docs: Add model download instructions to README

[sammcj]

Added instructions to download the model for the tutorial.

[daaain]

Thanks for this, I was wondering how to do it!

How big is the actual download? The HF repo says 498 GB for total size 😅

[sammcj]

134GB, still huge for an image generation model, I suspect it could be filtered down further perhaps (or a GGUF version would be nice).

h3.c/MiniMax-H3 ▸ du -sh .
134G    .


h3.c/MiniMax-H3/FL2VA ▸ tree
[ 288]  ./
├── [ 480]  audio_vae/
│   ├── [1.9K]  config.json
│   ├── [  91]  config.yaml
│   ├── [2.1K]  dac_activations.py
│   ├── [ 835]  dac_alias_free_act.py
│   ├── [3.2K]  dac_alias_free_filter.py
│   ├── [1.7K]  dac_alias_free_resample.py
│   ├── [3.2K]  dac_attn_proj.py
│   ├── [7.1K]  dac_audio_vae.py
│   ├── [6.9K]  dac_bigvgan.py
│   ├── [ 362]  dac_utils.py
│   ├── [ 440]  metadata.json
│   ├── [3.4K]  minimax_h3_audio_vae.py
│   └── [577M]  model.safetensors
├── [ 719]  model_index.json
├── [ 288]  processor/
│   ├── [5.4K]  chat_template.json
│   ├── [1.6M]  merges.txt
│   ├── [ 390]  preprocessor_config.json
│   ├── [ 11K]  tokenizer_config.json
│   ├── [6.7M]  tokenizer.json
│   ├── [ 385]  video_preprocessor_config.json
│   └── [2.6M]  vocab.json
├── [ 800]  text_encoder/
│   ├── [5.4K]  chat_template.json
│   ├── [1.4K]  config.json
│   ├── [1.6M]  merges.txt
│   ├── [4.6G]  model-00001-of-00014.safetensors
│   ├── [4.5G]  model-00002-of-00014.safetensors
│   ├── [4.5G]  model-00003-of-00014.safetensors
│   ├── [4.5G]  model-00004-of-00014.safetensors
│   ├── [4.5G]  model-00005-of-00014.safetensors
│   ├── [4.5G]  model-00006-of-00014.safetensors
│   ├── [4.5G]  model-00007-of-00014.safetensors
│   ├── [4.5G]  model-00008-of-00014.safetensors
│   ├── [4.5G]  model-00009-of-00014.safetensors
│   ├── [4.5G]  model-00010-of-00014.safetensors
│   ├── [4.5G]  model-00011-of-00014.safetensors
│   ├── [4.5G]  model-00012-of-00014.safetensors
│   ├── [4.5G]  model-00013-of-00014.safetensors
│   ├── [3.0G]  model-00014-of-00014.safetensors
│   ├── [ 96K]  model.safetensors.index.json
│   ├── [ 390]  preprocessor_config.json
│   ├── [ 11K]  tokenizer_config.json
│   ├── [6.7M]  tokenizer.json
│   ├── [ 385]  video_preprocessor_config.json
│   └── [2.6M]  vocab.json
├── [ 192]  tokenizer/
│   ├── [1.6M]  merges.txt
│   ├── [ 11K]  tokenizer_config.json
│   ├── [6.7M]  tokenizer.json
│   └── [2.6M]  vocab.json
├── [ 544]  transformer/
│   ├── [ 604]  config.json
│   ├── [4.9G]  model-00001-of-00013.safetensors
│   ├── [4.8G]  model-00002-of-00013.safetensors
│   ├── [4.8G]  model-00003-of-00013.safetensors
│   ├── [4.8G]  model-00004-of-00013.safetensors
│   ├── [4.8G]  model-00005-of-00013.safetensors
│   ├── [4.8G]  model-00006-of-00013.safetensors
│   ├── [4.8G]  model-00007-of-00013.safetensors
│   ├── [4.8G]  model-00008-of-00013.safetensors
│   ├── [4.8G]  model-00009-of-00013.safetensors
│   ├── [4.8G]  model-00010-of-00013.safetensors
│   ├── [4.8G]  model-00011-of-00013.safetensors
│   ├── [4.8G]  model-00012-of-00013.safetensors
│   ├── [4.0G]  model-00013-of-00013.safetensors
│   └── [ 37K]  model.safetensors.index.json
└── [ 608]  video_vae/
    ├── [5.6K]  attention.py
    ├── [9.3K]  base_module.py
    ├── [1.8K]  config.json
    ├── [4.4K]  conv.py
    ├── [5.6K]  flash.py
    ├── [5.7K]  func.py
    ├── [ 47K]  klvae.py
    ├── [5.0K]  minimax_h3_video_vae.py
    ├── [ 10K]  norm.py
    ├── [1.2K]  normalize.py
    ├── [ 13K]  parallel.py
    ├── [ 128]  source/
    │   ├── [1.1K]  config.json
    │   └── [9.7G]  model.safetensors
    ├── [ 764]  utils.py
    ├── [8.6K]  vae_cnn.py
    ├── [1.8K]  vae_module.py
    ├── [8.2K]  vae_processor.py
    └── [ 13K]  vae_vit.py

8 directories, 81 files

[daaain]

Right, so that's probably the full precision model!

This is very helpful, I think I'll wait for quantisation...