File size: 1,648 Bytes
4fc5b15
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
---
license: other
language:
- en
tags:
- recommendation
- information-retrieval
- feature-extraction
base_model: Qwen/Qwen3-Embedding-0.6B
---

# Amazon recommendation tools

Frozen catalog artifacts and behavior-trained recommendation tools used by RecoAtlas. This is an
inference-artifact repository, not a standalone Transformers model and not a record of training
runs.

## Categories

- `Musical_Instruments`
- `Electronics`
- `Video_Games`

## Uniform layout

Each category contains exactly the files needed by the released tools:

```text
<category>/
  data/product_texts.json
  data/subcategories.json
  embeddings/item_embeddings_qwen.pt
  embeddings/complementarity_embeddings_qwen.pt
  models/retrieval_model_qwen.pt
  manifest.json
```

- `retrieval_model_qwen.pt` is the selected query-side LoRA retrieval checkpoint.
- `item_embeddings_qwen.pt` contains the frozen catalog item vectors used by semantic search and
  near-duplicate pruning.
- `complementarity_embeddings_qwen.pt` contains the selected catalog-aligned `anchor_embs` and
  `complement_embs` used by the utility-aligned complementarity tool. It is the complementarity
  inference artifact; historical epoch checkpoints and training runs are intentionally omitted.
- `manifest.json` records exact sizes and SHA-256 checksums.

The underlying query/reference benchmark is [iaouali/amazon-benchmark](https://huggingface.co/datasets/iaouali/amazon-benchmark).
The base query encoder is [Qwen/Qwen3-Embedding-0.6B](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B). Catalog content,
source datasets, and the base model remain subject to their respective terms. See `NOTICE`.