| Base model | Qwen/Qwen3.8-27B (Qwen3_5ForConditionalGeneration, 27B Dense Hybrid Gated DeltaNet + Full Softmax Attention + Multimodal Vision + MTP) |
| Previous public release | kataguru/Industry-Coder-27B-v2.0-W4A16-AWQ (Σ15=68.73 %) |
| Kataguru v3.0 modifications | Triple-Adaptive Multi-Dynamic Pipeline (v4.1): SOMA/ARA residual orthogonalization + 3-Phase Split-Subspace Orthonormal QLoRA (r=16 qr_frozen_a, indices [0:8] vs. [8:16]) + Phase 4 Multi-Dynamic Thinking, Competitive Coding & Parallel Tool-Calling annealing |
| Finnish & English mastery | Native Bilingual (fi & en): Kotus-verified Finnish inflection and technical compound terminology alongside English |
| Quantization architecture | Universal W4A16-AWQ (compressed-tensors, pack-quantized INT4, group_size=32, MSE observer) for NVIDIA Ampere (RTX 3090 / A100), Ada Lovelace (RTX 4090 / L40S), Hopper (H100), and Blackwell (RTX 5090).• Preserved in full BF16: linear_attn.in_proj_a/b, model.visual.* (333 tensors), mtp.* (15 tensors), lm_head, embed_tokens |
| Disk & VRAM footprint | 28.2 GB on disk (~14.1 GB/GPU on TP=2, or fits on a single 32 GB GPU with fp8 KV cache) |
| What was NOT changed | Base 64-layer architecture, 27-layer multimodal vision encoder (333 BF16 tensors), Multi-Token Prediction (mtp.*, 15 BF16 tensors), native 262,144 context window |
| Primary use case | Autonomous coding agents (IDE/CLI), competitive algorithms, parallel <tool_call> orchestration, graduate-level STEM, and industrial automation (Siemens SCL, Beckhoff TwinCAT 3 ST, CNC G-code) |
| Recommended runtime | vLLM (TP=2 on 2x 24–32 GB GPUs or TP=1 on 1x 32–48 GB GPU, CUDAGraphs enabled) |