Update README.md

README.md

@@ -1,4 +1,4 @@
-# 🧠 Phi-4 Reasoning -- Rust Dataset LoRA (Merged)
+# 🧠 Rust-Master-thinking
 
 This repository contains a fine-tuned version of
 **unsloth/phi-4-reasoning**, trained with **LoRA** on the
@@ -45,13 +45,13 @@ The training format follows:
   Alpha            32
   Dropout          0.05
   Target Modules   q/k/v/o proj, mlp (up/down/gate)
-  Max Length       2048
-  Precision        4-bit QLoRA (merged later to BF16/FP16)
-  Batch Size       4
+  Max Length       512
+  Precision        4-bit QLoRA
+  Batch Size       16
   Grad Accum       8
   LR               2e-4
   Scheduler        cosine
-  Epochs           2
+  Epochs           1
 
 ## 📚 Dataset
 
@@ -72,18 +72,35 @@ accurate explanations for Rust programming tasks.
 
 ``` python
 from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
-model_id = "YOUR_USERNAME/YOUR_MODEL_NAME"
+model_id = "SkyAsl/Rust-Master-thinking"
 
 tokenizer = AutoTokenizer.from_pretrained(model_id)
-model = AutoModelForCausalLM.from_pretrained(model_id)
+model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16, device_map="auto")
+model.eval()
 
-prompt = "Explain ownership in Rust with examples."
+prompt = "Explain why Rust ownership prevents data races."
 
-inputs = tokenizer(prompt, return_tensors="pt")
-outputs = model.generate(**inputs, max_new_tokens=300)
+input_text = (
+    f"<|user|>\n{test_data[0]['prompt']}\n"
+    f"<|assistant|>\n<think>\n"
+)
+
+inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
+
+with torch.no_grad():
+    output = model.generate(
+        **inputs,
+        max_new_tokens=500,
+        temperature=0.7,
+        top_p=0.9,
+        do_sample=True,
+        eos_token_id=tokenizer.convert_tokens_to_ids("</think>")
+    )
+
+print(tokenizer.decode(output[0], skip_special_tokens=False))
 
-print(tokenizer.decode(outputs[0], skip_special_tokens=True))
 ```
 
 ## 🔍 Notes on Reasoning Tags
@@ -96,23 +113,6 @@ This model preserves **hidden reasoning structure**:
 ⚠️ Users should NOT expect the `<think>` content to be revealed; the
 model is aligned to hide reasoning by default.
 
-## 📦 Files Included
-
--   `config.json`\
--   `generation_config.json`\
--   `pytorch_model.bin`\
--   `tokenizer.json`
-
-If this is a LoRA-only repo (not merged), then the repo contains:
-
--   `adapter_config.json`\
--   `adapter_model.bin`
-
-## 🔒 License
-
-This model inherits the license of the base model:\
-**Microsoft Phi License / Reasoning Model Terms**
-
 ## ✨ Acknowledgements
 
 -   **Unsloth** for optimized model training\