README.md

---
base_model:
- skatardude10/SnowDrogito-RpR-32B
library_name: transformers
tags:
- merge
- mergekit
- quantization
---
<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_IQ4-XS
</h1>

<p align="center">
  <img src="https://cdn-uploads.huggingface.co/production/uploads/633e3b4136e87ddc64ad584d/XriPrqbrwSAju1XrNoxLK.png" alt="SnowDrogito-RpR-32B Banner" width="600"/>
</p>

## <span style="color: #CCFFCC;">Updates and Description of Files</span>
- Recent files uploaded use ArliAI RpR V3 instead of V1 as indicated in the name.
- All quantizations in this repo use IQ4_XS as a base with Q8 embedding and output tensors.
- (Recommended) SnowDrogito-RpR3-32B_IQ4-XS+Enhanced_Tensors.gguf - largest, highest quality, Q4KM size, quant using recalibrated imatrix on Bartowki's dataset+RP+Tao at 8k context, uses selective quantization with llama-quantize --tensor-type flags to bump up select FFN/self attention tensors between Q6 and Q8 as <a href="https://github.com/ggml-org/llama.cpp/pull/12718" style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">described here.</a>
- SnowDrogito-RpRv3-32B_IQ4-XS-Q8InOut-Q56Attn.gguf - Q6 and Q5 Attention tensors. This and all quants uploaded prior used imatrix from Snowdrop.

## <span style="color: #CCFFCC;">MORE SPEED!</span>
Improve inference speed offloading tensors instead of layers as referenced <a href="https://www.reddit.com/r/LocalLLaMA/comments/1ki7tg7/dont_offload_gguf_layers_offload_tensors_200_gen/"  style="color: #E6E6FA; text-decoration: none;" onmouseover="this.style.color='#ADD8E6'" onmouseout="this.style.color='#E6E6FA'">HERE</a>.
--overridetensors "\.[13579]\.ffn_up|\.[1-3][13579]\.ffn_up=CPU Restricts offloading of every third FFN up tensor, saving enough space on GPU to offload all layers on 24gb, taking me from 3.9tps to 10.6 tps. Example:
```
python koboldcpp.py --gpulayers 65 --quantkv 1 --overridetensors "\.[13579]\.ffn_up|\.[1-3][13579]\.ffn_up=CPU" --threads 10 --usecublas --contextsize 40960 --flashattention --model ~/Downloads/SnowDrogito-RpR3-32B_IQ4-XS+Enhanced_Tensors.gguf
```
...obviously editing threads, filepaths, etc...

## <span style="color: #CCFFCC;">Overview</span>
SnowDrogito-RpR-32B_IQ4-XS is my shot at an optimized imatrix quantization for my QwQ RP Reasoning merge, goal is 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, quantized to IQ4_XS with <span style="color: #E6E6FA;">Q8_0 embeddings and output layers</span> for enhanced quality, to plus it up just a bit. 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

Getting great reasoning results with ST's *Start Reply With*: 
```
<think>
Chain-of-thought: Alright, what just happened is
```
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
```
- 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
- Quantization: IQ4_XS with <span style="color: #E6E6FA;">Q8_0 embeddings and output layers</span> for better quality.
- Used .imatrix file from Snowdrop.

## <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;">Quantization Details</span>
- Primary Quantization: IQ4_XS (4-bit integer with extra-small blocks) using an importance matrix (trashpanda-org_QwQ-32B-Snowdrop-v0.imatrix) for high quality at reduced size.
- Embeddings & Output Layers: Quantized to <span style="color: #E6E6FA;">Q8_0</span> (8-bit) to preserve precision in token embeddings and final output weights, differing from the standard IQ4_XS body. This boosts quality with a modest size increase.

## <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>.