⚡ Fuel the Lab: Keep the Optimization Loops Running
Every single ZeroDigest YMQ-MTP release is handcrafted and manually calibrated via intensive importance-matrix 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 consumes massive local infrastructure overhead out-of-pocket. Consider checking out our compiler or supporting our compute costs!
👉 Developer Resources & Support Paths:
🛠️ View Compiler Source on GitHub (YMQ v2.0) 🚀 Deploy weights on RunPod Cloud Compute (Affiliate Link) ☕ Support the Lab on Ko-fi (One-Time / Monthly) 🧡 BTC:14Fmic9z3VA1ZU11bWoP6JtU7csTAApwZo
🔷 ETH/USDT: 0xbe4cdc3adc27c21ef1c27c6b403311db07b35ed2
Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-YMQ-GGUF
Source Model: DavidAU/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1
⚖️ An Architecture-Aware, AutoRound-Inspired Mixed Precision Layout
This repository features advanced, custom architecture-aware quantizations of Qwen3.8-27B Cold-Fusion GAIN V1.1 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 / VRAM Profile | Cognitive Real-World Coding Quality |
|---|---|---|---|
| XXS-Pro | ~9.1 GB | ⚠️ Experimental Low-VRAM Sandbox | The Ultra-Compact Frontier. |
| XS-TI | ~9.8 GB | ⚡ 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. |
| XS-Pro | ~10.6 GB | ⚡ Dedicated 12GB VRAM Champion | The 12GB Efficiency Miracle. Features a highly optimized IQ4_XS/IQ3_XS gradient that keeps the model running fully inside VRAM with a clean 1.6GB context headroom buffer. |
| S-Pro | ~12.5 GB | 💎 Premium 16GB Workstation Driver | The Efficiency Miracle. Holds an elite 6.50 PPL for pristine conversational fluidness while using full Q5_K attention armor. |
| M-TI | ~12.6 GB | 💎 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 | 🎯 High-Context Agent Workspace (Recommended) | The Flagship Masterpiece. Heavy structural shielding for maximum logical durability across massive enterprise repositories. |
| L-TI | ~14.1 GB | 🔥 Heavy 16GB Workstation Choice (Precision) | The 16GB High-Fidelity Champion. Features a robust Q6_K/IQ4_NL armored peak [local]. Minimizes information entropy drift to keep structural logic stable over long, multi-turn agent execution loops. |
🎯 Quick Selection Guide
| Use Case | Recommended Preset | Why |
|---|---|---|
| 12GB GPU | XS-TI or XS-Pro |
Best 12GB card performance with high-context support |
| 16GB GPU (Recommended) | M-TI or S-Pro |
The sweet spot for coding tasks with optimal VRAM headroom |
| 24GB GPU | L-TI or M |
~14GB models fit well on 24GB VRAM with headroom for context |
| Low-VRAM / Experimental | XXS-Pro |
Sub-10GB footprint for testing and lightweight workflows |
📉 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-Pro |
~9.1 GB | 7.2336 | 0.185361 ± 0.0021 |
IQ3_XXS ➔ IQ3_XXS ➔ IQ2_S ➔ IQ2_S ➔ IQ2_XXS |
XS-TI |
~9.8 GB | 7.0068 | 0.168378 ± 0.001935 |
IQ3_XXS ➔ IQ3_XXS ➔ IQ3_XXS ➔ IQ3_XXS ➔ IQ2_XXS |
XS-Pro |
~10.6 GB | 6.7987 | 0.135866 ± 0.0018 |
IQ4_XS ➔ IQ3_XXS ➔ IQ3_XXS ➔ IQ3_XXS ➔ IQ2_XXS |
S-Pro |
~12.5 GB | 6.5087 | 0.080403 ± 0.0014 |
Q5_K ➔ IQ3_S ➔ IQ3_XXS ➔ IQ3_XXS (No Floor) |
M-TI |
~12.9 GB | 6.6290 | 0.104087 ± 0.0016 |
Q5_K ➔ IQ4_XS ➔ IQ3_S ➔ IQ3_XXS ➔ IQ2_XXS |
M |
~14.5 GB | 6.3663 | 0.053286 ± 0.0013 |
Q5_K ➔ IQ4_XS ➔ IQ3_S ➔ IQ3_XXS (No Floor) |
L-TI |
~14.1 GB | 6.5786 | 0.086571 ± 0.0013 |
Q6_K ➔ IQ4_NL ➔ IQ4_XS ➔ IQ3_S ➔ IQ2_XS |
💡 The Multi-Tier Grid Breakthrough: Standard S vs. S-Pro
During intensive local workspace validation passes, our architecture-aware YMQ-Compiler successfully mapped out a radical new bit-allocation matrix. By splitting the layer distribution, we created a premium, high-fidelity alternative to our standard budget tier:
S-ProPreset (~12.5 GB): Perplexity = 6.5087. By aggressively compressing auxiliary tensor lanes down toQ2_Kbut raising the background baseline floor toIQ3_XXS, S-Pro eliminates a massive wave of background quantization noise while only adding a few megabytes of file weight.
The practical result is a premium, low-overhead everyday driver for 16GB GPU setups. Backed by full Q5_K reasoning armor.
⚖️ 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:
- 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. - 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. - 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/Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-YMQ-M.gguf -ctk q8_0 -ctv q4_0 --ctx-size 245760 --mmproj models/mmproj-Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-Q6_K.gguf \
--spec-type draft-mtp --spec-draft-n-max 3 --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) | mmproj-F16.gguf | ~928 MB | Full-precision vision tower. Maximum fidelity for image reasoning tasks. |
| Q6_K | mmproj-Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-Q6_K.gguf (this repo) | ~587 MB | High-fidelity quantized vision tower. Excellent quality-to-size balance with minimal perceptible degradation. |
| Q4_K_S | mmproj-Qwen3.8-27B-Cold-Fusion-GAIN-V1.1-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 <path-to-file> 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
📦 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: