Hugging Face's logo

Marqo
/
marqo-ecommerce-embeddings-B

OpenCLIP
Safetensors
Transformers
siglip
clip
ecommerce
multimodal retrieval
openCLIP
custom_code
Model card Files Community
elliesleightholm commited on
Commit
005825e
•
1 Parent(s): 9f0a42f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +180 -53
README.md CHANGED
@@ -1,50 +1,55 @@
1
- ---
2
- license: apache-2.0
3
- tags:
4
- - clip
5
- - ecommerce
6
- - multimodal retrieval
7
- - transformers
8
- - openCLIP
9
- ---
 
 
 
10
 
11
- # Marqo E-commerce Embedding Models
12
- In this work, we introduce two state-of-the-art embedding models for e-commerce:
13
- Marqo-Ecommerce-B and Marqo-Ecommerce-L.
14
- Trained on over 60M unique products, they are over 30% better compared to Amazon Titan Embedding services for e-commerce retrieval tasks.
15
 
16
- <img src="bar_chart_1.png" alt="multi split visual" width="1024"/>
17
 
 
 
 
18
 
19
  **Released Content**:
20
  1) Marqo-Ecommerce-B and Marqo-Ecommerce-L embedding models
21
  2) GoogleShopping-1m and AmazonProducts-3m for evaluation
22
  3) Evaluation Code
23
 
24
-
25
  ## Models
26
- Two models are available. The B model is smaller and faster for inference with a snaller embedding dimension. The L model is larger, has a larger embedding dimension, but has better retrieval performance.
27
- The models can be loaded in several ways. See below for available options.
28
  | **Embedding Model** | **#Params (m)** | **Dimension** | **HuggingFace** | **Download .pt** |
29
  |---------------------| --- |---------------|------------------------------------|-------------------------------------------------------------------------------------------------------------|
30
- | Marqo-Ecommerce-B | 203 | 768 | Marqo/marqo-ecommerce-embeddings-B | [link](https://marqo-gcl-public.s3.us-west-2.amazonaws.com/marqo-general-ecomm/marqo-ecomm-embeddings-b.pt) |
31
- | Marqo-Ecommerce-L | 652 | 1024 | Marqo/marqo-ecommerce-embeddings-L | [link](https://marqo-gcl-public.s3.us-west-2.amazonaws.com/marqo-general-ecomm/marqo-ecomm-embeddings-l.pt) |
 
 
 
32
 
33
- ### HuggingFace with transformers
34
  ```python
35
  from transformers import AutoModel, AutoProcessor
36
  import torch
37
  from PIL import Image
38
  import requests
39
- # model_name= 'Marqo/marqo-ecommerce-embeddings-L'
40
- model_name = 'Marqo/marqo-ecommerce-embeddings-B'
 
41
 
42
  model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
43
  processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)
44
 
45
- img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-FashionCLIP/main/docs/fashion-hippo.png', stream=True).raw).convert("RGB")
46
  image = [img]
47
- text = ["a hat", "a t-shirt", "shoes"]
48
  processed = processor(text=text, images=image, padding='max_length', return_tensors="pt")
49
  processor.image_processor.do_rescale = False
50
  with torch.no_grad():
@@ -54,10 +59,12 @@ with torch.no_grad():
54
  text_probs = (100 * image_features @ text_features.T).softmax(dim=-1)
55
 
56
  print(text_probs)
57
- # [9.9955e-01, 4.4712e-04, 4.4010e-06]]
58
  ```
59
 
60
- ### HuggingFace with OpenCLIP
 
 
61
  ```
62
  pip install open_clip_torch
63
  ```
@@ -69,14 +76,16 @@ import torch
69
 
70
  # Specify model from Hugging Face Hub
71
  model_name = 'hf-hub:Marqo/marqo-ecommerce-embeddings-B'
 
 
72
  model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms(model_name)
73
  tokenizer = open_clip.get_tokenizer(model_name)
74
 
75
  # Preprocess the image and tokenize text inputs
76
  # Load an example image from a URL
77
- img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-FashionCLIP/main/docs/fashion-hippo.png', stream=True).raw)
78
  image = preprocess_val(img).unsqueeze(0)
79
- text = tokenizer(["a hat", "a t-shirt", "shoes"])
80
 
