Update README.md
Browse files
README.md
CHANGED
|
@@ -12,11 +12,15 @@ tags:
|
|
| 12 |
|
| 13 |
# constructai/taylor-m1-38m-128
|
| 14 |
|
|
|
|
|
|
|
| 15 |
This model is a fine-tuned version of a custom BERT-like encoder (hidden_size=384, 6 layers) trained on MS MARCO passage ranking dataset.
|
| 16 |
It uses **triplet hard negatives** from [`sentence-transformers/msmarco-msmarco-distilbert-base-v3`](https://huggingface.co/datasets/sentence-transformers/msmarco-msmarco-distilbert-base-v3) (Apache-2.0 license). The base MS MARCO data is subject to [Microsoft Research License](https://github.com/microsoft/MSMARCO-Passage-Ranking/blob/master/LICENSE).
|
| 17 |
|
| 18 |
The model produces **384-dimensional** embeddings (CLS token) optimized for cosine similarity.
|
| 19 |
|
|
|
|
|
|
|
| 20 |
# Training details
|
| 21 |
|
| 22 |
* ~38.3M parameters
|
|
@@ -31,6 +35,8 @@ The model produces **384-dimensional** embeddings (CLS token) optimized for cosi
|
|
| 31 |
|
| 32 |
* Data: ~500k triplets from MS MARCO
|
| 33 |
|
|
|
|
|
|
|
| 34 |
# Evaluation
|
| 35 |
|
| 36 |
On a small test set, the model achieves:
|
|
@@ -41,6 +47,8 @@ On a small test set, the model achieves:
|
|
| 41 |
|
| 42 |
* Margin: 0.44
|
| 43 |
|
|
|
|
|
|
|
| 44 |
# Usage
|
| 45 |
|
| 46 |
**Option 1: Via custom Python package (recommended)**
|
|
@@ -91,6 +99,8 @@ print("Similarity matrix:")
|
|
| 91 |
print(similarities)
|
| 92 |
```
|
| 93 |
|
|
|
|
|
|
|
| 94 |
**Requirements:**
|
| 95 |
|
| 96 |
* Python ≥ 3.9
|
|
@@ -99,6 +109,8 @@ print(similarities)
|
|
| 99 |
* huggingface_hub ≥ 0.20.0
|
| 100 |
* numpy
|
| 101 |
|
|
|
|
|
|
|
| 102 |
# License
|
| 103 |
|
| 104 |
This model is released under the **Apache 2.0** License.
|
|
|
|
| 12 |
|
| 13 |
# constructai/taylor-m1-38m-128
|
| 14 |
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
This model is a fine-tuned version of a custom BERT-like encoder (hidden_size=384, 6 layers) trained on MS MARCO passage ranking dataset.
|
| 18 |
It uses **triplet hard negatives** from [`sentence-transformers/msmarco-msmarco-distilbert-base-v3`](https://huggingface.co/datasets/sentence-transformers/msmarco-msmarco-distilbert-base-v3) (Apache-2.0 license). The base MS MARCO data is subject to [Microsoft Research License](https://github.com/microsoft/MSMARCO-Passage-Ranking/blob/master/LICENSE).
|
| 19 |
|
| 20 |
The model produces **384-dimensional** embeddings (CLS token) optimized for cosine similarity.
|
| 21 |
|
| 22 |
+
---
|
| 23 |
+
|
| 24 |
# Training details
|
| 25 |
|
| 26 |
* ~38.3M parameters
|
|
|
|
| 35 |
|
| 36 |
* Data: ~500k triplets from MS MARCO
|
| 37 |
|
| 38 |
+
---
|
| 39 |
+
|
| 40 |
# Evaluation
|
| 41 |
|
| 42 |
On a small test set, the model achieves:
|
|
|
|
| 47 |
|
| 48 |
* Margin: 0.44
|
| 49 |
|
| 50 |
+
---
|
| 51 |
+
|
| 52 |
# Usage
|
| 53 |
|
| 54 |
**Option 1: Via custom Python package (recommended)**
|
|
|
|
| 99 |
print(similarities)
|
| 100 |
```
|
| 101 |
|
| 102 |
+
---
|
| 103 |
+
|
| 104 |
**Requirements:**
|
| 105 |
|
| 106 |
* Python ≥ 3.9
|
|
|
|
| 109 |
* huggingface_hub ≥ 0.20.0
|
| 110 |
* numpy
|
| 111 |
|
| 112 |
+
---
|
| 113 |
+
|
| 114 |
# License
|
| 115 |
|
| 116 |
This model is released under the **Apache 2.0** License.
|