Create README.md

README.md

@@ -0,0 +1,62 @@
+---
+license: mit
+datasets:
+- nvidia/Llama-Nemotron-Post-Training-Dataset
+language:
+- en
+base_model:
+- microsoft/phi-4
+- unsloth/phi-4
+pipeline_tag: text-generation
+library_name: transformers
+tags:
+- phi-4
+- deepseek
+- r1
+- reasoning
+- code
+- math
+- science
+- unsloth
+---
+# Model Card for Model ID
+
+Phi-4 unsloth model trained to generate deepseek-r1 styled reasoning based on a system prompt "detailed thinking on"
+
+
+## Model Details
+
+### Model Description
+
+This fine-tuned model generated ehanced chain-of-thoughts, reasoning and produce "Aha-moments" akin to deepseek whenever the system prompt is set to "detailed thinking on". 
+Test any questions from trending datasets about code, math and science, with the system prompt set and unset and you can clearly see the difference in the generated output. 
+A Modelfile is included with the gguf files that can be used to load the model into Ollama. You have to set the system prompt manually after loading the model in Ollama since by
+default there is no system prompt. You can use /set SYSTEM "detailed thinking on" and then input your prompt. The Modelfile includes optimal parameters but you can experimment 
+with different set of parameters based on your desired goal/output.
+
+
+## Uses
+
+Tasks requiring reasoning, chain-of-thoughts, several approaches etc...
+
+
+### Recommendations
+
+Use the parameters in the Modefile and set system prompt to "detailed thinking on" whenever you require long reasoning outputs. Set system parameter off when you want a direct
+to the point quick answer without reasoning chains.
+It is important to utilize the chat-template embedded in the Modelfile to ensure optimal generations and avoid endless generations or loops.
+
+## How to Get Started with the Model
+
+Download the gguf file and Modelfile into the same folder then use ollama create phi4-deepseek -f Modelfile. Then run the model using ollama run. Set the system parameter 
+Finally, start prompting. 
+
+## Training Details
+
+### Training Data
+
+Nvidia datasets containing reasoning context produced by DeeepSeek-R1.
+
+### Training Procedure
+
+Unsloth SFT Trainer