SnowDrogito-RpR-32B_4.5bpw_H6_EXL3

EXL3 4.5 bpw H6 quant of skatardude10/SnowDrogito-RpR-32B

Base models by

• @Qwen: QwQ-32B
• @Qwen: Qwen2.5-32B
• @ArliAI: QwQ-32B-ArliAI-RpR-v1
• @deepcogito: cogito-v1-preview-qwen-32B
• @trashpanda-org: QwQ-32B-Snowdrop-v0

Merge by @skatardude10: SnowDrogito-RpR-32B

Original model card provided by the author:

<h1 align="center"> <span style="color: #ADD8E6; font-weight: bold;">SnowDr</span><span style="color: #00FF00; font-weight: bold; font-style: italic;">ogito</span><span style="color: #FFFFFF; font-weight: bold;">-</span><span style="color: #FF9999; font-weight: bold;">RpR</span>-32B </h1> <p align="center"> <img src="https://cdn-uploads.huggingface.co/production/uploads/633e3b4136e87ddc64ad584d/XriPrqbrwSAju1XrNoxLK.png" alt="SnowDrogito-RpR-32B Banner" width="600"/> </p> <p align="center"> <a href="https://huggingface.co/skatardude10/SnowDrogito-RpR-32B_IQ4-XS" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">Download IQ4_XS IMATRIX GGUF</a> </p>

<span style="color: #CCFFCC;">Overview</span>

SnowDrogito-RpR-32B is a QwQ RP Reasoning merge to add smarts to the popular <span style="color: #ADD8E6;">Snowdrop</span> roleplay model, with a little <span style="color: #FF9999;">ArliAI RpR</span> and <span style="color: #00FF00;">Deepcogito</span> for the smarts. Built using the TIES merge method, it attempts to combine strengths from multiple fine-tuned QwQ-32B models. Uploading because the PPL was lower, have been getting more varied/longer/more creative responses with this, but maybe it lacks contextual awareness compared to snowdrop? Not sure.

<span style="color: #CCFFCC;">Setup for Reasoning and ChatML</span>

• ChatML Formatting: Use ChatML with <|im_start|>role\ncontent<|im_end|>\n (e.g., <|im_start|>user\nHello!<|im_end|>\n).
• Reasoning Settings: Set "include names" to "never." Start reply with <think>\n to enable reasoning.
• Sampler Settings: From Snowdrop: Try temperature 0.9, min_p 0.05, top_a 0.3, TFS 0.75, repetition_penalty 1.03, DRY if available.
• My Settings: Response (tokens): 2048 Context (tokens): 40960 Temperature: 3.25 Top P: 0.98 Min P: 0.04 Top nsigna: 2.5 Repetition Penalty: 1.03 (XTC) Threshold: 0.3 (XTC) Probability: 0.3 Dry Multiplier: 0.8 Dry Base: 1.75 Dry Allowed Length: 4 Dry Penalty Range: 1024

For more details, see the setup guides and master import for ST for <a href="https://huggingface.co/trashpanda-org/QwQ-32B-Snowdrop-v0" style="color: #ADD8E6; text-decoration: none;" onmouseover="this.style.color='#E6E6FA'" onmouseout="this.style.color='#ADD8E6'">Snowdrop</a> and other info on <a href="https://huggingface.co/ArliAI/QwQ-32B-ArliAI-RpR-v1" style="color: #FF9999; text-decoration: none;" onmouseover="this.style.color='#E6E6FA'" onmouseout="this.style.color='#FF9999'">ArliAI RpR</a>.

<span style="color: #CCFFCC;">Performance</span>

• Perplexity under identical conditions (IQ4_XS, 40,960 context, Q8_0 KV cache, on a 150K-token chat dataset) SnowDrogito-RpR-32B vs <span style="color: #ADD8E6;">QwQ-32B-Snowdrop-v0</span>:

  4.5597 ± 0.02554  
  4.6779 ± 0.02671

• IQ4_xs fits 40960 context 24GB VRAM using Q8 KV Cache with full GPU offload.

<span style="color: #CCFFCC;">Model Details</span>

• Base Model: <a href="https://huggingface.co/Qwen/Qwen2.5-32B" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">Qwen/Qwen2.5-32B</a>
• Architecture: Qwen 2.5 (32B parameters)
• Context Length: 40,960 tokens

<span style="color: #CCFFCC;">Merge Configuration</span>

This model was created using mergekit with the following TIES merge configuration:

models:
  - model: trashpanda-org/QwQ-32B-Snowdrop-v0
    parameters:
      weight: 0.75
      density: 0.5
  - model: deepcogito/cogito-v1-preview-qwen-32B
    parameters:
      weight: 0.15
      density: 0.5
  - model: ArliAI/QwQ-32B-ArliAI-RpR-v1
    parameters:
      weight: 0.1
      density: 0.5
merge_method: ties
base_model: Qwen/Qwen2.5-32B
parameters:
  weight: 0.9
  density: 0.9
  normalize: true
  int8_mask: true
tokenizer_source: Qwen/Qwen2.5-32B-Instruct
dtype: bfloat16

<span style="color: #CCFFCC;">Acknowledgments</span>

• <a href="https://github.com/arcee-ai/mergekit" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">mergekit</a> for merging.
• <a href="https://github.com/ggerganov/llama.cpp" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">llama.cpp</a> for quantization.
• Original model creators: <a href="https://huggingface.co/Qwen" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">Qwen</a>, <a href="https://huggingface.co/trashpanda-org" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">trashpanda-org</a>, <a href="https://huggingface.co/deepcogito" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">deepcogito</a>, <a href="https://huggingface.co/ArliAI" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">ArliAI</a>.