Upload README.md with huggingface_hub

README.md

@@ -29,7 +29,8 @@ Designed for coding agents and experienced .NET developers who need compilable,
 |---|---|
 | **Parameters** | 14.7B |
 | **Base Model** | Qwen2.5-Coder-14B-Instruct |
-| **Context Length** | 32,768 tokens (trained on 2,048) |
+| **Max Context** | 32,768 tokens (base model) |
+| **Trained Sequence Length** | 2,048 tokens |
 | **Training Method** | QLoRA SFT + Iterative DPO |
 | **Training Data** | 107K C# records |
 | **License** | Apache 2.0 |
@@ -208,7 +209,7 @@ The model handles both code generation and interactive debugging — it can diag
 
 **Inference parameters**: temperature=0.2, top_p=0.9, max_new_tokens=2048
 
-The model was fine-tuned on sequences up to 2048 tokens (prompt + response). It will stop generating when done (EOS token), so setting a higher limit won't cause unnecessary output. For tasks requiring longer output, the base model's 32K context still applies, but quality may vary beyond the training distribution.
+The base model supports up to 32,768 tokens, so you can use the full 32K context window. The fine-tuning was done on sequences up to 2,048 tokens — the model performs best within this range but still works beyond it thanks to the base model's capabilities. The model will stop generating when done (EOS token), so setting a higher limit won't cause unnecessary output.
 
 ## Usage