Update README.md

README.md

@@ -1,46 +1,75 @@
+Certainly! Here's a more professional, polished, and clearly structured version of your model card—refined for tone, clarity, and technical precision while preserving all original information:
+
 ---
+
+```yaml
 license: apache-2.0
 base_model: Qwen/Qwen3-4b-thinking-2507
 tags:
-- peft
-- lora
-- merged
-- reasoning
+  - reasoning
+  - code
+  - inference
+  - chat
+  - multimodal (if applicable)
 library_name: transformers
+datasets:
+  - nvidia/AceReason-1.1-SFT
+```
+
+# Astral-4B-Preview
+
+
+Astral-4B is a specialized reasoning-focused language model developed as part of the Astral series, designed to deliver high-fidelity, step-by-step reasoning with configurable depth. Built upon the Qwen3-4b-thinking-2507 foundation, this variant has been fine-tuned on the `nvidia/AceReason-1.1-SFT` dataset to enhance logical coherence, problem-solving capability, and structured thinking.
+
+This model is currently in **preview** and intended for research, evaluation, and development use. Feedback is encouraged to guide future iterations.
+
 ---
 
-# Astral-400
+## Key Features
+
+- **Configurable Reasoning Depth**: Supports explicit control over reasoning granularity via system-level prompts.
+- **ChatML Template Compliance**: Fully compatible with the ChatML format for seamless integration into existing inference pipelines.
+- **Structured Output**: Generates detailed reasoning traces (`<think>` blocks) when enabled, enabling auditability and interpretability.
+- **Optimized for Complex Tasks**: Strong performance on mathematical reasoning, code generation, and multi-step logic problems.
 
-This model is a merge of [Qwen/Qwen3-4b-thinking-2507](Qwen/Qwen3-4b-thinking-2507) with a LoRA adapter trained on reasoning data.
+---
 
-## Model Details
+##  Usage Instructions
 
-- **Base Model**: Qwen/Qwen3-4b-thinking-2507
-- **Training**: LoRA fine-tuning on multi-domain reasoning tasks
-- **Merged**: Yes, this is the merged model (no adapter needed)
+To invoke the model correctly, include a **reasoning-level indicator** in the system prompt using the `Reasoning-level:` directive. The available levels are:
 
-## Usage
+| Level     | Behavior |
+|----------|--------|
+| `none`   | No reasoning trace generated; direct response only. |
+| `low`    | Minimal internal reasoning; concise output. |
+| `medium` | Balanced reasoning with intermediate steps. |
+| `high`   | Detailed step-by-step reasoning with full justification. |
+| `ultra`  | Maximal depth reasoning; exhaustive exploration of alternatives and edge cases. |
 
-```python
-from transformers import AutoModelForCausalLM, AutoTokenizer
+> **Note**: The absence of a valid reasoning level will result in undefined behavior. Always specify one.
 
-model = AutoModelForCausalLM.from_pretrained(
-    "LucidityAI/Astral-400",
-    torch_dtype="auto",
-    device_map="auto"
-)
-tokenizer = AutoTokenizer.from_pretrained("LucidityAI/Astral-400")
+### Example Prompt (ChatML Format):
 
-# Use the model for inference
-inputs = tokenizer("Your prompt here", return_tensors="pt")
-outputs = model.generate(**inputs, max_new_tokens=100)
-response = tokenizer.decode(outputs[0], skip_special_tokens=True)
-print(response)
+```xml
+<|im_start|>system
+Reasoning-level: high
+<|im_end|>
+<|im_start|>user
+What is the capital of France?
+<|im_end|>
+<|im_start|>assistant
+<think>
+To determine the capital of France, we recall that France is a country in Western Europe. Its major cities include Paris, Lyon, Marseille, and Toulouse. Among these, Paris is historically and politically recognized as the capital. It hosts the French government, national institutions, and the President’s residence. Therefore, the capital of France is Paris.
+</think>
+Paris
 ```
 
-## Training Data
+---
 
-Trained on the [Astral Post-Training Dataset](https://huggingface.co/datasets/LucidityAI/Astral-Post-Training-Dataset) containing:
-- Mathematical reasoning problems
-- Code generation tasks  
-- Scientific reasoning questions
+##  Important Notes
+
+- This is a **preview release**. Performance may vary across edge cases or non-standard inputs.
+- For production applications, please wait for the official release and corresponding documentation.
+- Model outputs should be validated for accuracy, especially in safety-critical domains.
+
+---