Commit
·
1f0fdea
1
Parent(s):
6a21313
update readme
Browse files- README.md +3 -4
- config.json +0 -1
README.md
CHANGED
@@ -20,13 +20,11 @@ pipeline_tag: visual-document-retrieval
|
|
20 |
|
21 |
# llama-nemoretriever-colembed-1b-v1
|
22 |
|
23 |
-
# llama-nemoretriever-colembed-1b-v1
|
24 |
-
|
25 |
## Description
|
26 |
|
27 |
The **nvidia/llama-nemoretriever-colembed-1b-v1** is a late interaction embedding model fine-tuned for query-document retrieval. Users can input `queries`, which are text, or `documents` which are page images, to the model. The model outputs ColBERT-style multi-vector numerical representations for input queries and documents. It is the smaller version of [llama-nemoretriever-colembed-3b-v1](https://huggingface.co/nvidia/llama-nemoretriever-colembed-3b-v1), which achieved 1st place on ViDoRe V1 (nDCG@5), ViDoRe V2 (nDCG@5) and MTEB VisualDocumentRetrieval (Rank Borda) (as of 27th June, 2025). **nvidia/llama-nemoretriever-colembed-1b-v1** achieves 2nd place on the benchmarks.
|
28 |
|
29 |
-
This model is for non-commercial/research use only.
|
30 |
|
31 |
### License/Terms of Use
|
32 |
Governing Terms: [NVIDIA License](https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE)
|
@@ -114,11 +112,12 @@ from transformers import AutoModel
|
|
114 |
|
115 |
# Load Model
|
116 |
model = AutoModel.from_pretrained(
|
117 |
-
'nvidia/llama-
|
118 |
device_map='cuda',
|
119 |
trust_remote_code=True,
|
120 |
torch_dtype=torch.bfloat16,
|
121 |
attn_implementation="flash_attention_2",
|
|
|
122 |
).eval()
|
123 |
|
124 |
# Queries
|
|
|
20 |
|
21 |
# llama-nemoretriever-colembed-1b-v1
|
22 |
|
|
|
|
|
23 |
## Description
|
24 |
|
25 |
The **nvidia/llama-nemoretriever-colembed-1b-v1** is a late interaction embedding model fine-tuned for query-document retrieval. Users can input `queries`, which are text, or `documents` which are page images, to the model. The model outputs ColBERT-style multi-vector numerical representations for input queries and documents. It is the smaller version of [llama-nemoretriever-colembed-3b-v1](https://huggingface.co/nvidia/llama-nemoretriever-colembed-3b-v1), which achieved 1st place on ViDoRe V1 (nDCG@5), ViDoRe V2 (nDCG@5) and MTEB VisualDocumentRetrieval (Rank Borda) (as of 27th June, 2025). **nvidia/llama-nemoretriever-colembed-1b-v1** achieves 2nd place on the benchmarks.
|
26 |
|
27 |
+
This model is for non-commercial/research use only.
|
28 |
|
29 |
### License/Terms of Use
|
30 |
Governing Terms: [NVIDIA License](https://huggingface.co/nvidia/llama-nemoretriever-colembed-1b-v1/blob/main/LICENSE)
|
|
|
112 |
|
113 |
# Load Model
|
114 |
model = AutoModel.from_pretrained(
|
115 |
+
'nvidia/llama-nemoretriever-colembed-1b-v1',
|
116 |
device_map='cuda',
|
117 |
trust_remote_code=True,
|
118 |
torch_dtype=torch.bfloat16,
|
119 |
attn_implementation="flash_attention_2",
|
120 |
+
revision='6a21313a150a903bc522dc0d15ed47784a0d4c8d'
|
121 |
).eval()
|
122 |
|
123 |
# Queries
|
config.json
CHANGED
@@ -1,6 +1,5 @@
|
|
1 |
{
|
2 |
"_commit_hash": null,
|
3 |
-
"_name_or_path": "./model_1b_test/",
|
4 |
"architectures": [
|
5 |
"llama_NemoRetrieverColEmbed"
|
6 |
],
|
|
|
1 |
{
|
2 |
"_commit_hash": null,
|
|
|
3 |
"architectures": [
|
4 |
"llama_NemoRetrieverColEmbed"
|
5 |
],
|