kkknight
/

MiniOneRec

@@ -31,80 +31,57 @@ These checkpoints accompany the paper:
 ---
 ## 🗺️ Table of Contents
-1. Model Summary
-2. Intended Uses & Limitations
-3. Quick Start
-4. Training & Evaluation Details
-5. Citation
-6. Acknowledgements
 ---
-## 1️⃣ Model Summary
-MiniOneRec rewrites every catalogue item into a discrete **SID token**:
-1. **Text Encoder** (frozen PLM) →
-2. **3-level Residual Quantisation (RQ-VAE / RQ-KMeans)** → SID.
-User history ≙ SID sequence.
-Training pipeline:
-|
- Stage
-|
- Objective
-|
- Notes
-|
-|-------|-----------|-------|
-| **SFT** | Next-SID prediction + language alignment | inherits world knowledge while grounding in item space |
-| **RL (GRPO)** | KL-regularised policy optimisation | constrained beam search over the closed SID set |
-### Released checkpoints (examples)
-|
- Checkpoint
-|
- Base LLM
-|
-#
-Params
-|
- Precision
-|
- Stage
-|
-|-------------------------------------|---------------------|---------|-----------|-----------|
-| `MiniOneRec-SFT-industrial`         | Qwen-7B             | 7 B     | bf16      | SFT       |
-| `MiniOneRec-RL-industrial`          | Qwen-7B             | 7 B     | bf16      | SFT+RL    |
-*(Replace with the exact repo names you upload.)*
 ---
-## 2️⃣ Intended Uses & Limitations
-### ✅ Intended
-* Next-item prediction in e-commerce / content platforms.
-* Research on generative recommendation and RL-from-human-feedback variants.
-### ❌ Out-of-Scope
-* Safety-critical deployments without exhaustive evaluation.
-* Domains whose item catalogue is not covered by the released SID vocabulary.
-* Generation of content that violates the Apache-2.0 license or local regulations.
-### ⚖️ Ethical Considerations
-The model may inherit bias from the training corpus (user behaviour, language model).
-Please **audit for fairness, privacy and potential leakage** before production use.
 ---
-## 3️⃣ Citation
 ```
 @misc{MiniOneRec,
@@ -123,17 +100,3 @@ Please **audit for fairness, privacy and potential leakage** before production u
   year    = {2025}
 }
 ```
-## 4️⃣ Acknowledgements
-This repository reuses or adapts portions of code from the following open-source projects. We gratefully acknowledge their authors and contributors:
-- [ReRe](https://github.com/sober-clever/ReRe)
-- [LC-Rec](https://github.com/zhengbw0324/LC-Rec)
-## 5️⃣ Institutions  <!-- omit in toc -->
-This project is developed by the following institutions:
-- <img src="assets/lds.png" width="28px"> [LDS](https://data-science.ustc.edu.cn/_upload/tpl/15/04/5380/template5380/index.html)
-- <img src="assets/alphalab.jpg" width="28px"> [AlphaLab](https://alphalab-ustc.github.io/index.html)
-- <img src="assets/next.jpg" width="28px"> [NExT](https://www.nextcenter.org/)

 ---
 ## 🗺️ Table of Contents
+1. Key Techniques
+2. Evaluation
+3. Acknowledgements
+4. Institutions
+5. Citation
 ---
+## 1️⃣ Key Techniques
+<div align="center">
+<img src="./assets/minionerec_framework.png" width=100% ></img>
+</div>
+- **SID Construction: MiniOneRec begins by transforming every product into a compact, semantically meaningful token.** It concatenates an item’s title and description, feeds this sentence through a frozen text encoder, and then quantises the resulting embedding with a three-level RQ-VAE.
+- **SFT: With all items rewritten as SIDs, the model is first trained in a supervised fashion.** It views the chronologically ordered user history as a token sequence and learns, via next-token prediction, to generate the SID of the next product the user is likely to consume. Crucially, this stage is co-trained with a set of language-alignment objectives that map back and forth between natural language and SID space, allowing the recommender to inherit the world knowledge embedded in large language models while grounding that knowledge in discrete item codes.
+- **Recommendation-Oriented RL: After SFT, MiniOneRec is further polished with a recommendation-oriented RL phase based on GRPO.** Multiple candidate recommendations are generated for each prompt, their rewards are normalised within the group to stabilise gradients, and a KL penalty keeps the updated policy close to its reference. Because the action space is a closed list of item SIDs, the system switches to constrained beam search, which guarantees that every beam is unique and valid, greatly improving sampling efficiency and diversity. The reward signal itself blends a binary correctness term with a rank-aware component that penalises high-probability yet incorrect items more heavily, and can be augmented with collaborative-filtering scores. Together, this pipeline enables MiniOneRec to couple dense linguistic knowledge, achieving a high-performance, lightweight generative recommendation system.
 ---
+## 2️⃣ Evaluation
+<div align="center">
+<img src="./assets/minionerec_main_result.png" width=100% ></img>
+</div>
+---
+## 3️⃣ Acknowledgements
+This repository reuses or adapts portions of code from the following open-source projects. We gratefully acknowledge their authors and contributors:
+- [ReRe](https://github.com/sober-clever/ReRe)
+- [LC-Rec](https://github.com/zhengbw0324/LC-Rec)
+---
+## 4️⃣ Institutions  <!-- omit in toc -->
+This project is developed by the following institutions:
+- <img src="assets/lds.png" width="28px"> [LDS](https://data-science.ustc.edu.cn/_upload/tpl/15/04/5380/template5380/index.html)
+- <img src="assets/alphalab.jpg" width="28px"> [AlphaLab](https://alphalab-ustc.github.io/index.html)
+- <img src="assets/next.jpg" width="28px"> [NExT](https://www.nextcenter.org/)
 ---
+## 5️⃣ Citation
+If you find our code/paper/model helpful, please consider citing our papers 📝 and staring us ⭐️！
 ```
 @misc{MiniOneRec,
   year    = {2025}
 }
 ```