What changed
0 fixes1 addition0 changes0 removals
added*Adventurers and World Architects, welcome to the new era of RolemIAster!** In this titanic update, our AI-powered role-playing engine takes an unprecedented leap by integrating the powerful models of the **Qwen3.5** architecture, combined with the revolutionary **TurboQuant** memory technology. I've worked diligently and hard these past weeks to eradicate system hallucinations and push beyond the physical limits of your hardware to bring you the ultimate Game Master, running **100% locally, privately, and uncensored** on your own machine. There's still a long road ahead, but the leap is immense. Buckle up! 🚀 --- ## 🌟 Critical Engine Updates: ### 🧠 Native Intelligence (Qwen Fine-Tuned) Forged in a custom-built model trainer (**ABAFE Models Trainer**), the new models assimilate the game system and its rules into their own neural weights. Strict obedience without the need to inject heavy manuals at every turn. The AI no longer improvises: it *knows* the lore, respects the economy, and punishes your mistakes fairly. ### 🌌 Infinite Memory Campaigns (TurboQuant 3-bit) I've introduced a novel system for extreme KV Cache quantization. You can now deploy **massive contexts of over 100,000 tokens**—your AI will remember the lore, past decisions, and the "implicit gear" of each NPC without causing an Out-of-Memory collapse on your GPU. **For those with low VRAM, this is a game-changer**: if your RAM used to fall short, causing CPU dumping or direct crashes, now everything will fit in VRAM without getting so dramatic. Long campaigns no longer break mid-way! ### ⚡ Universal Performance From 8GB tactical graphics cards to colossal 32GB RTX 5090s, the game dynamically scales your context window to ensure **maximum immersion and token-per-second agility**. Your hardware doesn't matter: the adventure adapts to you, not the other way around. ## ⚠️ Developer Transparency (An Honest Read) ⚡ **AMD/Intel:** While there are technical indications that TurboQuant *might* work on AMD/Intel cards via Vulkan, in practice, there are too many bugs to enable it. For now, this technology is exclusive to NVIDIA RTX. I trust the community will work magic, and we'll have news soon—I'll be the first to integrate it when it's ready! 🧪 **Real Early Access:** This is still Early Access. The AI is now a much more competent table-top companion, but it can still stumble. If you see something odd, report it. Every bug we fix is a collective victory. --- ## 🎲 Ready to Roll the Dice? This isn't just an update: it's a **refoundation of the engine**. We're moving from a Game Master that needed to read the entire manual before every sentence, to a Role-playing Master who knows the rules by heart and applies them with surgical precision. Your legacy as a player—your decisions, your allies, your mistakes—now persists in the AI's memory without artificial limits. **Epic, tens-of-hours-long campaigns are finally possible without fear of the game forgetting who the main villain is.** *Thank you for trusting this solo project, made between coffee, code, and a lot of passion. See you at the virtual tables.* 🐉☕ **— The Developer (and sole inhabitant of this creative asylum)** 🎮✨ 👇 https://discord.gg/pKpu5H4F to share your campaigns.
RolemIAster changes
added*Adventurers and World Architects, welcome to the new era of RolemIAster!** In this titanic update, our AI-powered role-playing engine takes an unprecedented leap by integrating the powerful models of the **Qwen3.5** architecture, combined with the revolutionary **TurboQuant** memory technology. I've worked diligently and hard these past weeks to eradicate system hallucinations and push beyond the physical limits of your hardware to bring you the ultimate Game Master, running **100% locally, privately, and uncensored** on your own machine. There's still a long road ahead, but the leap is immense. Buckle up! 🚀 --- ## 🌟 Critical Engine Updates: ### 🧠 Native Intelligence (Qwen Fine-Tuned) Forged in a custom-built model trainer (**ABAFE Models Trainer**), the new models assimilate the game system and its rules into their own neural weights. Strict obedience without the need to inject heavy manuals at every turn. The AI no longer improvises: it *knows* the lore, respects the economy, and punishes your mistakes fairly. ### 🌌 Infinite Memory Campaigns (TurboQuant 3-bit) I've introduced a novel system for extreme KV Cache quantization. You can now deploy **massive contexts of over 100,000 tokens**—your AI will remember the lore, past decisions, and the "implicit gear" of each NPC without causing an Out-of-Memory collapse on your GPU. **For those with low VRAM, this is a game-changer**: if your RAM used to fall short, causing CPU dumping or direct crashes, now everything will fit in VRAM without getting so dramatic. Long campaigns no longer break mid-way! ### ⚡ Universal Performance From 8GB tactical graphics cards to colossal 32GB RTX 5090s, the game dynamically scales your context window to ensure **maximum immersion and token-per-second agility**. Your hardware doesn't matter: the adventure adapts to you, not the other way around. ## ⚠️ Developer Transparency (An Honest Read) ⚡ **AMD/Intel:** While there are technical indications that TurboQuant *might* work on AMD/Intel cards via Vulkan, in practice, there are too many bugs to enable it. For now, this technology is exclusive to NVIDIA RTX. I trust the community will work magic, and we'll have news soon—I'll be the first to integrate it when it's ready! 