81
  # Perform inference
82
  with torch.no_grad(), torch.cuda.amp.autocast():
@@ -88,18 +97,22 @@ with torch.no_grad(), torch.cuda.amp.autocast():
88
 
89
  # Display the label probabilities
90
  print("Label probs:", text_probs)
91
- # [9.9955e-01, 4.4712e-04, 4.4010e-06]]
92
  ```
93
 
94
- ### Evaluation with GCL
 
 
95
  ```
96
  git clone https://github.com/marqo-ai/GCL
97
  ```
98
  Install the packages required by GCL.
 
 
99
  ```
100
  cd ./GCL
101
  MODEL=hf-hub:Marqo/marqo-ecommerce-B
102
- outdir=/MarqoModels/GE/marqo-ecommerce-B/gs-title2image2
103
  hfdataset=Marqo/google-shopping-general-eval
104
  python evals/eval_hf_datasets_v1.py \
105
  --model_name $MODEL \
@@ -118,39 +131,153 @@ python evals/eval_hf_datasets_v1.py \
118
  --context-length "[[64], [0]]"
119
  ```
120
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
121
 
122
  ## Detailed Performance
123
- **GoogleShopping-Text2Image Retrieval.**
 
 
124
 
125
- | **Embedding Model** | **mAP** | **P@10** | **R@10** | **MRR** |
126
- | --- | --- | --- | --- | --- |
127
- | Marqo-Ecommerce-L | **0.682** | **0.089** | **0.878** | **0.683** |
128
- | Marqo-Ecommerce-B | 0.623 | 0.084 | 0.832 | 0.624 |
129
- | Amazon-Titan-MultiModal | 0.475 | 0.065 | 0.648 | 0.475 |
130
- | ViT-B-16-SigLip | 0.476 | 0.067 | 0.660 | 0.477 |
131
- | ViT-L-16-SigLip | 0.540 | 0.073 | 0.722 | 0.540 |
132
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
133
 
134
  **GoogleShopping-Category2Image Retrieval.**
135
 
136
- | **Embedding Model** | **mAP** | **P@10** | **MRR** | **nDCG@10** |
137
- | --- | --- | --- | --- | --- |
138
- | Marqo-Ecommerce-L | **0.463** | **0.652** | **0.822** | **0.666** |
139
- | Marqo-Ecommerce-B | 0.423 | 0.629 | 0.810 | 0.644 |
140
- | Amazon-Titan-MultiModal | 0.246 | 0.429 | 0.642 | 0.446 |
141
- | ViT-B-16-SigLip | 0.277 | 0.458 | 0.660 | 0.473 |
142
- | ViT-L-16-SigLip | 0.324 | 0.497 | 0.687 | 0.509 |
 
 
143
 
144
  **AmazonProducts-Text2Image Retrieval.**
145
 
146
- | **Embedding Model** | **mAP** | **P@10** | **R@10** | **MRR** |
147
- | --- | --- | --- | --- | --- |
148
- | Marqo-Ecommerce-L | **0.658** | **0.096** | **0.854** | **0.663** |
149
- | Marqo-Ecommerce-B | 0.592 | 0.089 | 0.795 | 0.597 |
150
- | Amazon-Titan-MultiModal | 0.456 | 0.064 | 0.627 | 0.457 |
151
- | ViT-B-16-SigLip | 0.480 | 0.070 | 0.650 | 0.484 |
152
- | ViT-L-16-SigLip | 0.544 | 0.077 | 0.715 | 0.548 |
 
 
153
 
 
 
154
 
 
155
 
 
 
 
 
 
 
 
 
 
 
 
156
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ tags:
4
+ - clip
5
+ - ecommerce
6
+ - multimodal retrieval
7
+ - transformers
8
+ - openCLIP
9
+ datasets:
10
