CompiwerAI/Mtrini-Imagine-1.0-7B-GGUF

⚡ Mtrini-Imagine-1.0-7B — GGUF

🎨 The full Mtrini-Imagine model.

🧠 Fused. Packed. Ready to experiment with.

🇲🇦 Built by Compiwer AI.

Welcome to the GGUF release of Mtrini-Imagine-1.0-7B! 🚀

This version contains the full fused model weights in GGUF format.

Unlike the lightweight LoRA release, the Mtrini-Imagine adapter has already been fused into the base model.

So instead of:

Base Model + Adapter

you get:

✨ Mtrini-Imagine — already fused

---

🚀 Model Specs

🔧	Details
🤖 Model	Mtrini-Imagine-1.0-7B
📦 Format	GGUF
🧮 Precision	BF16
🧠 Architecture	HiDream-O1 / UiT
🧪 Variant	Dev
🔢 Tensor Count	759
🧩 Fused LoRA Layers	360
💾 Size	~16.4 GB
📁 Source	SafeTensors

---

✨ What makes this different?

The original Mtrini-Imagine LoRA contained 360 LoRA layer pairs.

Those updates were natively fused into the base model before conversion.

🔥 Conversion pipeline

🧠 HiDream-O1-Image
          +
🎨 Mtrini-Imagine LoRA
          ↓
🔗 Native LoRA Fusion
          ↓
🚀 Mtrini-Imagine-1.0-7B
          ↓
📦 GGUF Conversion
          ↓
⚡ Mtrini-Imagine-1.0-7B-bf16.gguf

No separate adapter is required for the fused checkpoint. 🎉

---

📦 File

Mtrini-Imagine-1.0-7B-bf16.gguf

💾 Size

~16.4 GB

This is a BF16 GGUF, so make sure your system and runtime have enough memory/VRAM for the model and its image-generation pipeline.

---

🔐 Verified

We checked the generated GGUF before release.

✅ Verification

• 🟢 GGUF magic header: Valid
• 🟢 Tensor count: 759
• 🟢 Architecture metadata: hidream_o1
• 🟢 Variant: dev
• 🟢 Precision: BF16
• 🟢 Conversion: Successful
• 🟢 SHA256: Verified

🔑 SHA256

a73c6bc8c3acec091e2973a397251eff62b960f894b4d89c7ebdbb7364e324ad

A SHA256SUMS file is included in the repository too. 🔒

---

🧪 Runtime Status

The GGUF conversion itself was successful. ✅

We also attempted to compile the current o1.c runtime with CUDA 13 on an NVIDIA RTX PRO 6000 Blackwell system.

The build currently hits CUDA compatibility issues in the o1.c source, specifically around hd_cuda_errbuf.

Therefore:

⚠️ o1.c generation has not been officially validated for this release yet.

This is important: the failed runtime build did not modify the GGUF file.

The GGUF remains structurally valid and was verified after conversion.

---

🎨 What can it do?

Mtrini-Imagine is designed for:

🖼️ Text-to-image generation
🎨 Creative artwork
🧪 Image-generation experiments
🤖 Multimodal AI research
💻 Local inference experimentation
🔬 GGUF runtime development

---

⚠️ Limitations

Mtrini-Imagine 1.0 is an experimental release.

You may see:

✍️ Weird text
🔤 Misspelled words
🧩 Incorrect object relationships
🎨 Visual artifacts
🧠 Prompt interpretation mistakes
🔍 Inconsistent tiny details

Text rendering is currently one of the areas we'd most like to improve in future versions. 🛠️

---

🆚 Adapter vs GGUF

Not sure which one you want? 👀

	🧩 Adapter	⚡ GGUF
Size	~98 MB	~16.4 GB
LoRA	✅	✅ Fused
Base required	✅	❌
Native DiffSynth	✅	—
Standalone	❌	✅
Best for	🧑‍💻 Developers	🚀 Full-model users

🧩 Want the smallest download?

Get the Adapter.

⚡ Want the complete fused model?

Get the GGUF.

---

🛡️ Safety

Please use Mtrini-Imagine responsibly.

Do not use it to generate illegal, harmful, deceptive, or abusive content.

Users are responsible for complying with applicable laws, policies, and the license terms of the underlying model.

---

📜 License

This release is provided under Apache-2.0, subject to the licensing conditions of the underlying HiDream-O1-Image model.

Please review the original model license before commercial use.

---

❤️ Built by Compiwer AI

🇲🇦 Compiwer AI

Building AI For Everyone.

Mtrini is the model family developed by Compiwer AI.

🎨 Mtrini-Imagine — image generation
🧠 Mtrini-SVL — reasoning / multimodal experimentation
🌍 Mtrini-Tellus — language model experimentation

More coming. 🚀

---

⚡ Mtrini-Imagine-1.0-7B

🎨 Imagine it. Generate it. Build it.

Compiwer AI 🇲🇦 — Building AI For Everyone.