🧪 **Real Early Access:** This is still Early Access. The AI is now a much more competent table-top companion, but it can still stumble. If you see something odd, report it. Every bug we fix is a collective victory. --- ## 🎲 Ready to Roll the Dice? This isn't just an update: it's a **refoundation of the engine**. We're moving from a Game Master that needed to read the entire manual before every sentence, to a Role-playing Master who knows the rules by heart and applies them with surgical precision. Your legacy as a player—your decisions, your allies, your mistakes—now persists in the AI's memory without artificial limits. **Epic, tens-of-hours-long campaigns are finally possible without fear of the game forgetting who the main villain is.** *Thank you for trusting this solo project, made between coffee, code, and a lot of passion. See you at the virtual tables.* 🐉☕ **— The Developer (and sole inhabitant of this creative asylum)** 🎮✨ 👇 https://discord.gg/pKpu5H4F to share your campaigns.
[dynamiclink href="https://youtu.be/FNz1QRPKPFw"]
**Adventurers and World Architects, welcome to the new era of RolemIAster!** In this titanic update, our AI-powered role-playing engine takes an unprecedented leap by integrating the powerful models of the **Qwen3.5** architecture, combined with the revolutionary **TurboQuant** memory technology. I've worked diligently and hard these past weeks to eradicate system hallucinations and push beyond the physical limits of your hardware to bring you the ultimate Game Master, running **100% locally, privately, and uncensored** on your own machine. There's still a long road ahead, but the leap is immense. Buckle up! 🚀 --- ## 🌟 Critical Engine Updates: ### 🧠 Native Intelligence (Qwen Fine-Tuned) Forged in a custom-built model trainer (**ABAFE Models Trainer**), the new models assimilate the game system and its rules into their own neural weights. Strict obedience without the need to inject heavy manuals at every turn. The AI no longer improvises: it *knows* the lore, respects the economy, and punishes your mistakes fairly. ### 🌌 Infinite Memory Campaigns (TurboQuant 3-bit) I've introduced a novel system for extreme KV Cache quantization. You can now deploy **massive contexts of over 100,000 tokens**—your AI will remember the lore, past decisions, and the "implicit gear" of each NPC without causing an Out-of-Memory collapse on your GPU. **For those with low VRAM, this is a game-changer**: if your RAM used to fall short, causing CPU dumping or direct crashes, now everything will fit in VRAM without getting so dramatic. Long campaigns no longer break mid-way! ### ⚡ Universal Performance From 8GB tactical graphics cards to colossal 32GB RTX 5090s, the game dynamically scales your context window to ensure **maximum immersion and token-per-second agility**. Your hardware doesn't matter: the adventure adapts to you, not the other way around. ## ⚠️ Developer Transparency (An Honest Read) ⚡ **AMD/Intel:** While there are technical indications that TurboQuant *might* work on AMD/Intel cards via Vulkan, in practice, there are too many bugs to enable it. For now, this technology is exclusive to NVIDIA RTX. I trust the community will work magic, and we'll have news soon—I'll be the first to integrate it when it's ready! 🧪 **Real Early Access:** This is still Early Access. The AI is now a much more competent table-top companion, but it can still stumble. If you see something odd, report it. Every bug we fix is a collective victory. --- ## 🎲 Ready to Roll the Dice? This isn't just an update: it's a **refoundation of the engine**. We're moving from a Game Master that needed to read the entire manual before every sentence, to a Role-playing Master who knows the rules by heart and applies them with surgical precision. Your legacy as a player—your decisions, your allies, your mistakes—now persists in the AI's memory without artificial limits. **Epic, tens-of-hours-long campaigns are finally possible without fear of the game forgetting who the main villain is.** *Thank you for trusting this solo project, made between coffee, code, and a lot of passion. See you at the virtual tables.* 🐉☕ **— The Developer (and sole inhabitant of this creative asylum)** 🎮✨ 👇 https://discord.gg/pKpu5H4F to share your campaigns.
