ramgpt/Qwen3.8-27B-Human-KO-Enterprise-v0.2-GGUF overview
Qwen3.8 27B Human KO Enterprise v0.2 GGUF GGUF conversion of ThakiCloud/Qwen3.8 27B Human KO Enterprise v0.2 . Files Qwen3.8 27B Human KO Enterprise v0.2 Q4 K …
Runs locally from ~15.66 GB disk (16 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| Qwen3.8-27B-Human-KO-Enterprise-v0.2-Q4_K_M.gguf | GGUF | Q4_K_M | 15.66 GB | Download |
Model Details
Model README
---
base_model: ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.2
tags:
- gguf
- llama.cpp
---
Qwen3.8-27B-Human-KO-Enterprise-v0.2-GGUF
GGUF conversion of ThakiCloud/Qwen3.8-27B-Human-KO-Enterprise-v0.2.
Files
Qwen3.8-27B-Human-KO-Enterprise-v0.2-Q4_K_M.gguf— Q4_K_M — 15.66 GiB
SHA256: a34a57444c548172b72f46d7a9320b6230627812df74a6e2efede71286e3420d
Conversion scope
The source checkpoint is multimodal, but this GGUF is a text-only conversion. Vision/image weights and an mmproj artifact are not included. Do not treat this file as preserving the source model's multimodal capability.
Conversion note
No MTP/NextN mismatch was detected by the preflight checks. GGUF preserves the source MTP/NextN tensors. In the tested standard llama.cpp autoregressive server path, blk.64 / nextn tensors were logged as unused, so the smoke results validate the main 64-layer generation path rather than an MTP speculative path.
Validation gate
Before upload this exact file passed a real llama-server /v1/chat/completions smoke gate using the source chat template path. The gate checks:
- model loads successfully
- system/user chat template separation
- exact-output compliance for strict smoke prompts
- generation ends with finish_reason=stop instead of hitting the token cap
- no pseudo-role continuation such as /user or /assistant
- basic arithmetic generation
- no raw chat control-token leakage
- no thinking-tag leakage when reasoning is disabled
Reasoning mode used for smoke: auto.
Korean enterprise-action JSON smoke
Passed 3/3 strict JSON action cases. All tested generations ended normally: True. The cases exercised CALL_TOOL, ASK_CLARIFY, and REFUSE decisions with exact single-object JSON output.
Source
Source revision used for conversion: d03c8ead02103d6c1b737dfd384ecb9031ca620c.
Source HEAD checked immediately before publish: d03c8ead02103d6c1b737dfd384ecb9031ca620c.
If these revisions differ while the source safetensors fingerprints are unchanged, the difference is metadata-only and publishing remains allowed.
All model credit belongs to the original model authors.
Run ramgpt/Qwen3.8-27B-Human-KO-Enterprise-v0.2-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models