Qwen3.5-88B-YMQ-M-GGUF
Source Model: 0xSero/Qwen3.5-88B
⚖️ An Architecture-Aware, AutoRound-Inspired Mixed Precision Layout
This repository features an advanced, custom architecture-aware quantization of Qwen3.5-88B (Mixture-of-Experts) processed directly from official raw BF16 source files using the custom YMQ-Compiler (v2.0) log-space framework.
These builds natively support parallel multi-token prediction (MTP) speculation engines and utilize high-context optimization parameters tailored for demanding code development API execution environments (such as RooCode/Aider).
📊 Quantization Preset Tier Details
| Preset Tier | Total Size | Target Usage / Memory VRAM Profile | Cognitive Real-World Coding Quality |
|---|---|---|---|
M | ~37 GB | 💎 Multi-GPU / 48GB Workstation Driver (Recommended) | The MoE High-Context Sweet Spot. A IQ3_S attention & shared-expert armor shields every routing decision, while the massive background expert pool is flattened to a dense IQ2_XXS grid. Elite logical stability that fits comfortably onto 48GB cards or dual-GPU arrays. |
ℹ️ Why so large? Qwen3.5-88B is a sparse Mixture-of-Experts architecture. Even at aggressive 2-bit expert floors, the total expert pool keeps the file around ~37 GB. The active parameter count per token remains small, so inference speed stays comparable to much smaller dense models once loaded.
📉 Perplexity Evaluation Metrics (WikiText-2)
The following metrics demonstrate the mathematical quality preservation of the YMQ-Compiler log-space cluster analysis compared to standard linear quantization layouts. Tested natively via llama-perplexity over a 4096 context window using the official WikiText-2 test corpus.
| Model Variant | File Size | Perplexity | Mean KL-Divergence | Internal Bit Gradient (High ➔ Mid ➔ Low ➔ Default) |
|---|---|---|---|---|
M | ~37 GB | TBD | TBD | IQ3_S ➔ IQ3_XXS ➔ IQ2_XXS ➔ IQ2_XS |
⏳ Perplexity & KL-Divergence figures are pending the final evaluation sweep on the build server. Run
ymq-ppl.sh/ymq-kld.shagainst this GGUF and populate the table above.
💡 The MoE Attention-Armor Breakthrough
Standard quantization pipelines apply a blunt, uniform bit-depth across every layer, collapsing the delicate routing structures of large sparse MoE models. By keeping the Multi-Head Attention (Q/K/V/Output/Gate) and shared experts locked behind IQ3_S high-fidelity shields—while only crushing the rarely-active background expert pool to IQ2_XXS—the YMQ-Compiler preserves the model's routing clarity at a fraction of the dense-equivalent footprint.
⚖️ YMQ vs. Uniform Quantization (The AutoRound Philosophy)
Standard quantization pipelines apply a blunt, uniform bit-depth across every single layer in a model. This wastes valuable VRAM on silent background layers while starving critical logic anchors of necessary precision.
The YMQ-Compiler implements a philosophy similar to advanced weight-tuning frameworks like Intel's AutoRound:
- Targeted Bit Allocation: It strips bits away from low-leverage background tensors and automatically re-allocates that saved VRAM budget straight into full high-fidelity shields for the model's highest cognitive spikes and boundary pathways.
- Instant Optimization: Instead of running heavy, days-long optimization training loops, YMQ achieves a highly accurate mixed-precision layout instantly by analyzing layer importance metrics in log-space.
The result is a custom mixed-precision portfolio that matches the low perplexity and high context stability of premium optimized quants (like AutoRound), while maintaining an ultra-lightweight, high-speed cache footprint.
🛠️ The YMQ Compilation Architecture
Standard quantization pipelines treat network tensors like a flat dataset, applying destructive blanket low-bit compression to delicate tracking networks. The YMQ-Compiler solves high-context logic decay by parsing model files dynamically via an automated, multi-tiered protection matrix:
- Log-Space Gap Detection Clustering: Instead of flat percentage thresholds, the engine computes statistical cluster variances in log-space, isolating intermediate logical reasoning spikes and elevating them to stable non-linear formats, while compressing idle fact-storage expert layers to aggressive 2-bit baselines.
- Fading Boundary Tapering: Recognizes the extreme fragility of initial token entry data vectors, forcing an input wave cushion that gradually stabilizes parameters before hitting the fallback pools.
- Dedicated Gate Insulation: Hard-shields volatile parallel Transformer Multi-Head Attention and Mamba Linear State Space Model (SSM) routing paths, keeping context tracking perfectly noise-free.
- Asymmetric Vocabulary Shielding: Fixes tied-weight boundary errors by mapping the final logit classification exit heads to robust configurations to completely eliminate formatting loops and API tag leakage under deep contexts.
- Native Next-N Speculative Stripping: Processed with advanced pre-tokenizer stripping to ensure zero index offset drift or layer-shifting risks across hybrid configurations.
🚀 Recommended Runtime Parameters (llama.cpp / llama-server)
Need to scale up? Deploy this exact script on on-demand cloud GPUs via RunPod Cloud Compute.
$./llama-server -m models/Qwen3.5-88B-YMQ-M.gguf -ctk q8_0 -ctv q4_0 --ctx-size 131072 \
--n-predict -1 --temp 0.6 --top-p 0.95 --top-k 20 --repeat-penalty 1.05 --jinja -fa
☕ Support & Future R&D
If the YMQ-Compiler builds saved your context window from collapsing or optimized your active development cycle speeds, consider buying a coffee to fund further low-level optimization research. Your support keeps the server nodes baking future model scales!
👉 Support ZeroDigest Research on ko-fi
📦 Source Framework & Automation Code
The compiler pipeline automation engine, setup thresholds, and structural mapping rules are open-source. To view the implementation details or compile your own custom models natively using this profile layout, visit the official development hub: