--- license: apache-2.0 base_model: Qwen/Qwen3.8-27B tags: - text-generation - gguf - quantizer - autoround - architecture-aware - mamba - ssm - multi-token-prediction - mtp ---

⚡ Fuel the Lab: Keep the Optimization Loops Running

Every single ZeroDigest YMQ-MTP release is handcrafted and manually calibrated via intensive importance-matrix text dataset sweeps to protect critical logic pathways. This project is entirely independent research—no automation bots, no corporate backers, and no external funding.

Running multi-hour compute arrays and compiling hardware-native presets consumes massive local electricity and infrastructure overhead out-of-pocket. If these optimized files are saving your local developer workflow, consider dropping a support contribution to fuel the next batch of architecture-aware releases!

👉 Premium Support Paths:

☕ Support the Lab on Ko-fi (One-Time / Monthly)
🧡 BTC (Bitcoin Network): 14Fmic9z3VA1ZU11bWoP6JtU7csTAApwZo
🔷 ETH / USDT (ERC20 Network): 0xbe4cdc3adc27c21ef1c27c6b403311db07b35ed2

# Qwen3.8-27B-YMQ-MTP-GGUF > **Source Model:** [Qwen/Qwen3.8-27B](https://huggingface.co/Qwen/Qwen3.8-27B) ### ⚖️ An Architecture-Aware, AutoRound-Inspired Mixed Precision Layout This repository features advanced, custom architecture-aware quantizations of **Qwen3.8-27B** 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 | | :--- | :--- | :--- | :--- | | **`XXS`** | [~9.8 GB](./Qwen3.8-27B-YMQ-XXS.gguf) | Absolute VRAM Squeeze / 12GB Card Lifeline | Massive structural quantization noise. Best restricted to low-context, single-turn instructions. Fits 12GB cards with context cache breathing room. | | **`XS-TI`** | [~10.2 GB](./Qwen3.8-27B-YMQ-XS-TI.gguf) | ⚡ **12GB Card Lifeline (High-Context)** | **The 12GB Context Champion.** Safely pins core attention layers to a stable `IQ3_XXS` gradient while crushing non-critical auxiliary arrays to `IQ2_XS`. | | **`M-TI`** | [~12.6 GB](./Qwen3.8-27B-YMQ-M-TI.gguf) | 💎 **Premium 16GB GPU Workspace Driver** | **The 16GB Workstation Choice.** Employs a robust `Q5_K/IQ4_XS` mixed-precision gradient that protects logical reasoning focus while leaving over 3.4GB of VRAM wide open. | | **`M`** | [~14.5 GB](./Qwen3.8-27B-YMQ-M.gguf) | **The Ultimate Coding Sweet Spot (Recommended)** | **Elite logical stability.** Complete logic clarity. It crushes standard industry 4-bit alternatives. | | **`L`** | [~17.0 GB](./Qwen3.8-27B-YMQ-L.gguf) | Premium Single-GPU Processing / Heavy workloads | Near-lossless instruction formatting. Pristine multi-turn architecture safety. | | **`XL`** | [~19.0 GB](./Qwen3.8-27B-YMQ-XL.gguf) | Maximum VRAM Fill / No Compromises | Mathematical saturation ceiling. Full precision logic tracks for massive multi-file codebase operations. | ## 📉 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 ➔ Floor) | | :--- | :--- | :--- | :--- | :--- | | **`XXS`** | [~9.8 GB](./Qwen3.8-27B-YMQ-XXS.gguf) | 7.7848 | `0.193804 ± 0.0022` | `IQ3_XXS` ➔ `IQ2_S` ➔ `IQ2_XS` ➔ `IQ2_XS` *(No Floor)* | | **`XS-TI`** | [~10.2 GB](./Qwen3.8-27B-YMQ-XS-TI.gguf) | 7.6090 | `0.168378 ± 0.001935` | `IQ3_XXS` ➔ `IQ3_XXS` ➔ `IQ3_XXS` ➔ `IQ3_XXS` ➔ `IQ2_XXS` | | **`M-TI`** | [~12.9 GB](./Qwen3.8-27B-YMQ-M-TI.gguf) | **7.0109** | `0.104087 ± 0.0016` | `Q5_K` ➔ `IQ4_XS` ➔ `IQ3_S` ➔ `IQ3_XXS` ➔ `IQ2_XXS` | | **`M`** | [~14.5 GB](./Qwen3.8-27B-YMQ-M.gguf) | **6.8413** | `0.053286 ± 0.0013` | `Q5_K` ➔ `IQ4_XS` ➔ `IQ3_S` ➔ `IQ3_XXS` *(No Floor)* | | **`L`** | [~17.0 GB](./Qwen3.8-27B-YMQ-L.gguf) | 6.9791 | `0.031546 ± 0.0009` | `Q6_K` ➔ `Q5_K` ➔ `IQ4_NL` ➔ `IQ3_S` *(No Floor)* | | **`XL`** | [~19.0 GB](./Qwen3.8-27B-YMQ-XL.gguf) | 6.8196 | `0.011598 ± 0.0005` | `Q6_K` ➔ `Q6_K` ➔ `Q5_K` ➔ `IQ4_NL` *(No Floor)* | ### 💡 The Core Architectural Discovery Notice the dramatic performance leap between the `S` and `M` presets. The **YMQ-Compiler** log-space algorithm automatically detects the true data signals on the newly updated Qwen 3.8 hybrid Attention/Mamba routing nodes. By shifting the quantization boundaries slightly in the **`M` preset**, the engine safely promotes the model's high-leverage logical spikes straight into full high-fidelity precision layers. This drops the perplexity score down to an elite **6.8413**—matching the raw reasoning power of the massive 19GB `XL` file while clawing back a clean **5 Gigabytes of VRAM overhead cache space** for your local agent environments! --- ## ⚖️ 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 single-GPU 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: 1. **Log-Space Gap Detection Clustering**: Instead of flat percentage thresholds, the engine computes statistical cluster variances in log-space, successfully isolating intermediate logical reasoning spikes and elevating them to stable non-linear 4-bit (`IQ4_XS`) formats, while compressing idle fact-storage layers to aggressive 2-bit baselines. 2. **Fading Boundary Tapering**: Recognizes the extreme fragility of initial token entry data vectors, forcing an input wave cushion (`L00=IQ4_NL` → `L01=IQ4_XS` → `L02=IQ3_XXS`) that gradually stabilizes parameters before hitting the fallback pools. 3. **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. 4. **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. 5. **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) ``` $./llama-server -m models/Qwen3.8-27B-YMQ-M.gguf -ctk q8_0 -ctv q4_0 --ctx-size 245760 --mmproj models/Qwen3.8-27B.mmproj-Q6_K.gguf \ --spec-type draft-mtp --spec-draft-n-max 2 --timeout 36000 --checkpoint-min-step 2048 --ctx-checkpoints 4 \ --n-predict -1 --temp 0.6 --top-p 0.95 --top-k 20 --repeat-penalty 1.05 --jinja -fa ``` ### 🖼️ Vision Projection (`--mmproj`) For multimodal vision support, pair these builds with one of the following projection files: | Variant | File | Size | Notes | | :--- | :--- | :--- | :--- | | **Full Precision (F16)** | [Qwen3.8-27B.mmproj-f16.gguf](https://huggingface.co/mradermacher/Qwen3.8-27B-GGUF/blob/main/Qwen3.8-27B.mmproj-f16.gguf) | ~928 GB | Full-precision native vision tower. Maximum fidelity for image reasoning tasks. | | **Q6_K** | [Qwen3.8-27B.mmproj-Q6_K.gguf](./Qwen3.8-27B.mmproj-Q6_K.gguf) (this repo) | ~587 MB | High-fidelity quantized vision tower. Excellent quality-to-size balance with minimal perceptible degradation. | | **Q4_K_S** | [Qwen3.8-27B.mmproj-Q4_K_S.gguf](./Qwen3.8-27B.mmproj-Q4_K_S.gguf) (this repo) | ~478 MB | Compact vision projection for VRAM-constrained setups. Retains strong image understanding at reduced footprint. | Pass via `--mmproj ` in your llama-server invocation (see example above). --- ## ☕ 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](https://ko-fi.com/zerodigest)** --- ## 📦 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: 👉 **[GitHub: ZeroDigest / YMQ-Compiler](https://github.com/minyor/ymq-compiler)**