Update README.md

README.md

@@ -19,14 +19,14 @@ library_name: transformers
 
 **DeepBrainz-R1-4B-40K** is a compact, high-performance reasoning model engineered by **DeepBrainz AI & Labs**. It is part of the **DeepBrainz-R1 Series**, designed to deliver frontier-class reasoning capabilities in cost-effective parameter sizes.
 
-This specific variant offers a **40,960 token context window**, making it suitable for `maximum context version designed for repository-level code reasoning.`.
+This specific variant offers a **40,960 token context window**, making it suitable for extended-context evaluation and repository-level code reasoning.
 
 ---
 
 ## 🚀 Model Highlights
 
 - **Parameter Count:** ~4B
-- **Context Window:** 40,960 tokens
+- **Context Window:** up to 40,960 tokens (extended context; experimental)
 - **Context Type:** Extended (RoPE)
 - **Specialization:** STEM Reasoning, Logic, Code Analysis
 - **Architecture:** Optimized Dense Transformer
@@ -41,7 +41,8 @@ This specific variant offers a **40,960 token context window**, making it suitab
 - **Code Generation:** Writing and debugging algorithms.
 - **Structured Data Extraction:** Parsing and reasoning over unstructured text.
 
-> **Note:** This is a base reasoning model. For conversational chat, we recommend using a specific instruct template or fine-tuning on your domain data.
+> **Note:** This is a post-trained reasoning variant intended for evaluation and experimentation.  
+> It is not production-validated and is not optimized for open-ended conversational chat.
 
 ---
 
@@ -70,9 +71,9 @@ print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 
 ## 🏗️ Technical Summary
 
-The model was produced using a **multi-stage optimization process** involving large-scale supervision and iterative refinement. It is designed to maximize reasoning quality while maintaining instruction robustness.
+This model has undergone **post-training** to improve structured reasoning behavior, mathematical problem solving, and robustness in agentic workflows.
 
-*Specific training methodologies and dataset compositions are proprietary.*
+*Detailed post-training recipes and dataset compositions are not fully disclosed.*
 
 ---