Improve model card for Variational Reasoning for Language Models

README.md

@@ -1,37 +1,34 @@
 ---
 library_name: transformers
 tags: []
+pipeline_tag: text-generation
 ---
 
 # Model Card for Model ID
 
-<!-- Provide a quick summary of what the model is/does. -->
-
-
+This model is based on the work presented in the paper [Variational Reasoning for Language Models](https://huggingface.co/papers/2509.22637). It introduces a variational reasoning framework for language models that treats thinking traces as latent variables and optimizes them through variational inference. The method aims to improve the reasoning ability of language models.
 
 ## Model Details
 
 ### Model Description
 
-<!-- Provide a longer summary of what this model is. -->
-
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+This model card describes a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
 
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
+-   **Developed by:** Xiangxin Zhou, Zichen Liu, Haonan Wang, Chao Du, Min Lin, Chongxuan Li, Liang Wang, Tianyu Pang (Authors of the paper)
+-   **Funded by [optional]:** [More Information Needed]
+-   **Shared by [optional]:** [More Information Needed]
+-   **Model type:** Causal Language Model (Qwen3ForCausalLM), as indicated by `config.json`.
+-   **Language(s) (NLP):** English (as implied by common reasoning tasks and the source material.)
+-   **License:** [More Information Needed]
+-   **Finetuned from model [optional]:** The paper mentions "Qwen 2.5 and Qwen 3 model families". The `config.json` indicates `model_type: qwen3`. The GitHub README table lists backbones such as `Qwen3-4B-Base`, `Qwen3-8B-Base`, `Qwen2.5-7B-Instruct`, `Qwen2.5-32B-Instruct`.
 
 ### Model Sources [optional]
 
 <!-- Provide the basic links for the model. -->
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+-   **Repository:** [https://github.com/sail-sg/variational-reasoning](https://github.com/sail-sg/variational-reasoning)
+-   **Paper [optional]:** [https://huggingface.co/papers/2509.22637](https://huggingface.co/papers/2509.22637)
+-   **Demo [optional]:** [More Information Needed]
 
 ## Uses
 
@@ -39,37 +36,32 @@ This is the model card of a 🤗 transformers model that has been pushed on the
 
 ### Direct Use
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
-
-[More Information Needed]
+The model is intended for improving the reasoning abilities of large language models, particularly on tasks requiring explicit thinking traces, as demonstrated in the paper.
 
 ### Downstream Use [optional]
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
-
 [More Information Needed]
 
 ### Out-of-Scope Use
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
-
-[More Information Needed]
+The model is primarily designed for reasoning tasks. Its performance on general conversational or creative generation tasks without further fine-tuning may vary.
 
 ## Bias, Risks, and Limitations
 
 <!-- This section is meant to convey both technical and sociotechnical limitations. -->
 
-[More Information Needed]
+General limitations of large language models apply, including potential biases inherited from training data, tendencies for hallucination, and sensitivity to prompt phrasing. The paper focuses on a principled probabilistic perspective for improving reasoning. Users should be aware that while reasoning is improved, these foundational limitations might still be present.
 
 ### Recommendations
 
 <!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Thorough testing and evaluation on specific target tasks and data are recommended to ensure suitability and mitigate potential issues.
 
 ## How to Get Started with the Model
 
 Use the code below to get started with the model.
+For detailed instructions on setting up the environment, training, and evaluation, please refer to the official [GitHub repository](https://github.com/sail-sg/variational-reasoning).
 
 [More Information Needed]
 
@@ -79,12 +71,23 @@ Use the code below to get started with the model.
 
 <!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
 
-[More Information Needed]
+The models are trained using various datasets, including `Variational-Posterior-4B-Acc-mix`, `Variational-Posterior-4B-GML-mix`, `Variational-Posterior-8B-Acc-mix`, `Variational-Posterior-8B-GML-mix`, etc., as indicated in the GitHub repository's "Models and Datasets" table.
 
