class CpmOPlugin(BasePlugin):
image_start_tokens += 1 image_start_tokens += 1
image_end_tokens = torch.where(end_cond)[0] image_end_tokens = torch.where(end_cond)[0]
valid_image_nums = max(len(image_start_tokens), len(image_end_tokens)) valid_image_nums = max(len(image_start_tokens), len(image_end_tokens))
if valid_image_nums > 0:
flag = True
valid_image_nums_ls.append(valid_image_nums) valid_image_nums_ls.append(valid_image_nums)
image_bounds = torch.hstack( image_bounds = torch.hstack(
[ [
@ -406,10 +403,6 @@ class CpmOPlugin(BasePlugin):
) )
image_bounds_list.append(image_bounds) image_bounds_list.append(image_bounds)
if not flag and len(images) > 0:
valid_image_nums_ls = [1 for _ in range(len(batch_ids))]
image_bounds_list = [torch.arange(64) for _ in range(len(batch_ids))]
mm_inputs = self._get_mm_inputs(images, videos, processor, valid_image_nums_ls=valid_image_nums_ls) mm_inputs = self._get_mm_inputs(images, videos, processor, valid_image_nums_ls=valid_image_nums_ls)
mm_inputs.update({"image_bound": image_bounds_list}) mm_inputs.update({"image_bound": image_bounds_list})
return mm_inputs return mm_inputs

30
tests/data/test_mm_plugin.py
View File

@ -79,17 +79,6 @@ def _is_close(batch_a: Dict[str, Any], batch_b: Dict[str, Any]) -> None:
assert len(batch_a[key]) == len(batch_b[key]) assert len(batch_a[key]) == len(batch_b[key])
for tensor_a, tensor_b in zip(batch_a[key], batch_b[key]): for tensor_a, tensor_b in zip(batch_a[key], batch_b[key]):
assert torch.allclose(tensor_a, tensor_b, rtol=1e-4, atol=1e-5) assert torch.allclose(tensor_a, tensor_b, rtol=1e-4, atol=1e-5)
elif (
isinstance(batch_a[key], list)
and all(isinstance(item, list) for item in batch_a[key])
and len(batch_a[key]) > 0
and len(batch_a[key][0]) > 0
and isinstance(batch_a[key][0][0], torch.Tensor)
):
for item_a, item_b in zip(batch_a[key], batch_b[key]):
assert len(item_a) == len(item_a)
for tensor_a, tensor_b in zip(item_a, item_b):
assert torch.allclose(tensor_a, tensor_b, rtol=1e-4, atol=1e-5)
else: else:
assert batch_a[key] == batch_b[key] assert batch_a[key] == batch_b[key]
@ -138,25 +127,6 @@ def test_base_plugin():
_check_plugin(**check_inputs) _check_plugin(**check_inputs)
@pytest.mark.skipif(not HF_TOKEN, reason="Gated model.")
def test_cpm_o_plugin():
tokenizer_module = _load_tokenizer_module(model_name_or_path="/data/fengzc/LLM/checkpoints/MiniCPM-V-2_6")
cpm_o_plugin = get_mm_plugin(name="cpm_o", image_token="<image>")
check_inputs = {"plugin": cpm_o_plugin, **tokenizer_module}
image_seqlen = 64
check_inputs["expected_mm_messages"] = [
{
key: value.replace("<image>", f"<image_id>0</image_id><image>{'<unk>' * image_seqlen}</image>")
for key, value in message.items()
}
for message in MM_MESSAGES
]
check_inputs["expected_mm_inputs"] = _get_mm_inputs(tokenizer_module["processor"])
check_inputs["expected_mm_inputs"]["image_bound"] = [torch.arange(64)]
check_inputs["expected_no_mm_inputs"] = {"image_bound": [torch.tensor([], dtype=torch.int64).reshape(0, 2)]}
_check_plugin(**check_inputs)
def test_llava_plugin(): def test_llava_plugin():
image_seqlen = 576 image_seqlen = 576
tokenizer_module = _load_tokenizer_module(model_name_or_path="llava-hf/llava-1.5-7b-hf") tokenizer_module = _load_tokenizer_module(model_name_or_path="llava-hf/llava-1.5-7b-hf")