Rename model to ShenZhen (神珍)

README.md

@@ -25,13 +25,13 @@ extra_gated_description: >-
 
 <div align="center">
 
-[🤗 Model](https://huggingface.co/sais-org/Polaris_Pro) &nbsp;•&nbsp; [💻 GitHub](https://github.com/Shanghai-Academy-of-AI-For-Science/Polaris-Pro) &nbsp;•&nbsp; [📜 Technical Report (coming soon)](#) &nbsp;•&nbsp; [⚖️ License: Apache-2.0 + SAM License](https://github.com/Shanghai-Academy-of-AI-For-Science/Polaris-Pro/blob/main/LICENSE)
+[🤗 Model](https://huggingface.co/sais-org/ShenZhen) &nbsp;•&nbsp; [💻 GitHub](https://github.com/Shanghai-Academy-of-AI-For-Science/ShenZhen) &nbsp;•&nbsp; [📜 Technical Report (coming soon)](#) &nbsp;•&nbsp; [⚖️ License: Apache-2.0 + SAM License](https://github.com/Shanghai-Academy-of-AI-For-Science/ShenZhen/blob/main/LICENSE)
 
 </div>
 
-# Polaris-Pro
+# 神珍 (ShenZhen)
 
-**Polaris-Pro is a unified scientific multimodal foundation model** that
+**神珍 is a unified scientific multimodal foundation model** that
 supports scientific **understanding and generation** across Earth science,
 proteins, RNA, DNA, and small molecules within a single **8B** model. Native
 scientific encoders/decoders wrap a shared **Qwen3-VL-8B-Instruct** backbone, so
@@ -67,13 +67,13 @@ space — natural language in and out, no per-task fine-tuning.
 
 ## Benchmarks
 
-**Polaris-Pro** (**8B**) vs **Biology-Instructions** (Llama-3.1-**8B**, text-token,
+**神珍** (**8B**) vs **Biology-Instructions** (Llama-3.1-**8B**, text-token,
 no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
 **Bold** = best; <u>underline</u> = second-best.
 
 ### Biological sequence understanding
 
-| Task | Metric | Polaris-Pro (8B) | Biology-Instructions (8B) | Intern-S1-Pro (~1T) |
+| Task | Metric | 神珍 (8B) | Biology-Instructions (8B) | Intern-S1-Pro (~1T) |
 |:-----|:------:|:----------------:|:-------------------------:|:-------------------:|
 | DNA · Epigenetic marks (EMP) | MCC | **71.99** | 3.64 | <u>14.02</u> |
 | DNA · Promoter det. 300bp (PD300) | MCC | **91.17** | 58.18 | <u>82.65</u> |
@@ -91,11 +91,11 @@ no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
 | Cross-modal · AAN (antibody–antigen) | MCC | <u>42.96</u> | 1.06 | **44.76** |
 | Cross-modal · EPI (enhancer–promoter) | MCC | <u>-0.03</u> | **3.37** | -1.30 |
 
-<sub>Aggregate over 20 biological-understanding benchmarks: Polaris-Pro matches or beats the ~1T Intern-S1-Pro on 10/20 and the same-scale 8B text-token baseline on 16/20.</sub>
+<sub>Aggregate over 20 biological-understanding benchmarks: 神珍 matches or beats the ~1T Intern-S1-Pro on 10/20 and the same-scale 8B text-token baseline on 16/20.</sub>
 
 ### Molecule understanding (SMolInstruct)
 
-| Task | Metric | Polaris-Pro (8B) | LlaSMol |
+| Task | Metric | 神珍 (8B) | LlaSMol |
 |:-----|:------:|:----------------:|:-------:|
 | BBBP | Acc | **96.95** | 74.60 |
 | HIV | Acc | **97.00** | 96.70 |
@@ -106,20 +106,20 @@ no scientific encoders) and **Intern-S1-Pro** (**~1T** MoE scientific model).
 
 ### Earth-science forecasting — vs ECMWF HRES (day-10, global ERA5 0.25°)
 
-| Variable | Metric | Polaris-Pro (8B) | ECMWF HRES (NWP) |
+| Variable | Metric | 神珍 (8B) | ECMWF HRES (NWP) |
 |:---------|:------:|:----------------:|:----------------:|
 | Z500 | RMSE ↓ | **≈740** | ≈810 |
 | T2M | RMSE ↓ (K) | **≈2.65** | ≈2.90 |
 | MSL | RMSE ↓ (Pa) | **≈680** | ≈745 |
 
-<sub>Polaris-Pro tracks or beats the operational physics-based HRES system, with the advantage growing at longer lead times.</sub>
+<sub>神珍 tracks or beats the operational physics-based HRES system, with the advantage growing at longer lead times.</sub>
 
 ### Medical-image segmentation
 
 Mean Dice (%) on the BiomedParse test splits, 102,855 image–prompt pairs across
 nine imaging modalities, versus six modality-native segmentation specialists.
 
-| Modality | # Samples | Polaris-Pro | BiomedParse | MedSAM | SAM | SAM3 | DINO+MedSAM | DINO+SAM |
+| Modality | # Samples | 神珍 | BiomedParse | MedSAM | SAM | SAM3 | DINO+MedSAM | DINO+SAM |
 |:---------|----------:|:-----------:|:-----------:|:------:|:---:|:----:|:-----------:|:--------:|
 | **All**    | 102,855 | **91.20** | <u>90.73</u> | 83.55 | 71.29 | 35.40 | 15.37 | 15.10 |
 | CT         |  45,306 | **93.36** | <u>92.25</u> | 83.87 | 74.10 | 28.93 |  9.59 | 10.34 |
@@ -139,9 +139,9 @@ nine imaging modalities, versus six modality-native segmentation specialists.
 Runs via the accompanying code repository (custom multimodal architecture).
 
 ```bash
-git clone https://github.com/Shanghai-Academy-of-AI-For-Science/Polaris-Pro && cd Polaris-Pro
+git clone https://github.com/Shanghai-Academy-of-AI-For-Science/ShenZhen && cd ShenZhen
 pip install -r requirements.txt          # Python 3.10; transformers==5.0.0
-hf download sais-org/Polaris_Pro --local-dir ./model
+hf download sais-org/ShenZhen --local-dir ./model
 
 export PYTHONPATH=$PWD/code
 python code/inference.py --model_path model --greedy --max_new_tokens 64 \
@@ -160,7 +160,7 @@ Each task has a specific `--system` prompt that fixes the output format; see
 
 ## License
 
-**Composite license.** Polaris-Pro's own components — the code, and all weights
+**Composite license.** 神珍's own components — the code, and all weights
 except the SAM 3 branch — are **Apache-2.0**, built on Qwen3-VL (Apache-2.0) and
 including merged ESM-2 (MIT) and Polaris/Suiren-derived encoders.
 
@@ -173,10 +173,10 @@ See `THIRD_PARTY_LICENSES.md` / `NOTICE` for the full third-party breakdown.
 ## Citation
 
 ```bibtex
-@misc{polarispro2026,
-  title  = {Polaris-Pro: A Unified Scientific Multimodal Foundation Model},
+@misc{shenzhen2026,
+  title  = {ShenZhen (神珍): A Unified Scientific Multimodal Foundation Model},
   author = {Hesen Chen and Xinyu Su and Xiaomeng Yang and Yuetan Lin and Zixiong Yang and Zhiyu Tan and Hao Li},
   year   = {2026},
-  note   = {https://huggingface.co/sais-org/Polaris_Pro}
+  note   = {https://huggingface.co/sais-org/ShenZhen}
 }
 ```