Llama-3-Kimodo-SMPLX-RP-v1-GGUF

Llama-3-Kimodo-SMPLX-RP-v1-GGUF

Pre-converted GGUF weights and LLM2Vec text encoder bundle for kimodo.cpp, enabling native, high-performance CPU / Vulkan GPU text-to-motion generation of SMPL-X22 3D animation skeletons.

Included Files

• kimodo-smplx-rp-v1-f32.gguf (~1.13 GB): Motion diffusion denoiser and SMPL-X22 kinematic decoder in F32 GGUF format.
• llm2vec-text-bundle/ (~15 GB): Chunked Llama-3-8B + LLM2Vec bidirectional text encoder components (Tokenizer, Embedding, Final-norm, and Layers 00–31 in BF16/F32 GGUF format).

Quick Usage with kimodo.cpp

1. Download Model Files

You can download this entire repository directly using huggingface-cli or hf:

huggingface-cli download k0ta0uchi/Llama-3-Kimodo-SMPLX-RP-v1-GGUF --local-dir models/

Or download with Python:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="k0ta0uchi/Llama-3-Kimodo-SMPLX-RP-v1-GGUF",
    local_dir="./models",
)

2. Generate Motion (CLI)

# Windows
kmd-generate.exe models/kimodo-smplx-rp-v1-f32.gguf models/llm2vec-text-bundle prompt.txt 150 100 42 output_dir/

# Linux / macOS
./build/release/kmd-generate models/kimodo-smplx-rp-v1-f32.gguf models/llm2vec-text-bundle prompt.txt 150 100 42 output_dir/

3. Run Web Demo

go run ./demo -generator build/Release/kmd-generate.exe -motion-model models/kimodo-smplx-rp-v1-f32.gguf -text-bundle models/llm2vec-text-bundle -addr 0.0.0.0:8094

---

Attribution & License Information

This model package contains derivative works created from:

1. NVIDIA Kimodo-SMPLX-RP-v1
   • Licensed by NVIDIA Corporation under the NVIDIA Open Model License.
   • See LICENSE-NVIDIA.txt for details.
2. Meta Llama-3-8B-Instruct
   • Built with Meta Llama 3.
   • Meta Llama 3 is licensed under the Meta Llama 3 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved.
   • See LICENSE-LLAMA3.txt and Meta Llama 3 Acceptable Use Policy for details.
3. McGill-NLP LLM2Vec Adapters
   • Derived from McGill-NLP/LLM2Vec-Meta-Llama-3-8B-Instruct-mntp and ...-supervised.