Improve model card: add abstract, project page, update paper link, refine tags

Hello team,

This PR addresses several improvements to the model card for `Llama3.1-8B-Middo-Alpaca`:

* **Updated Paper Link**: The link to the paper has been updated to point to the Hugging Face Papers page (`https://huggingface.co/papers/2508.21589`) for better integration within the Hub.
* **Added Project Page**: A dedicated link to the [Middo Hugging Face Collection](https://huggingface.co/collections/Word2Li) has been added to provide users with a central location for related artifacts.
* **Included Abstract**: The paper's abstract has been added to the model card, offering a comprehensive overview of the model's approach and results.
* **Refined Tags**: Added `llm-finetuning` and `data-optimization` tags to improve discoverability and accurately reflect the model's core contributions. The less descriptive `full` tag has been removed.
* **Removed File Information**: The "File information" section, which is internal context, has been removed from the public-facing model card.

These changes enhance the model card's informativeness and adherence to best practices for documentation on the Hugging Face Hub.

README.md

@@ -1,106 +1,111 @@
 ---
-library_name: transformers
-license: llama3.1
 base_model: meta-llama/Llama-3.1-8B
-language: en
 datasets:
 - Word2Li/MiddOptimized
+language: en
+library_name: transformers
+license: llama3.1
+metrics:
+- accuracy
+pipeline_tag: text-generation
 tags:
 - llama-factory
-- full
-pipeline_tag: text-generation
+- llm-finetuning
+- data-optimization
 model-index:
 - name: Llama3.1-8B-Middo-Alpaca
   results:
-    - task:
-        type: text-generation
-      dataset:
-        name: MMLU
-        type: MMLU
-      metrics:
-        - name: weighted accuracy
-          type: weighted accuracy
-          value: 51.32
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: IFEval
-        type: IFEval
-      metrics:
-        - name: overall accuracy
-          type: overall accuracy
-          value: 43.20
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: GSM8K
-        type: GSM8K
-      metrics:
-        - name: accuracy
-          type: accuracy
-          value: 51.18
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: MATH
-        type: MATH
-      metrics:
-        - name: accuracy
-          type: accuracy
-          value: 12.92
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: HumanEval
-        type: HumanEval
-      metrics:
-        - name: humaneval_pass@1
-          type: humaneval_pass@1
-          value: 39.63
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: MBPP
-        type: MBPP
-      metrics:
-        - name: score
-          type: score
-          value: 41.80
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: Hellaswag
-        type: Hellaswag
-      metrics:
-        - name: accuracy
-          type: accuracy
-          value: 58.78
-          verified: true
-    - task:
-        type: text-generation
-      dataset:
-        name: GPQA
-        type: GPQA
-      metrics:
-        - name: accuracy
-          type: accuracy
-          value: 16.67
-          verified: true
-metrics:
-- accuracy
+  - task:
+      type: text-generation
+    dataset:
+      name: MMLU
+      type: MMLU
+    metrics:
+    - type: weighted accuracy
+      value: 51.32
+      name: weighted accuracy
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: IFEval
+      type: IFEval
+    metrics:
+    - type: overall accuracy
+      value: 43.2
+      name: overall accuracy
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: GSM8K
+      type: GSM8K
+    metrics:
+    - type: accuracy
+      value: 51.18
+      name: accuracy
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: MATH
+      type: MATH
+    metrics:
+    - type: accuracy
+      value: 12.92
+      name: accuracy
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: HumanEval
+      type: HumanEval
+    metrics:
+    - type: humaneval_pass@1
+      value: 39.63
+      name: humaneval_pass@1
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: MBPP
+      type: MBPP
+    metrics:
+    - type: score
+      value: 41.8
+      name: score
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: Hellaswag
+      type: Hellaswag
+    metrics:
+    - type: accuracy
+      value: 58.78
+      name: accuracy
+      verified: true
+  - task:
+      type: text-generation
+    dataset:
+      name: GPQA
+      type: GPQA
+    metrics:
+    - type: accuracy
+      value: 16.67
+      name: accuracy
+      verified: true
 ---
 
 # Llama3.1-8B-Middo-Alpaca
 
-Paper: [Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning](https://arxiv.org/abs/2508.21589)
-
+Paper: [Middo: Model-Informed Dynamic Data Optimization for Enhanced LLM Fine-Tuning via Closed-Loop Learning](https://huggingface.co/papers/2508.21589)
 Code: https://github.com/Word2VecT/Middo
+Project page: [Middo Hugging Face Collection](https://huggingface.co/collections/Word2Li)
+
+## Abstract
+
+Supervised Fine-Tuning (SFT) Large Language Models (LLM) fundamentally rely on high-quality training data. While data selection and data synthesis are two common strategies to improve data quality, existing approaches often face limitations in static dataset curation that fail to adapt to evolving model capabilities. In this paper, we introduce Middo, a self-evolving Model-informed dynamic data optimization framework that uses model-aware data selection and context-preserving data refinement. Unlike conventional one-off filtering/synthesis methods, our framework establishes a closed-loop optimization system: (1) A self-referential diagnostic module proactively identifies suboptimal samples through tri-axial model signals - loss patterns (complexity), embedding cluster dynamics (diversity), and self-alignment scores (quality); (2) An adaptive optimization engine then transforms suboptimal samples into pedagogically valuable training points while preserving semantic integrity; (3) This optimization process continuously evolves with model capability through dynamic learning principles. Experiments on multiple benchmarks demonstrate that our Middo consistently enhances the quality of seed data and boosts LLM's performance with improving accuracy by 7.15% on average while maintaining the original dataset scale. This work establishes a new paradigm for sustainable LLM training through dynamic human-AI co-evolution of data and models. Our datasets, models, and code are coming soon. Our datasets, models, and code are publicly available at this https URL
 
 ## Model description