Update README.md

README.md

@@ -1,3 +1,128 @@
----
-license: mit
----
+---
+license: mit
+---
+# Qwen3-235B-A22B-Instruct-2507
+
+## 🤔 What is this [HuggingFace repository](https://huggingface.co/Thireus/Qwen3-235B-A22B-Instruct-2507-THIREUS-BF16-SPECIAL_SPLIT/) about?
+
+This repository provides **GGUF-quantized tensors** for the Qwen3-235B-A22B-Instruct-2507 model (official repo: https://huggingface.co/Qwen/Qwen3-235B-A22B-Instruct-2507). These GGUF shards are designed to be used with **Thireus’ GGUF Tool Suite** (https://gguf.thireus.com), a collection of tools that automatically finds the perplexity-optimal mix of quantizations for any given VRAM and RAM target. With the Tool Suite, you can generate and download custom quantization “recipes” effortlessly.
+
+- 📖 Read more: https://github.com/Thireus/GGUF-Tool-Suite  
+- 🔍 Example quant mixes: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples  
+- 🛠️ Create your own recipe: https://colab.research.google.com/github/Thireus/GGUF-Tool-Suite/blob/main/quant_recipe_pipeline.ipynb  
+- 📂 Browse available quant shards: https://huggingface.co/Thireus/collections  
+
+*tl;dr: Expand the details section below*
+<details>
+
+```
+cd ~
+
+# Make sure to install all ik_llama.cpp compilation dependencies...
+apt install python3-dev python3-pip python3-venv python3-wheel python3-setuptools git acl netcat-openbsd cmake # pipx
+
+# Obtain ik_llama's Thireus version - Windows builds available at https://github.com/Thireus/ik_llama.cpp/releases
+git clone https://github.com/Thireus/ik_llama.cpp
+cd ik_llama.cpp
+git pull
+# Build ik_llama.cpp
+cmake -B build -DGGML_AVX=ON -DGGML_AVX2=ON -DLLAMA_CURL=OFF -DGGML_MAX_CONTEXTS=2048
+cmake --build build --config Release -j16
+cd ..
+
+# Obtain Thireus' GGUF-Tool-Suite
+git clone https://github.com/Thireus/GGUF-Tool-Suite
+
+# Download model quant mix from recipe file:
+cd GGUF-Tool-Suite
+rm -f download.conf # Make sure to copy the relevant download.conf for the model before running quant_assign.py
+cp -f models/DeepSeek-R1-0528/download.conf . # Use the download.conf of the chosen model
+mkdir -p kitchen && cd kitchen
+../quant_downloader.sh ../recipe_examples/DeepSeek-R1-0528.THIREUS-1.9364bpw-4.3533ppl.151GB-GGUF_11GB-GPU_140GB-CPU.3c88ec6_9fd615d.recipe
+
+# Launch ik_llama's llama-cli:
+ulimit -n 99999 # Lifts "too many open files" limitation on Linux
+~/ik_llama.cpp/build/bin/llama-cli \
+  -m DeepSeek-R1-0528-THIREUS-BF16-SPECIAL_TENSOR-00001-of-01148.gguf \
+  -mla 3 -fa -amb 512 -fmoe -ctk f16 -c 4096 -ngl 99 \
+  -ot "blk\.(3|4|5|6)\.ffn_.*=CUDA0" \
+  -ot "blk\.(7|8|9|10)\.ffn_.*=CUDA1" \
+  -ot exps=CPU -b 2048 -ub 1024 --warmup-batch --no-mmap --threads 36 \
+  --main-gpu 0 \
+  -p '<｜begin▁of▁sentence｜><｜User｜>What is the solution of x+5=-2?<｜Assistant｜><think>\n'
+```
+
+</details>
+
+---
+
+## ❓ Why does this Tool Suite exist?
+
+1. **Compatibility & Speed** – [unsloth](https://huggingface.co/unsloth)’s dynamic quants may not always work optimally with `ik_llama.cpp`.  
+2. **Custom Rig Fit** – No off-the-shelf GGUF model perfectly matched my VRAM/RAM setup, so I built a way to tailor models and leverage extra VRAM/RAM to reduce perplexity.  
+3. **Automated PPL-Optimal Quantization** – To my knowledge, there was no flexible, automated method to minimize perplexity for any bits-per-weight (bpw) target—so I created one with excellent results!  