+ - Marqo/amazon-products-eval
11
+ - Marqo/google-shopping-general-eval
12
+ ---
13
 
14
+ # Marqo Ecommerce Embedding Models
15
+ In this work, we introduce two state-of-the-art embedding models for ecommerce products: Marqo-Ecommerce-B and Marqo-Ecommerce-L.
 
 
16
 
17
+ The benchmarking results highlight a remarkable performance by marqo-ecommerce models, which both consistently outperformed all other models across various metrics. Specifically, for the Google Shopping Text-to-Image task, marqo-ecommerce-L achieved an improvement of 43% in MRR, 41% in nDCG@10 and 33% in Recall@10 when compared to ViT-B-16-SigLIP which is our baseline model for these benchmarks. For the Google Shopping Category-to-Image task, we saw an improvement of 67% in mAP, 41% in nDCG@10 and 42% in Precision@10.
18
 
19
+ <img src="https://raw.githubusercontent.com/marqo-ai/marqo-ecommerce-embeddings/refs/heads/main/performance.png?token=GHSAT0AAAAAACZY3OVLSH7USBXOC3SYCQ3OZZL6LVQ" alt="multi split visual" width="700"/>
20
+
21
+ More benchmarking results can be found below.
22
 
23
  **Released Content**:
24
  1) Marqo-Ecommerce-B and Marqo-Ecommerce-L embedding models
25
  2) GoogleShopping-1m and AmazonProducts-3m for evaluation
26
  3) Evaluation Code
27
 
 
28
  ## Models
29
+
 
30
  | **Embedding Model** | **#Params (m)** | **Dimension** | **HuggingFace** | **Download .pt** |
31
  |---------------------| --- |---------------|------------------------------------|-------------------------------------------------------------------------------------------------------------|
32
+ | Marqo-Ecommerce-B | 203 | 768 | [Marqo/marqo-ecommerce-embeddings-B](https://huggingface.co/Marqo/marqo-ecommerce-embeddings-B) | [link](https://marqo-gcl-public.s3.us-west-2.amazonaws.com/marqo-general-ecomm/marqo-ecomm-embeddings-b.pt) |
33
+ | Marqo-Ecommerce-L | 652 | 1024 | [Marqo/marqo-ecommerce-embeddings-L](https://huggingface.co/Marqo/marqo-ecommerce-embeddings-L) | [link](https://marqo-gcl-public.s3.us-west-2.amazonaws.com/marqo-general-ecomm/marqo-ecomm-embeddings-l.pt) |
34
+
35
+ ### Load from HuggingFace with transformers
36
+ To load the models in Transformers, see below. The models are hosted on [Hugging Face](https://huggingface.co/collections/Marqo/marqo-ecommerce-embeddings-66f611b9bb9d035a8d164fbb) and loaded using [Transformers](https://github.com/huggingface/transformers).
37
 
 
38
  ```python
39
  from transformers import AutoModel, AutoProcessor
40
  import torch
41
  from PIL import Image
42
  import requests
43
+
44
+ model_name= 'Marqo/marqo-ecommerce-embeddings-B'
45
+ # model_name = 'Marqo/marqo-ecommerce-embeddings-L'
46
 