 ### Training Procedure
 
 <!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
 
+The training procedure, detailed in the GitHub README, involves several steps:
+1.  Training an initial reasoning model ($\pi_{\theta_0}$).
+2.  Training a variational posterior ($q_\phi$).
+3.  Sampling from the variational posterior.
+4.  Estimating log likelihoods using both the initial reasoning model and the variational posterior.
+5.  (Optional) Sampling from the initial reasoning model and verification (for accuracy-based estimator).
+6.  Building the dataset for training the final reasoning model ($\pi_\theta$) using either geometric mean of token likelihood (GML) or accuracy-based estimators.
+7.  Training the final reasoning model ($\pi_\theta$).
+
+Detailed scripts are available in the GitHub repository under `LLaMA-Factory/variational_reasoning/train/`.
+
 #### Preprocessing [optional]
 
 [More Information Needed]
@@ -92,7 +95,7 @@ Use the code below to get started with the model.
 
 #### Training Hyperparameters
 
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+Training is performed using DeepSpeed and assumes a distributed setup (e.g., 2 nodes x 8 H100 GPUs). `gradient_accumulation_steps` are adjusted to maintain effective batch size across different setups. Specific hyperparameters are outlined in `.yaml` files in the GitHub repository.
 
 #### Speeds, Sizes, Times [optional]
 
@@ -106,6 +109,8 @@ Use the code below to get started with the model.
 
 ### Testing Data, Factors & Metrics
 
+Evaluation is performed on a wide range of reasoning tasks. The GitHub repository provides details and scripts for evaluation under `SkyThought/variational_reasoning/eval/eval.sh`.
+
 #### Testing Data
 
 <!-- This should link to a Dataset Card if possible. -->
@@ -122,11 +127,11 @@ Use the code below to get started with the model.
 
 <!-- These are the evaluation metrics being used, ideally with a description of why. -->
 
-[More Information Needed]
+Evaluation metrics typically include accuracy or other task-specific performance measures relevant to reasoning tasks.
 
 ### Results
 
-[More Information Needed]
+Results are empirically validated on Qwen 2.5 and Qwen 3 model families across various reasoning tasks. For detailed quantitative results, please refer to the [paper](https://huggingface.co/papers/2509.22637) and the evaluation scripts in the [GitHub repository](https://github.com/sail-sg/variational-reasoning).
 
 #### Summary
 
@@ -144,17 +149,17 @@ Use the code below to get started with the model.
 
 Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
 
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
+-   **Hardware Type:** [More Information Needed]
+-   **Hours used:** [More Information Needed]
+-   **Cloud Provider:** [More Information Needed]
+-   **Compute Region:** [More Information Needed]
+-   **Carbon Emitted:** [More Information Needed]
 
 ## Technical Specifications [optional]
 
 ### Model Architecture and Objective
 
-[More Information Needed]
+The model employs a variational reasoning framework that treats thinking traces as latent variables. It extends the Evidence Lower Bound (ELBO) to a multi-trace objective and proposes a forward-KL formulation for stable training. This framework unifies variational inference with RL-style methods to enhance language models' reasoning ability.
 
 ### Compute Infrastructure
 
@@ -172,13 +177,22 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
 
 <!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
 
+If you find this work useful, please consider citing the paper:
+
 **BibTeX:**
 
-[More Information Needed]
+```bib
+@article{zhou2025variationalreasoninglanguagemodels,
+      title={Variational Reasoning for Language Models}, 
+      author={Xiangxin Zhou and Zichen Liu and Haonan Wang and Chao Du and Min Lin and Chongxuan Li and Liang Wang and Tianyu Pang},
+      journal={arXiv preprint arXiv:2509.22637},
+      year={2025}
+}
+```
 
 **APA:**
 
-[More Information Needed]
+Zhou, X., Liu, Z., Wang, H., Du, C., Lin, M., Li, C., Wang, L., & Pang, T. (2025). *Variational Reasoning for Language Models*. arXiv preprint arXiv:2509.22637.
 
 ## Glossary [optional]
 
@@ -196,4 +210,4 @@ Carbon emissions can be estimated using the [Machine Learning Impact calculator]
 
 ## Model Card Contact
 
-[More Information Needed]
+For questions related to this model, please refer to the contact information provided in the original paper or on the [GitHub repository](https://github.com/sail-sg/variational-reasoning).