Skip to content

[mieb] Fix vlm2vec #1380

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
gowitheflow-1998 merged 2 commits into from
Nov 5, 2024
Merged

[mieb] Fix vlm2vec #1380

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
e92e6ab
fix vlm2vec return dtype
isaac-chung Nov 4, 2024
f10dffa
make lint
isaac-chung Nov 4, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
57 changes: 26 additions & 31 deletions mteb/models/vlm2vec_models.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -149,7 +149,7 @@ def get_image_embeddings(
}

image_outputs = self.encode_input(inputs)
all_image_embeddings.append(image_outputs.cpu())
all_image_embeddings.append(image_outputs.cpu().to(torch.float32))

else:
with torch.no_grad():
Expand Down Expand Up @@ -186,7 +186,7 @@ def get_image_embeddings(
}

image_outputs = self.encode_input(inputs)
all_image_embeddings.append(image_outputs.cpu())
all_image_embeddings.append(image_outputs.cpu().to(torch.float32))

all_image_embeddings = torch.cat(all_image_embeddings, dim=0)
return all_image_embeddings
Expand Down Expand Up @@ -221,7 +221,7 @@ def get_text_embeddings(self, texts: list[str], batch_size: int = 32):
}

text_outputs = self.encode_input(inputs)
all_text_embeddings.append(text_outputs.cpu())
all_text_embeddings.append(text_outputs.cpu().to(torch.float32))

all_text_embeddings = torch.cat(all_text_embeddings, dim=0)
return all_text_embeddings
Expand All @@ -239,39 +239,34 @@ def get_fused_embeddings(
text_embeddings = None
image_embeddings = None

if texts is not None:
if texts is not None and images is None:
text_embeddings = self.get_text_embeddings(texts, batch_size)
return text_embeddings

if images is not None:
if images is not None and texts is None:
image_embeddings = self.get_image_embeddings(images, batch_size)

if text_embeddings is not None and image_embeddings is not None:
if len(text_embeddings) != len(image_embeddings):
raise ValueError(
"The number of texts and images must have the same length"
)
texts = iter(texts)
all_fused_embeddings = []
if isinstance(images, DataLoader):
import torchvision.transforms.functional as F
with torch.no_grad():
for batch in images:
for b in batch:
text = next(texts)
inputs = self.processor(
f"<|image_1|> Represent the given image with the following question: {text}",
[F.to_pil_image(b.to("cpu"))],
)
inputs = {k: v.to(self.device) for k, v in inputs.items()}
outputs = self.encode_input(inputs)
all_fused_embeddings.append(outputs.cpu())
fused_embeddings = torch.cat(all_fused_embeddings, dim=0)
return fused_embeddings
elif text_embeddings is not None:
return text_embeddings
elif image_embeddings is not None:
return image_embeddings

# text_embeddings is not None and image_embeddings is not None
texts = iter(texts)
all_fused_embeddings = []
if isinstance(images, DataLoader):
import torchvision.transforms.functional as F

with torch.no_grad():
for batch in images:
for b in batch:
text = next(texts)
inputs = self.processor(
f"<|image_1|> Represent the given image with the following question: {text}",
[F.to_pil_image(b.to("cpu"))],
)
inputs = {k: v.to(self.device) for k, v in inputs.items()}
outputs = self.encode_input(inputs)
all_fused_embeddings.append(outputs.cpu().to(torch.float32))
fused_embeddings = torch.cat(all_fused_embeddings, dim=0)
return fused_embeddings