47
  model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
48
  processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)
49
 
50
+ img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-ecommerce-embeddings/refs/heads/main/images/dining-chairs.png', stream=True).raw).convert("RGB")
51
  image = [img]
52
+ text = ["dining chairs", "a laptop", "toothbrushes"]
53
  processed = processor(text=text, images=image, padding='max_length', return_tensors="pt")
54
  processor.image_processor.do_rescale = False
55
  with torch.no_grad():
 
59
  text_probs = (100 * image_features @ text_features.T).softmax(dim=-1)
60
 
61
  print(text_probs)
62
+ # [1.0000e+00, 8.3131e-12, 5.2173e-12]
63
  ```
64
 
65
+ ### Load from HuggingFace with OpenCLIP
66
+ To load the models in OpenCLIP, see below. The models are hosted on [Hugging Face](https://huggingface.co/collections/Marqo/marqo-ecommerce-embeddings-66f611b9bb9d035a8d164fbb) and loaded using [OpenCLIP](https://github.com/mlfoundations/open_clip). You can also find this code inside `run_models.py`.
67
+
68
  ```
69
  pip install open_clip_torch
70
  ```
 
76
 
77
  # Specify model from Hugging Face Hub
78
  model_name = 'hf-hub:Marqo/marqo-ecommerce-embeddings-B'
79
+ # model_name = 'hf-hub:Marqo/marqo-ecommerce-embeddings-L'
80
+
81
  model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms(model_name)
82
  tokenizer = open_clip.get_tokenizer(model_name)
83
 
84
  # Preprocess the image and tokenize text inputs
85
  # Load an example image from a URL
86
+ img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-ecommerce-embeddings/refs/heads/main/images/dining-chairs.png', stream=True).raw)
87
  image = preprocess_val(img).unsqueeze(0)
88
+ text = tokenizer(["dining chairs", "a laptop", "toothbrushes"])
89
 
90
  # Perform inference
91
  with torch.no_grad(), torch.cuda.amp.autocast():
 
97
 
98
  # Display the label probabilities
99
  print("Label probs:", text_probs)
100
+ # [1.0000e+00, 8.3131e-12, 5.2173e-12]
101
  ```
102
 
103
+ ### Evaluation
104
+ [Generalised Contrastiove Learning](https://github.com/marqo-ai/GCL) (GCL) is used for the evaluation. The following code can also be found in `scripts`.
105
+
106
  ```
107
  git clone https://github.com/marqo-ai/GCL
108
  ```
109
  Install the packages required by GCL.
110
+
111
+ **1. GoogleShopping-Text2Image Retrieval.**
112
  ```
113
  cd ./GCL
114
  MODEL=hf-hub:Marqo/marqo-ecommerce-B
115
+ outdir=/MarqoModels/GE/marqo-ecommerce-B/gs-title2image
116
  hfdataset=Marqo/google-shopping-general-eval
117
  python evals/eval_hf_datasets_v1.py \
118
  --model_name $MODEL \
 
131
  --context-length "[[64], [0]]"
132
  ```
133
 
134
+ **2. GoogleShopping-Category2Image Retrieval.**
135
+ ```
136
+ cd ./GCL
137
+ MODEL=hf-hub:Marqo/marqo-ecommerce-B
138
+ outdir=/MarqoModels/GE/marqo-ecommerce-B/gs-cat2image
139
+ hfdataset=Marqo/google-shopping-general-eval
140
+ python evals/eval_hf_datasets_v1.py \
141
+ --model_name $MODEL \
142
+ --hf-dataset $hfdataset \
143
+ --output-dir $outdir \
144
+ --batch-size 1024 \
145
+ --num_workers 8 \
146
+ --left-key "['query']" \
147
+ --right-key "['image']" \
148
+ --img-or-txt "[['txt'], ['img']]" \
149
+ --left-weight "[1]" \
150
+ --right-weight "[1]" \
151
+ --run-queries-cpu \
152
+ --top-q 4000 \
153
+ --doc-id-key item_ID \
154
+ --context-length "[[64], [0]]"
155
+ ```
156
+
157
+ **3. AmazonProducts-Category2Image Retrieval.**
158
+ ```
159
+ cd ./GCL
160
+ MODEL=hf-hub:Marqo/marqo-ecommerce-B
161
+ outdir=/MarqoModels/GE/marqo-ecommerce-B/ap-title2image
162
+ hfdataset=Marqo/amazon-products-eval
163
+ python evals/eval_hf_datasets_v1.py \
164
+ --model_name $MODEL \
165
+ --hf-dataset $hfdataset \
166
+ --output-dir $outdir \
167
+ --batch-size 1024 \
168
+ --num_workers 8 \
169
+ --left-key "['title']" \
170
+ --right-key "['image']" \
171
+ --img-or-txt "[['txt'], ['img']]" \
172
+ --left-weight "[1]" \
173
+ --right-weight "[1]" \
174
+ --run-queries-cpu \
175
+ --top-q 4000 \
176
+ --doc-id-key item_ID \
177
+ --context-length "[[64], [0]]"
178
+ ```
179
 
180
  ## Detailed Performance
181
+ The benchmarks are separated into 'Marqo-Ecommerce-Hard' and '100k-Marqo-Ecommerce-Easy'. The "easy" dataset is about 10-30 times smaller, and designed to accommodate rate-limited models, specifically Cohere-Embeddings-v3 and GCP-Vertex. The "hard" dataset represents the true challenge, since it contains four million ecommerce product listings, which pushes these models to their limits in a real-world, ecommerce scenario.
182
+
183
+ Within both these scenarios, the models were benchmarked against three different tasks:
184
 
185
+ * Google Shopping Text-to-Image
186
+ * Google Shopping Category-to-Image
187
+ * Amazon Products Text-to-Image
 
 
 
 
188
 
189
+ ### Marqo-Ecommerce-Hard
190
+ Marqo-Ecommerce-Hard looks into the comprehensive evaluation conducted using the full 4 million dataset, highlighting the robust performance of our models in a real-world context.
191
+
192
+ **GoogleShopping-Text2Image Retrieval.**
193
+
194
+ | **Embedding Model** | **mAP** | **R@10** | **MRR** | **nDCG@10** |
195
+ |-------------------------|------|-------|------|---------|
196
+ | **Marqo-Ecommerce-L** | **0.682**| **0.878** | **0.683**| **0.726** |
197
+ | Marqo-Ecommerce-B | 0.623| 0.832 | 0.624| 0.668 |
198
+ | ViT-SO400M-14-SigLip | 0.573| 0.763 | 0.574| 0.613 |
199
+ | ViT-L-16-SigLip | 0.540| 0.722 | 0.540| 0.577 |
200
+ | ViT-B-16-SigLip | 0.476| 0.660 | 0.477| 0.513 |
201
+ | Amazon-Titan-MultiModal | 0.475| 0.648 | 0.475| 0.509 |
202
+ | Jina-V1-CLIP | 0.285| 0.402 | 0.285| 0.306 |
203
 
204
  **GoogleShopping-Category2Image Retrieval.**
205
 
206
+ | **Embedding Model** | **mAP** | **P@10** | **MRR** | **nDCG@10** |
207
+ |-----------------------------|---------|----------|---------|-------------|
208
+ | **Marqo-Ecommerce-L** | **0.463** | **0.652** | **0.822** | **0.666** |
209
+ | Marqo-Ecommerce-B | 0.423 | 0.629 | 0.810 | 0.644 |
210
+ | ViT-SO400M-14-SigLip | 0.352 | 0.516 | 0.707 | 0.529 |
211
+ | ViT-L-16-SigLip | 0.324 | 0.497 | 0.687 | 0.509 |
212
+ | ViT-B-16-SigLip | 0.277 | 0.458 | 0.660 | 0.473 |
213
+ | Amazon-Titan-MultiModal | 0.246 | 0.429 | 0.642 | 0.446 |
214
+ | Jina-V1-CLIP | 0.123 | 0.275 | 0.504 | 0.294 |
215
 
216
  **AmazonProducts-Text2Image Retrieval.**
217
 
218
+ | **Embedding Model** | **mAP** | **R@10** | **MRR** | **nDCG@10** |
219
+ |-----------------------------|---------|----------|---------|-------------|
220
+ | **Marqo-Ecommerce-L** | **0.658** | **0.854** | **0.663** | **0.703** |
221
+ | Marqo-Ecommerce-B | 0.592 | 0.795 | 0.597 | 0.637 |
222
+ | ViT-SO400M-14-SigLip | 0.560 | 0.742 | 0.564 | 0.599 |
223
+ | ViT-L-16-SigLip | 0.544 | 0.715 | 0.548 | 0.580 |
224
+ | ViT-B-16-SigLip | 0.480 | 0.650 | 0.484 | 0.515 |
225
+ | Amazon-Titan-MultiModal | 0.456 | 0.627 | 0.457 | 0.491 |
226
+ | Jina-V1-CLIP | 0.265 | 0.378 | 0.266 | 0.285 |
227
 
228
+ ### 100k-Marqo-Ecommerce-Easy
229
+ This dataset is about 10-30 times smaller than the Marqo-Ecommerce-Hard, and designed to accommodate rate-limited models, specifically Cohere-Embeddings-v3 and GCP-Vertex.
230
 
