---
license: apache-2.0
library_name: llama.cpp
base_model:
- bottlecapai/ThinkingCap-Qwen3.6-27B
base_model_relation: quantized
pipeline_tag: text-generation
quantized_by: MagicQuant
language:
- en
tags:
- gguf
- quantized
- magicquant
---

# ThinkingCap-Qwen3.6-27B-MagicQuant-GGUF

Derivative of [ThinkingCap-Qwen3.6-27B](https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B), quantized using MagicQuant hybrid evolutionary per-tensor search.

Sibling repo with AMD-native (ROCmFPX fork-only) builds: [lmcoleman/ThinkingCap-Qwen3.6-27B-ROCmFPX-GGUF](https://huggingface.co/lmcoleman/ThinkingCap-Qwen3.6-27B-ROCmFPX-GGUF).

## Base Model

This is a derivative of [ThinkingCap-Qwen3.6-27B](https://huggingface.co/bottlecapai/ThinkingCap-Qwen3.6-27B).
All credit for the base model architecture and weights goes to the original authors.
The base model's license applies to this derivative.

## Quantization Method

Quantized using **[MagicQuant](https://github.com/lucasmcoleman/MagicQuant)** hybrid evolutionary per-tensor quantization,
based on the methodology by **[magiccodingman](https://github.com/magiccodingman/MagicQuant-Wiki)**:

- Tensors are classified into sensitivity groups (Embeddings, Head, Query, Key, Output, FFN Up/Down, MoE Experts, Router)
- An evolutionary search finds the optimal quantization type per group, balancing size vs. perplexity
- **Q4/Q5/Q6 tier targets** are searched, and each one ships only if it earns its place (see below)
- Small-row tensors and sensitivity-critical layers (embeddings, output head, router) are kept at F32/F16/BF16
- This is NOT a uniform quantization -- each tensor group gets its own optimal type

A tier name here is a **size band**, not a promise that every tensor uses that
exact type. A "Q5" is whatever mix of schemes landed in the Q5 size band with
the lowest measured perplexity loss -- which is the point of the search.

## Files from an earlier build

These files were produced by a **previous quantization run**, not the one this card describes:

- `ThinkingCap-Qwen3.6-27B-Q6_K.gguf` (23.00 GiB) -- size verified as Q6 band

They are kept because they are correctly sized for their tier and remain usable. But they were selected by an earlier version of the search, so their quality was not measured on the same footing as the other files here, and the per-group scheme breakdown above does not describe them.

If you are comparing tiers against each other, prefer the files from the current run -- the comparison is only apples-to-apples within a single search.

## GGUF Files

| File | Size | Quant | Perplexity vs BF16 |
|------|------|-------|--------------------|
| [ThinkingCap-Qwen3.6-27B-Q4_K_M.gguf](./ThinkingCap-Qwen3.6-27B-Q4_K_M.gguf) | 17.9 GB | Q4 hybrid | 6.9051 (+1.84%) |
| [ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf](./ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf) | 20.2 GB | Q5 hybrid | 6.7867 (+0.09%) |
| [ThinkingCap-Qwen3.6-27B-Q6_K.gguf](./ThinkingCap-Qwen3.6-27B-Q6_K.gguf) | 24.7 GB | Q6 hybrid | earlier build, not measured here |
| [mmproj-ThinkingCap-Qwen3.6-27B-f16.gguf](./mmproj-ThinkingCap-Qwen3.6-27B-f16.gguf) | 0.9 GB | F16 (unquantized) | not measured |


Perplexity measured on wikitext-2 (100 chunks, ctx 512) against the BF16 baseline of **6.7804**. Lower is better; the percentage is the increase over BF16. These are the same measurements the tier selection is based on, so a tier that shipped is one that earned its size.

**Recommended: Q5 (18.78 GiB).** It has the best measured quality on offer, and the next size down (Q4) gives up a real 1.75 percentage points of perplexity rather than a difference lost in noise.

## Usage

### LM Studio

1. Download the GGUF file of your preferred quantization tier
2. Place it in your LM Studio models directory
3. Load the model in LM Studio -- it will auto-detect the chat template
4. The model supports the base model's full context length

### llama.cpp

```bash
# Interactive chat (--jinja uses the model's embedded chat template, not a hardcoded one)
llama-cli -m ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf -c 8192 --jinja -cnv

# Single prompt
llama-cli -m ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf -c 8192 -p "Your prompt here"

# Server mode
llama-server -m ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf -c 8192 --port 8080 --jinja
```

### Python (llama-cpp-python)

```python
from llama_cpp import Llama

llm = Llama(model_path="./ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf", n_ctx=8192)
output = llm.create_chat_completion(
    messages=[
        {"role": "user", "content": "Hello, how are you?"}
    ]
)
print(output["choices"][0]["message"]["content"])
```

### Vision (image input)

```bash
llama-server -m ThinkingCap-Qwen3.6-27B-Q5_K_M.gguf --mmproj mmproj-ThinkingCap-Qwen3.6-27B-f16.gguf -c 8192 --port 8080 -ngl 99 -fa on
```

## Caveats

- The base model's license (apache-2.0) applies to all derivative files
- Quantization reduces precision -- verify outputs for your specific use case
- The hybrid quantization assigns different precision to different tensor groups, which means quality characteristics may differ from uniform quantizations

## Limitations

- Quantized models may exhibit subtle differences from the full-precision fine-tune
- This model inherits any limitations and biases present in the base model

---
*Generated with [MagicQuant](https://github.com/lucasmcoleman/MagicQuant)*