vlm2vec_lora = ModelMeta(
loader=partial(
Expand Down
186 changes: 186 additions & 0 deletions .../TIGER-Lab__VLM2Vec-LoRA/7403b6327958071c1e33c822c7453adadccc7298/BLINKIT2IRetrieval.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,186 @@
{
"dataset_revision": "359b66f11c25d19bc8f7108d98e660a5857f3d26",
"evaluation_time": 224.3045289516449,
"kg_co2_emissions": null,
"mteb_version": "1.14.21",
"scores": {
"test": [
{
"cv_recall_at_1": 0.33582,
"cv_recall_at_10": 0.63184,
"cv_recall_at_100": 0.90547,
"cv_recall_at_1000": 1.0,
"cv_recall_at_20": 0.72637,
"cv_recall_at_3": 0.52736,
"cv_recall_at_5": 0.57711,
"hf_subset": "default",
"languages": [
"eng-Latn"
],
"main_score": 0.48677,
"map_at_1": 0.33582,
"map_at_10": 0.43983,
"map_at_100": 0.45072,
"map_at_1000": 0.45136,
"map_at_20": 0.44629,
"map_at_3": 0.4204,
"map_at_5": 0.43209,
"mrr_at_1": 0.3358208955223881,
"mrr_at_10": 0.43982666034904844,
"mrr_at_100": 0.45071603631711843,
"mrr_at_1000": 0.4513620025552411,
"mrr_at_20": 0.44629389891710264,
"mrr_at_3": 0.42039800995024873,
"mrr_at_5": 0.43208955223880596,
"nauc_cv_recall_at_1000_diff1": NaN,
"nauc_cv_recall_at_1000_max": NaN,
"nauc_cv_recall_at_1000_std": NaN,
"nauc_cv_recall_at_100_diff1": 0.2698006811472229,
"nauc_cv_recall_at_100_max": -0.8185369176412147,
"nauc_cv_recall_at_100_std": -1.0075573331300645,
"nauc_cv_recall_at_10_diff1": 0.1788243841076317,
"nauc_cv_recall_at_10_max": -0.612746605676537,
"nauc_cv_recall_at_10_std": -0.9530034638964193,
"nauc_cv_recall_at_1_diff1": 0.2822466928213175,
"nauc_cv_recall_at_1_max": -0.33285460282946566,
"nauc_cv_recall_at_1_std": -0.5640630200327668,
"nauc_cv_recall_at_20_diff1": 0.12960665874064498,
"nauc_cv_recall_at_20_max": -0.8021930752909697,
"nauc_cv_recall_at_20_std": -1.071727841531356,
"nauc_cv_recall_at_3_diff1": 0.2019990705156492,
"nauc_cv_recall_at_3_max": -0.4459723288652778,
"nauc_cv_recall_at_3_std": -0.7870326458461854,
"nauc_cv_recall_at_5_diff1": 0.23846869773197804,
"nauc_cv_recall_at_5_max": -0.47064146280491637,
"nauc_cv_recall_at_5_std": -0.8505490027952863,
"nauc_map_at_1000_diff1": 0.2409455106402186,
"nauc_map_at_1000_max": -0.407384704974283,
"nauc_map_at_1000_std": -0.6813392812409054,
"nauc_map_at_100_diff1": 0.24104039617379056,
"nauc_map_at_100_max": -0.408250752001871,
"nauc_map_at_100_std": -0.682039826842636,
"nauc_map_at_10_diff1": 0.2411578991609476,
"nauc_map_at_10_max": -0.4044975322245294,
"nauc_map_at_10_std": -0.6840284218652672,
"nauc_map_at_1_diff1": 0.2822466928213175,
"nauc_map_at_1_max": -0.33285460282946566,
"nauc_map_at_1_std": -0.5640630200327668,
"nauc_map_at_20_diff1": 0.2388129109858802,
"nauc_map_at_20_max": -0.4111868495020421,
"nauc_map_at_20_std": -0.6861574124125431,
"nauc_map_at_3_diff1": 0.2415896795362189,
"nauc_map_at_3_max": -0.3816620658278678,
"nauc_map_at_3_std": -0.6597288106948174,
"nauc_map_at_5_diff1": 0.2490237529189825,
"nauc_map_at_5_max": -0.3874920448893905,
"nauc_map_at_5_std": -0.6729363831482362,
"nauc_mrr_at_1000_diff1": 0.2409455106402186,
"nauc_mrr_at_1000_max": -0.407384704974283,
"nauc_mrr_at_1000_std": -0.6813392812409054,
"nauc_mrr_at_100_diff1": 0.24104039617379056,
"nauc_mrr_at_100_max": -0.408250752001871,
"nauc_mrr_at_100_std": -0.682039826842636,
"nauc_mrr_at_10_diff1": 0.2411578991609476,
"nauc_mrr_at_10_max": -0.4044975322245294,
"nauc_mrr_at_10_std": -0.6840284218652672,
"nauc_mrr_at_1_diff1": 0.2822466928213175,
"nauc_mrr_at_1_max": -0.33285460282946566,
"nauc_mrr_at_1_std": -0.5640630200327668,
"nauc_mrr_at_20_diff1": 0.2388129109858802,
"nauc_mrr_at_20_max": -0.4111868495020421,
"nauc_mrr_at_20_std": -0.6861574124125431,