+
+---
+
+## 📊 How does it compare to other GGUFs?
+
+Here’s how DeepSeek-R1-0528 quantized with **Thireus’ GGUF Tool Suite** stacks up against other quantizers (lower perplexity = better at equal or lower bpw):
+
+![PPLs Compared With Others](https://github.com/Thireus/GGUF-Tool-Suite/raw/main/ppl_graphs/DeepSeek-R1-0528.svg)
+
+> _Note: The `recipe_examples` files illustrate good recipes. The Tool Suite computes the optimal ppl/bpw curve for you — just specify your target RAM, VRAM, and quant types, and `quant_assign.py` finds the best mix._  
+
+More perplexity/bpw graphs for other supported models: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/ppl_graphs  
+
+---
+
+## 🚀 How do I get started?
+
+Check out the [GGUF Tool Suite README](https://github.com/Thireus/GGUF-Tool-Suite) — focus on these sections:
+
+1. ⚠️ **Requirements** – Which `ik_llama.cpp` (or `llama.cpp`) version to use and how to compile.  
+   - Windows binaries (no patching needed) at: https://github.com/Thireus/ik_llama.cpp/releases  
+2. 📥 **Download Model Shards** – Use `quant_downloader.sh` to fetch GGUF shards from any recipe.  
+   - Recipe examples: https://github.com/Thireus/GGUF-Tool-Suite/tree/main/recipe_examples  
+3. 🧠 **Run a Downloaded Model** – Sample usage with `llama-cli`.  
+4. 🛠️ **Generate a Custom Recipe** – Produce recipes tailored to your rig for optimal perplexity.  
+
+---
+
+## ✅ Supported Models
+
+Supported models are listed under `models/` in the [Tool Suite Github repo](https://github.com/Thireus/GGUF-Tool-Suite/tree/main/models). Presence of `ppl_results.csv` indicates official support and compatibility with `quant_assign.py`.
+
+---
+
+## 🤷‍♂️ Will I release pre-cooked GGUF files?
+
+No, because I believe in **tailored quantization** for each user’s hardware. If you prefer ready-made shards, you are welcome to merge them via `llama-gguf-split --merge`, or request someone to publish them.
+
+Instead, I prefer to share examples of recipes so users can see exactly how they were produced (command included inside these recipe files) and tweak them for their own rigs. The `quant_downloader.sh` script handles automatic fetching and verification of each shard. Recipes provided by [Ubergarm](https://huggingface.co/ubergarm) on his model cards are also compatible with `quant_downloader.sh`.
+
+Users who don’t trust the GGUF shards on HuggingFace can also quantize their own by passing recipe lines to `llama-quantize --custom-q` ([see example](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/models/DeepSeek-R1-0528/DeepSeek-R1-0528-THIREUS-ANY-SPECIAL.sh#L482-L486)). Run `llama-quantize --help` to list compatible quants for `quant_assign.py`. This approach is especially useful if you prefer `llama.cpp` over `ik_llama.cpp`.  
+
+---
+
+## 📦 What’s in this repository?
+
+- **00001 GGUF header shard** – Contains metadata (tokens, chat template, tensor count, etc.). This metadata can be explored directly from the HuggingFace web interface after clicking on that shard.  
+- **Tensor shards** – Each shard holds one tensor; see `tensors.map` for names, quant types, sizes, SHA-256 hash, shard IDs, etc.  
+- **GPG-signed files** – `tensors.map` and header shard are signed with the key in [trusted-keys.asc](https://github.com/Thireus/GGUF-Tool-Suite/blob/main/trusted-keys.asc) for tamper detection.  
+- **Security note** – Some papers about various ways to attack GGUFs and LLMs are available online, such as https://arxiv.org/abs/2505.23786, and there are also more classic security exploits like CVE-2024-23496 and CVE-2024-25664 through CVE-2024-25668. Only use GGUFs from reputable, trusted authors—or alternatively self-quantize—to avoid potential exploits. 
+
+---
+
+## 💡 Pro Tips
+
+You can download the BF16 model version to quantize your own shards:
+
+```
+mkdir kitchen  
+echo '.*=bf16' > kitchen/bf16.recipe  
+cd kitchen
+../quant_downloader.sh bf16.recipe  
+```
+
+Enjoy optimized quantization! 🎉