Update README.md

README.md

@@ -32,25 +32,16 @@ library_name: transformers
 <h1 style="margin-top:0rem; margin-bottom: 0rem;">🐋 DeepSeek-R1-0528-Qwen3-8B Usage Guidelines</h1>
 </div>
 
-| Setting       | Non-Thinking Mode | Thinking Mode |
-|---------------|-------------------|----------------|
-| Temperature   | 0.7               | 0.6            |
-| Min_P         | 0.0  | 0.0            |
-| Top_P         | 0.8               | 0.95           |
-| TopK          | 20                | 20             |
-
-<h4 style="margin-top:0rem;">Chat template/prompt format:</h4>
- 
+- Set the temperature between **0.5–0.7 (0.6 recommended)** to reduce repetition and incoherence.
+- Set Top_P value of **0.95 (recommended)**
+- R1-0528 uses the same chat template as the original R1 model:
 ```
-<|im_start|>user\nWhat is 2+2?<|im_end|>\n<|im_start|>assistant\n
+<｜begin▁of▁sentence｜><｜User｜>What is 1+1?<｜Assistant｜>It's 2.<｜end▁of▁sentence｜><｜User｜>Explain more!<｜Assistant｜>
 ```
-- For NON thinking mode, we purposely enclose <think> and </think> with nothing:
-
+- For llama.cpp / GGUF inference, you should skip the BOS since it’ll auto add it:
 ```
-<|im_start|>user\nWhat is 2+2?<|im_end|>\n<|im_start|>assistant\n<think>\n\n</think>\n\n
+<｜User｜>What is 1+1?<｜Assistant｜>
 ```
-- For Thinking-mode, DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions.
-
 - For complete detailed instructions, see our guide: [unsloth.ai/blog/deepseek-r1-0528](https://docs.unsloth.ai/basics/deepseek-r1-0528-how-to-run-locally)
 
 ---