"nauc_mrr_at_3_diff1": 0.2415896795362189,
"nauc_mrr_at_3_max": -0.3816620658278678,
"nauc_mrr_at_3_std": -0.6597288106948174,
"nauc_mrr_at_5_diff1": 0.2490237529189825,
"nauc_mrr_at_5_max": -0.3874920448893905,
"nauc_mrr_at_5_std": -0.6729363831482362,
"nauc_ndcg_at_1000_diff1": 0.2344092745371185,
"nauc_ndcg_at_1000_max": -0.4328446999637412,
"nauc_ndcg_at_1000_std": -0.709474885782739,
"nauc_ndcg_at_100_diff1": 0.23614140799135372,
"nauc_ndcg_at_100_max": -0.45059659752764697,
"nauc_ndcg_at_100_std": -0.7232249956899242,
"nauc_ndcg_at_10_diff1": 0.22732780326872332,
"nauc_ndcg_at_10_max": -0.44870094654610376,
"nauc_ndcg_at_10_std": -0.7429856899210265,
"nauc_ndcg_at_1_diff1": 0.2822466928213175,
"nauc_ndcg_at_1_max": -0.33285460282946566,
"nauc_ndcg_at_1_std": -0.5640630200327668,
"nauc_ndcg_at_20_diff1": 0.21923921046650915,
"nauc_ndcg_at_20_max": -0.4772077987509219,
"nauc_ndcg_at_20_std": -0.7553659757754395,
"nauc_ndcg_at_3_diff1": 0.23134535418875965,
"nauc_ndcg_at_3_max": -0.39788809593745006,
"nauc_ndcg_at_3_std": -0.691825805701685,
"nauc_ndcg_at_5_diff1": 0.24552033909645363,
"nauc_ndcg_at_5_max": -0.4076854503791393,
"nauc_ndcg_at_5_std": -0.7156597141210952,
"nauc_precision_at_1000_diff1": NaN,
"nauc_precision_at_1000_max": NaN,
"nauc_precision_at_1000_std": NaN,
"nauc_precision_at_100_diff1": 0.2698006811472233,
"nauc_precision_at_100_max": -0.818536917641218,
"nauc_precision_at_100_std": -1.007557333130069,
"nauc_precision_at_10_diff1": 0.17882438410763166,
"nauc_precision_at_10_max": -0.6127466056765365,
"nauc_precision_at_10_std": -0.9530034638964187,
"nauc_precision_at_1_diff1": 0.2822466928213175,
"nauc_precision_at_1_max": -0.33285460282946566,
"nauc_precision_at_1_std": -0.5640630200327668,
"nauc_precision_at_20_diff1": 0.12960665874064572,
"nauc_precision_at_20_max": -0.8021930752909691,
"nauc_precision_at_20_std": -1.0717278415313547,
"nauc_precision_at_3_diff1": 0.20199907051564917,
"nauc_precision_at_3_max": -0.44597232886527743,
"nauc_precision_at_3_std": -0.7870326458461852,
"nauc_precision_at_5_diff1": 0.238468697731978,
"nauc_precision_at_5_max": -0.47064146280491626,
"nauc_precision_at_5_std": -0.8505490027952862,
"nauc_recall_at_1000_diff1": NaN,
"nauc_recall_at_1000_max": NaN,
"nauc_recall_at_1000_std": NaN,
"nauc_recall_at_100_diff1": 0.2698006811472229,
"nauc_recall_at_100_max": -0.8185369176412147,
"nauc_recall_at_100_std": -1.0075573331300645,
"nauc_recall_at_10_diff1": 0.1788243841076317,
"nauc_recall_at_10_max": -0.612746605676537,
"nauc_recall_at_10_std": -0.9530034638964193,
"nauc_recall_at_1_diff1": 0.2822466928213175,
"nauc_recall_at_1_max": -0.33285460282946566,
"nauc_recall_at_1_std": -0.5640630200327668,
"nauc_recall_at_20_diff1": 0.12960665874064498,
"nauc_recall_at_20_max": -0.8021930752909697,
"nauc_recall_at_20_std": -1.071727841531356,
"nauc_recall_at_3_diff1": 0.2019990705156492,
"nauc_recall_at_3_max": -0.4459723288652778,
"nauc_recall_at_3_std": -0.7870326458461854,
"nauc_recall_at_5_diff1": 0.23846869773197804,
"nauc_recall_at_5_max": -0.47064146280491637,
"nauc_recall_at_5_std": -0.8505490027952863,
"ndcg_at_1": 0.33582,
"ndcg_at_10": 0.48677,
"ndcg_at_100": 0.54341,
"ndcg_at_1000": 0.55646,
"ndcg_at_20": 0.51055,
"ndcg_at_3": 0.44788,
"ndcg_at_5": 0.46865,
"precision_at_1": 0.33582,
"precision_at_10": 0.06318,
"precision_at_100": 0.00905,
"precision_at_1000": 0.001,
"precision_at_20": 0.03632,
"precision_at_3": 0.17579,
"precision_at_5": 0.11542,
"recall_at_1": 0.33582,
"recall_at_10": 0.63184,
"recall_at_100": 0.90547,
"recall_at_1000": 1.0,
"recall_at_20": 0.72637,
"recall_at_3": 0.52736,
"recall_at_5": 0.57711
}
]
},
"task_name": "BLINKIT2IRetrieval"
}