Update README.md

README.md

@@ -43,6 +43,42 @@ outputs = model.generate(**inputs, max_new_tokens=128)
 print(AutoTokenizer.from_pretrained(model_name).decode(outputs[0], skip_special_tokens=True))
 ```
 
+## Model Description
+
+Kazakh text generation model fine-tuned from `google/mt5-base` on the `Darmm/darmm-text-generation-kazakh` dataset.
+
+## Training (summary)
+
+- Base model: `google/mt5-base`
+- Epochs: 3
+- Batch size: 2
+- Learning rate: 1e-4
+- Max input length: 256
+- Max target length: 256
+
+## Metrics (summary)
+
+```
+{
+  "eval_loss": 0.08725570142269135,
+  "eval_exact_match": 0.5547445255474452,
+  "eval_rouge1": 0.10948905109489052,
+  "eval_rouge2": 0.10583941605839416,
+  "eval_rougeL": 0.10948905109489052,
+  "epoch": 3.0
+}
+```
+
+## Intended use
+
+- Instruction-style Kazakh text generation for short responses.
+- Educational and informational content generation prototypes.
+
+## Limitations
+
+- Limited dataset size may reduce generalization to unseen domains.
+- Outputs may be generic for short prompts.
+
 ## Paper & Documentation
 
 <details>