README.md

metadata

base_model:
  - benhaotang/Phi-4-llama-t1-full
  - prithivMLmods/Phi-4-QwQ
  - win10/Phi-4-llama-t1-lora
library_name: transformers
tags:
  - mergekit
  - merge
datasets:
  - NovaSky-AI/Sky-T1_data_17k
license: mit
model-index:
  - name: phi4-qwq-sky-t1
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: wis-k/instruction-following-eval
          split: train
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 4.6
            name: averaged accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: SaylorTwift/bbh
          split: test
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 52.61
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: lighteval/MATH-Hard
          split: test
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 39.58
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          split: train
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 19.35
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MuSR (0-shot)
          type: TAUR-Lab/MuSR
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 21.38
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU-PRO (5-shot)
          type: TIGER-Lab/MMLU-Pro
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 47.16
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard#/?search=benhaotang%2Fphi4-qwq-sky-t1
          name: Open LLM Leaderboard

merge

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the TIES merge method using prithivMLmods/Phi-4-QwQ as a base.

Models Merged

The following models were included in the merge:

• benhaotang/Phi-4-llama-t1-full but actually win10/Phi-4-llama-t1-lora, this is who and where you should really thank.
• prithivMLmods/Phi-4-QwQ

Eval

IFEval is broken due to the Sky-T1 strict system prompt format, but other than that, seems to have recreated qwq at 14B.

Running

• With Ollama

ollama run hf.co/benhaotang/phi4-qwq-sky-t1-Q4_K_M-GGUF

I suggest adding SYSTEM "You are a helpful AI asistent. You always think step by step." to triger step by step reasoning.

• With pytorch

import transformers
tokenizer = AutoTokenizer.from_pretrained("mircosoft/phi-4")
pipeline = transformers.pipeline(
    "text-generation",
    model="benhaotang/phi4-qwq-sky-t1",
    tokenizer=tokenizer,
    device_map="auto",
)
messages = [
    {"role": "system", "content": "You are a helpful AI asistent. You always think step by step."},
    {"role": "user", "content": "Give me a short intodcution to renormalization group(RG) flow in physcis?"},
]
outputs = pipeline(messages, max_new_tokens=128)
print(outputs[0]["generated_text"])

Configuration

The following YAML configuration was used to produce this model:

models:
  - model: prithivMLmods/Phi-4-QwQ
    #no parameters necessary for base model
  - model: benhaotang/Phi-4-llama-t1-full
    parameters:
      density: 0.5
      weight: 0.5
  - model: prithivMLmods/Phi-4-QwQ
    parameters:
      density: 0.5
      weight: 0.5

merge_method: ties
base_model: prithivMLmods/Phi-4-QwQ
parameters:
  normalize: false
  int8_mask: true
dtype: float16

Open LLM Leaderboard Evaluation Results

Detailed results can be found here! Summarized results can be found here!

Metric	Value (%)
Average	30.78
IFEval (0-Shot)	4.60
BBH (3-Shot)	52.61
MATH Lvl 5 (4-Shot)	39.58
GPQA (0-shot)	19.35
MuSR (0-shot)	21.38
MMLU-PRO (5-shot)	47.16