bunnycore/Phi-4-ReasoningRP

This model is Phi-4 with a reasoning fine-tuned LoRA applied. While it can follow a reasoning format, it's important to understand that its "thinking" isn't the same as more advanced reasoning models (like R1 or O1). Think of it as Phi-4 with a helpful reasoning boost.

What can it do?

This model is designed for roleplay and other reasoning-related tasks. It's not intended to be a replacement for specialized reasoning models; it has its own strengths and limitations.

To activate the reasoning format, use the tag within the system prompt. This will encourage the model to structure its response in a step-by-step or explanatory manner.

Chat Template:

<|im_start|>system<|im_sep|>{system_prompt}<|im_end|>
<|im_start|>user<|im_sep|>{user}<|im_end|>
<|im_start|>assistant<|im_sep|>

Example System Prompt (with reasoning):

You are a helpful assistant. <think> Let's break this down step by step. First, we need to consider... Then, we can look at... Finally, we arrive at the answer. </think> Strengths:

• Capable of roleplay.
• Can follow a reasoning format when prompted.
• Based on the Phi-4 architecture.

Benchmark:

Merge Details

Merge Method

This model was merged using the Passthrough merge method using bunnycore/Phi-4-Model-Stock-v4 + bunnycore/Phi-4-14B-1M-RRP-v1-lora as a base.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

base_model: bunnycore/Phi-4-Model-Stock-v4+bunnycore/Phi-4-14B-1M-RRP-v1-lora
dtype: bfloat16
merge_method: passthrough
models:
  - model: bunnycore/Phi-4-Model-Stock-v4+bunnycore/Phi-4-14B-1M-RRP-v1-lora

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	40.73
IFEval (0-Shot)	67.36
BBH (3-Shot)	55.88
MATH Lvl 5 (4-Shot)	44.34
GPQA (0-shot)	12.53
MuSR (0-shot)	15.14
MMLU-PRO (5-shot)	49.12