Instructions to use corechan/MiniMax-H3_NF4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use corechan/MiniMax-H3_NF4 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("corechan/MiniMax-H3_NF4", dtype=torch.bfloat16, device_map="cuda") prompt = "Hi, what can you help me with?" image = pipe(prompt).images[0] - Notebooks
- Google Colab
- Kaggle
update manifest
Browse files- h3_upload_manifest.json +7 -20
h3_upload_manifest.json
CHANGED
|
@@ -1,30 +1,17 @@
|
|
| 1 |
{
|
| 2 |
"app": "MiniMax-H3 WebUI",
|
| 3 |
-
"preset": "メモリ上の量子化モデル(セル2の実行後)",
|
| 4 |
"source": "memory (save_pretrained from the running session)",
|
| 5 |
-
"uploaded_at": "2026-
|
| 6 |
"layout": "quantized",
|
|
|
|
|
|
|
| 7 |
"components": [
|
| 8 |
-
{
|
| 9 |
-
"name": "transformer",
|
| 10 |
-
"quant": "nf4",
|
| 11 |
-
"path_in_repo": "quantized/transformer-nf4",
|
| 12 |
-
"bytes": 18383777418
|
| 13 |
-
},
|
| 14 |
-
{
|
| 15 |
-
"name": "transformer_ref",
|
| 16 |
-
"quant": "nf4",
|
| 17 |
-
"path_in_repo": "quantized/transformer_ref-nf4",
|
| 18 |
-
"bytes": 18383777412
|
| 19 |
-
},
|
| 20 |
{
|
| 21 |
"name": "text_encoder",
|
| 22 |
-
"quant": "
|
| 23 |
-
"path_in_repo": "quantized/text_encoder-
|
| 24 |
-
"bytes":
|
| 25 |
}
|
| 26 |
],
|
| 27 |
-
"
|
| 28 |
-
"text_encoder_quant": "nf4",
|
| 29 |
-
"note": "セル2の _save_quant_cache() と同じ形式。{CACHE_DIR}/quantized/ 配下へ置けば USE_DRIVE_QUANT_CACHE = True でそのまま読める。vae / audio_vae / tokenizer などの軽量コンポーネントは含まれていないので、本家 MiniMaxAI/MiniMax-H3 から取るか、別途ミラーを上げること"
|
| 30 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"app": "MiniMax-H3 WebUI",
|
|
|
|
| 3 |
"source": "memory (save_pretrained from the running session)",
|
| 4 |
+
"uploaded_at": "2026-09-12T09:24:02+0000",
|
| 5 |
"layout": "quantized",
|
| 6 |
+
"torchao": "0.15.0",
|
| 7 |
+
"torch": "2.11.0+cu128",
|
| 8 |
"components": [
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
{
|
| 10 |
"name": "text_encoder",
|
| 11 |
+
"quant": "int8",
|
| 12 |
+
"path_in_repo": "quantized/text_encoder-int8",
|
| 13 |
+
"bytes": 27622285372
|
| 14 |
}
|
| 15 |
],
|
| 16 |
+
"note": "セル2の _save_quant_cache() と同じ形式。USE_HF_QUANT_CACHE = True でそのまま読める。⚠️ pickle 形式(torchao 系)は保存時の torchao の版に縛られるので、.complete の torchao と実行環境が一致していること。vae / audio_vae / tokenizer は含まれていない(本家から取る)"
|
|
|
|
|
|
|
| 17 |
}
|