231
+ **GoogleShopping-Text2Image Retrieval.**
232
 
233
+ | **Embedding Model** | **mAP** | **R@10** | **MRR** | **nDCG@10** |
234
+ |-----------------------------|---------|----------|---------|-------------|
235
+ | **Marqo-Ecommerce-L** | **0.879** | **0.971** | **0.879** | **0.901** |
236
+ | Marqo-Ecommerce-B | 0.842 | 0.961 | 0.842 | 0.871 |
237
+ | ViT-SO400M-14-SigLip | 0.792 | 0.935 | 0.792 | 0.825 |
238
+ | GCP-Vertex | 0.740 | 0.910 | 0.740 | 0.779 |
239
+ | ViT-L-16-SigLip | 0.754 | 0.907 | 0.754 | 0.789 |
240
+ | ViT-B-16-SigLip | 0.701 | 0.870 | 0.701 | 0.739 |
241
+ | Amazon-Titan-MultiModal | 0.694 | 0.868 | 0.693 | 0.733 |
242
+ | Jina-V1-CLIP | 0.480 | 0.638 | 0.480 | 0.511 |
243
+ | Cohere-embedding-v3 | 0.358 | 0.515 | 0.358 | 0.389 |
244
 
245
+ **GoogleShopping-Category2Image Retrieval.**
246
+
247
+ | **Embedding Model** | **mAP** | **P@10** | **MRR** | **nDCG@10** |
248
+ |-----------------------------|---------|----------|---------|-------------|
249
+ | **Marqo-Ecommerce-L** | **0.515** | **0.358** | **0.764** | **0.590** |
250
+ | Marqo-Ecommerce-B | 0.479 | 0.336 | 0.744 | 0.558 |
251
+ | ViT-SO400M-14-SigLip | 0.423 | 0.302 | 0.644 | 0.487 |
252
+ | GCP-Vertex | 0.417 | 0.298 | 0.636 | 0.481 |
253
+ | ViT-L-16-SigLip | 0.392 | 0.281 | 0.627 | 0.458 |
254
+ | ViT-B-16-SigLip | 0.347 | 0.252 | 0.594 | 0.414 |
255
+ | Amazon-Titan-MultiModal | 0.308 | 0.231 | 0.558 | 0.377 |
256
+ | Jina-V1-CLIP | 0.175 | 0.122 | 0.369 | 0.229 |
257
+ | Cohere-embedding-v3 | 0.136 | 0.110 | 0.315 | 0.178 |
258
+
259
+ **AmazonProducts-Text2Image Retrieval.**
260
+
261
+ | **Embedding Model** | **mAP** | **R@10** | **MRR** | **nDCG@10** |
262
+ |-----------------------------|---------|----------|---------|-------------|
263
+ | **Marqo-Ecommerce-L** | **0.92** | **0.978** | **0.928** | **0.940** |
264
+ | Marqo-Ecommerce-B | 0.897 | 0.967 | 0.897 | 0.914 |
265
+ | ViT-SO400M-14-SigLip | 0.860 | 0.954 | 0.860 | 0.882 |
266
+ | ViT-L-16-SigLip | 0.842 | 0.940 | 0.842 | 0.865 |
267
+ | GCP-Vertex | 0.808 | 0.933 | 0.808 | 0.837 |
268
+ | ViT-B-16-SigLip | 0.797 | 0.917 | 0.797 | 0.825 |
269
+ | Amazon-Titan-MultiModal | 0.762 | 0.889 | 0.763 | 0.791 |
270
+ | Jina-V1-CLIP | 0.530 | 0.699 | 0.530 | 0.565 |
271
+ | Cohere-embedding-v3 | 0.433 | 0.597 | 0.433 | 0.465 |
272
+
273
+ ## Citation
274
+ ```
275
+ @software{zhu2024marqoecommembed_2024,
276
+ author = {Tianyu Zhu and and Jesse Clark},
277
+ month = oct,
278
+ title = {{Marqo Ecommerce Embeddings - Foundation Model for Product Embeddings}},
279
+ url = {https://github.com/marqo-ai/marqo-ecommerce-embeddings/},
280
+ version = {1.0.0},
281
+ year = {2024}
282
+ }
283
+ ```