Improve model card: add library_name, paper link, and clean up structure

README.md

@@ -1,105 +1,34 @@
 ---
+base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
 language:
 - zh
 - en
+license: other
 pipeline_tag: text-generation
+library_name: transformers
 tags:
 - deepscaler
 - reasoning
 - grpo
 - qwen2
-base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
-license: other
 ---
 
 # DECS_7B
 
-This is the official model for ICLR 2026 Oral "Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling".
-DECS_7B is a reasoning-focused causal language model built from `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B` and further trained with DECS algorithm, focused on 50% fewer tokens when answering a reasoning-required problem.
-
-## Model Summary
-
-- Base model: `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B`
-- Upload date: `2026-02-24`
-- Recommended use: long-form reasoning and mathematical/problem-solving style generation
-
-## Quick Start (Transformers)
-
-```python
-import torch
-from transformers import AutoModelForCausalLM, AutoTokenizer
-
-model_id = "pixas/DECS_7B"
-tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
-model = AutoModelForCausalLM.from_pretrained(
-    model_id,
-    torch_dtype=torch.bfloat16,
-    device_map="auto",
-)
-
-messages = [
-    {"role": "user", "content": "Solve: If x^2 - 5x + 6 = 0, what are x values?"}
-]
-prompt = tokenizer.apply_chat_template(
-    messages, tokenize=False, add_generation_prompt=True
-)
-inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
-
-with torch.no_grad():
-    outputs = model.generate(
-        **inputs,
-        max_new_tokens=512,
-        temperature=0.6,
-        top_p=0.95,
-    )
-
-new_tokens = outputs[0][inputs["input_ids"].shape[-1]:]
-print(tokenizer.decode(new_tokens, skip_special_tokens=True))
-```
+This is the official model repository for **DECS_7B**, presented in the ICLR 2026 Oral paper: **"Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling"**.
 
-## Quick Start (vLLM)
+[**Paper**](https://huggingface.co/papers/2509.25827) | [**Code**](https://github.com/pixas/DECS) | [**Project Page**](https://pixas.github.io/decs-iclr26-site/)
 
-```python
-from vllm import LLM, SamplingParams
+## Model Description
+DECS_7B is a reasoning-focused causal language model built from `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B` and further trained with the **DECS** (Decoupled Rewards and Curriculum Scheduling) algorithm. 
 
-llm = LLM(model="pixas/DECS_7B", trust_remote_code=True)
-sampling = SamplingParams(temperature=0.6, top_p=0.95, max_tokens=512)
-prompt = "Please reason step by step: what is 37 * 48?"
-outputs = llm.generate([prompt], sampling_params=sampling)
-print(outputs[0].outputs[0].text)
-```
-
-## Notes
-
-- This model may produce incorrect or unverifiable reasoning. Always validate outputs in high-stakes settings.
-- Performance can vary by prompt style and decoding parameters.
-- License and acceptable-use constraints should follow the upstream base model and your deployment policy.
-
-
-## Citation
----
-language:
-- zh
-- en
-pipeline_tag: text-generation
-tags:
-- deepscaler
-- reasoning
-- grpo
-- qwen2
-base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
-license: other
----
-
-# DECS_1.5B
-This is the official model for ICLR 2026 Oral "Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling".
-DECS_1.5B is a reasoning-focused causal language model built from `deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B` and further trained with DECS algorithm, focused on 50% fewer tokens when answering a reasoning-required problem.
+The DECS framework addresses the "overthinking" problem in large reasoning models—where models generate excessively long reasoning paths without performance benefits. DECS achieves a reduction in reasoning tokens by over 50% across multiple benchmarks while maintaining or improving accuracy. It introduces a decoupled token-level reward mechanism and a curriculum batch scheduling strategy to optimize the efficiency-efficacy equilibrium.
 
 ## Model Summary
 
-- Base model: `deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B`
-- Upload date: `2026-02-24`
-- Recommended use: long-form reasoning and mathematical/problem-solving style generation
+- **Base model:** `deepseek-ai/DeepSeek-R1-Distill-Qwen-7B`
+- **Upload date:** `2026-02-24`
+- **Recommended use:** Long-form reasoning, mathematical problem solving, and efficient step-by-step logic generation.
 
 ## Quick Start (Transformers)
 
@@ -107,7 +36,7 @@ DECS_1.5B is a reasoning-focused causal language model built from `deepseek-ai/D
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
-model_id = "pixas/DECS_1.5B"
+model_id = "pixas/DECS_7B"
 tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
 model = AutoModelForCausalLM.from_pretrained(
     model_id,
@@ -140,7 +69,7 @@ print(tokenizer.decode(new_tokens, skip_special_tokens=True))
 ```python
 from vllm import LLM, SamplingParams
 
-llm = LLM(model="pixas/DECS_1.5B", trust_remote_code=True)
+llm = LLM(model="pixas/DECS_7B", trust_remote_code=True)
 sampling = SamplingParams(temperature=0.6, top_p=0.95, max_tokens=512)
 prompt = "Please reason step by step: what is 37 * 48?"
 outputs = llm.generate([prompt], sampling_params=sampling)
@@ -149,20 +78,20 @@ print(outputs[0].outputs[0].text)
 
 ## Notes
 
-- This model may produce incorrect or unverifiable reasoning. Always validate outputs in high-stakes settings.
-- Performance can vary by prompt style and decoding parameters.
-- License and acceptable-use constraints should follow the upstream base model and your deployment policy.
-
+- **Reasoning Accuracy:** While optimized for efficiency, this model may produce incorrect or unverifiable reasoning. Always validate outputs in high-stakes settings.
+- **Licensing:** License and acceptable-use constraints follow the upstream base model and your deployment policy.
 
 ## Citation
 
 If you use this model, please cite our paper:
+
 ```bibtex
-@inproceedings{jiang2026overthinking,
-title={Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling},
-author={Shuyang Jiang and Yusheng Liao and Ya Zhang and Yanfeng Wang and Yu Wang},
-booktitle={The Fourteenth International Conference on Learning Representations},
-year={2026},
-url={https://openreview.net/forum?id=kdeiRledV6}
+@inproceedings{jiang2026decs,
+  title     = {Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling},
+  author    = {Jiang, Shuyang and Tao, Xiaofeng and Zhang, Kui and Xiao, Yanghua},
+  booktitle = {International Conference on Learning Representations (ICLR)},
+  year      = {2026},
+  note      = {Oral},
+  url       = {https://arxiv.org/abs/2509.25827}
 }
-```
+```