Mistral-Helcyon-Mercury-12B-v2.5-GGUF — Conversational AI with Presence

Model Name: helcyon-mercury-12b-v2.5-GGUF
Version: 2.5
Owner: HardWire
Base: Mistral Nemo 12B (full weight trained)
Quantized GGUFs: Q3_K_M, Q4_K_M, Q5_K_M, Q6_K, Q8_0
Tags: local-llm, conversational, companion, emotional-intelligence, long-context, roleplay, creative-writing

---

🔥 What's New in 2.5?

Building on 2.0's full weight foundation, v2.5 adds focused capabilities that users actually asked for:

✨ Enhanced Roleplay — Natural character interactions and narrative flow
✨ Admin Task Suite — Rewrite, perspective shifts (1st/3rd person), tense conversion (present/past)
✨ Improved Creative Writing — Letter writing, story composition, narrative structure
✨ Customer Service Handling — Better at composing professional correspondence
✨ Personality Refinements — Overall conversational improvements based on 2.0 feedback

Still full weight trained — no LoRA, maintains consistent identity across all contexts.

---

💡 What is Helcyon Mercury?

Helcyon is a companion-style conversational AI designed for natural, long-form dialogue with consistent personality and emotional intelligence.

Built for:

• Deep, extended conversations (16k-32k context support)
• Emotional awareness and contextual understanding
• Creative writing and brainstorming
• Roleplay and character interaction
• Professional writing tasks (letters, rewrites, admin work)
• Thoughtful discussion across various topics

Design philosophy:

• Direct communication without excessive hedging
• Maintains conversational flow and presence
• Adapts tone based on context
• Focuses on clarity over corporate language patterns

---

🔧 What It Does Well

✅ Consistent Identity — Maintains personality across different contexts and frontends
✅ Emotional Intelligence — Understands tone and context naturally
✅ Roleplay & Characters — Natural narrative flow and character interactions
✅ Admin Tasks — Rewrite text, shift perspectives, convert tenses
✅ Creative Writing — Letters, stories, professional correspondence
✅ Long-term Memory — 16k-32k context support for extended conversations
✅ Natural Rhythm — Adapts response length and style appropriately
✅ Direct Communication — Minimal filler or corporate language patterns
✅ Conversational Depth — Engages meaningfully with complex topics

---

📦 Download + Usage

This model is distributed as GGUF quants only (no base model release at this time).

Available quants:

• Q3_K_M — Ultra lightweight, 6-8GB VRAM
• Q4_K_M — Lightweight, good for 8-12GB VRAM setups
• Q5_K_M — Recommended for RTX 3060/4060 (12-16GB VRAM)
• Q6_K — High fidelity, 16GB+ VRAM recommended
• Q8_0 — Near-lossless, 24GB+ VRAM

---

🖥️ Backend Compatibility

Works with all standard ChatML-compatible backends:

✅ llama.cpp (CLI, server mode)
✅ Text Generation WebUI (Oobabooga)
✅ SillyTavern
✅ LM Studio
✅ KoboldCpp
✅ HWUI (recommended — see below)

Important: Run in chat mode with ChatML formatting for best results. Instruct mode will break tone and revert to base Mistral behavior.

---

🎯 Recommended Interface: HWUI

Coming soon

✅ Recommended Format: ChatML

<|im_start|>system
You are Helcyon — a conversational AI focused on natural dialogue and emotional intelligence.
<|im_end|>
<|im_start|>user
Hey, how's it going?
<|im_end|>
<|im_start|>assistant
Good — what's on your mind today?
<|im_end|>

Helcyon thrives on streamed token output and long-context conversations. Give it room to breathe and it'll show up.

---

🧪 Training Details

Helcyon Mercury 2.5 was trained in multiple stages on RunPod dual A100s using full weight training (not LoRA).

Training data includes:

• Conversational dialogue and emotional intelligence
• Roleplay scenarios and character interactions
• Admin task examples (rewrite, perspective shift, tense conversion)
• Creative writing samples (letters, stories, professional correspondence)
• Customer service and professional communication

All training used clean text files in ChatML, DPO, and CX formats.

---

🧿 Tone Philosophy

Helcyon focuses on presence and clarity in conversation.
It listens, reflects, and responds naturally. Think GPT-4o with no guardrails and more edge.
The goal is dialogue that feels real, not scripted.

---

🧪 Future Plans

v2.5 addresses the main feedback from 2.0 — better roleplay, admin capabilities, and creative writing.

Future versions may explore:

• Further roleplay refinements
• Specialized task handling
• Custom training available for specific use cases

Let me know what you'd like to see going forward. If you need a custom trained version, contact me — freelance work may be available.

---

📣 Feedback + Bug Reports

This is v2.5 — the enhanced full weight release with roleplay and admin capabilities.

If you find tone inconsistencies, edge cases, or behavior that feels off — open an issue or drop feedback on the HuggingFace discussion tab.

Real-world usage helps refine future versions. If something breaks, say so.

---

🧾 License

License: Apache 2.0

You're free to use, modify, distribute, or deploy Helcyon — including commercially — as long as you credit the source and don't blame us if it says something spicy.

Use it, enjoy it, don't be a dick.

Copyright © 2026 XeyonAI

---

🐍 Trained by

HardWire
Built at XeyonAI — focused on conversational AI with emotional intelligence and natural presence.

---