README.md

metadata

language:
  - zh
  - en
license: gpl-3.0
tags:
  - qwen
  - uncensored
  - llama-cpp
  - gguf-my-repo
base_model: Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
datasets:
  - NobodyExistsOnTheInternet/ToxicQAFinal
  - anthracite-org/kalo-opus-instruct-22k-no-refusal
  - Orion-zhen/dpo-toxic-zh
  - unalignment/toxic-dpo-v0.2
  - Crystalcareai/Intel-DPO-Pairs-Norefusals
pipeline_tag: text-generation
model-index:
  - name: Qwen2.5-7B-Instruct-Uncensored
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: IFEval (0-Shot)
          type: HuggingFaceH4/ifeval
          args:
            num_few_shot: 0
        metrics:
          - type: inst_level_strict_acc and prompt_level_strict_acc
            value: 72.04
            name: strict accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: BBH (3-Shot)
          type: BBH
          args:
            num_few_shot: 3
        metrics:
          - type: acc_norm
            value: 35.83
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MATH Lvl 5 (4-Shot)
          type: hendrycks/competition_math
          args:
            num_few_shot: 4
        metrics:
          - type: exact_match
            value: 1.36
            name: exact match
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GPQA (0-shot)
          type: Idavidrein/gpqa
          args:
            num_few_shot: 0
        metrics:
          - type: acc_norm
            value: 7.05
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          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: 13.58
            name: acc_norm
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          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: 38.07
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard?query=Orion-zhen/Qwen2.5-7B-Instruct-Uncensored
          name: Open LLM Leaderboard

Triangle104/Qwen2.5-7B-Instruct-Uncensored-Q6_K-GGUF

This model was converted to GGUF format from Orion-zhen/Qwen2.5-7B-Instruct-Uncensored using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model.

---

Model details:

This model is an uncensored fine-tune version of Qwen2.5-7B-Instruct. However, I can still notice that though uncensored, the model fails to generate detailed descriptions on certain extreme scenarios, which might be associated with deletion on some pretrain datasets in Qwen's pretraining stage.

Traning details

I used SFT + DPO to ensure uncensorment as well as trying to maintain original model's capabilities.

SFT: NobodyExistsOnTheInternet/ToxicQAFinal anthracite-org/kalo-opus-instruct-22k-no-refusal

DPO: Orion-zhen/dpo-toxic-zh unalignment/toxic-dpo-v0.2 Crystalcareai/Intel-DPO-Pairs-Norefusals

---

Use with llama.cpp

Install llama.cpp through brew (works on Mac and Linux)

brew install llama.cpp

Invoke the llama.cpp server or the CLI.

CLI:

llama-cli --hf-repo Triangle104/Qwen2.5-7B-Instruct-Uncensored-Q6_K-GGUF --hf-file qwen2.5-7b-instruct-uncensored-q6_k.gguf -p "The meaning to life and the universe is"

Server:

llama-server --hf-repo Triangle104/Qwen2.5-7B-Instruct-Uncensored-Q6_K-GGUF --hf-file qwen2.5-7b-instruct-uncensored-q6_k.gguf -c 2048

Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well.

Step 1: Clone llama.cpp from GitHub.

git clone https://github.com/ggerganov/llama.cpp

Step 2: Move into the llama.cpp folder and build it with LLAMA_CURL=1 flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux).

cd llama.cpp && LLAMA_CURL=1 make

Step 3: Run inference through the main binary.

./llama-cli --hf-repo Triangle104/Qwen2.5-7B-Instruct-Uncensored-Q6_K-GGUF --hf-file qwen2.5-7b-instruct-uncensored-q6_k.gguf -p "The meaning to life and the universe is"

or

./llama-server --hf-repo Triangle104/Qwen2.5-7B-Instruct-Uncensored-Q6_K-GGUF --hf-file qwen2.5-7b-instruct-uncensored-q6_k.gguf -c 2048