Thump604
/

Qwen3.5-4B-MLX-4bit

vision-language-model

4-bit precision

Model card Files Files and versions

Qwen3.5-4B-MLX-4bit

3.15 GB

Ctrl+K

Ctrl+K

1 contributor

History: 4 commits

Thump604's picture

fix: revert to quantized MTP weights (BF16 MTP produces 0% acceptance on quantized backbones)

5f71025 verified about 1 month ago

.gitattributes

1.57 kB
Upload folder using huggingface_hub about 2 months ago
README.md

2.27 kB
Upload folder using huggingface_hub about 2 months ago
chat_template.jinja

7.76 kB
Upload folder using huggingface_hub about 2 months ago
config.json

2.87 kB
Upload folder using huggingface_hub about 2 months ago
model-mtp.safetensors

86.7 MB
xet

fix: revert to quantized MTP weights (BF16 MTP produces 0% acceptance on quantized backbones) about 1 month ago
model.safetensors

3.03 GB
xet

Upload folder using huggingface_hub about 2 months ago
model.safetensors.index.json

99.4 kB
Upload folder using huggingface_hub about 2 months ago
preprocessor_config.json

390 Bytes
Upload folder using huggingface_hub about 2 months ago
processor_config.json

1.3 kB
Upload folder using huggingface_hub about 2 months ago
tokenizer.json

20 MB
xet

Upload folder using huggingface_hub about 2 months ago
tokenizer_config.json

1.14 kB
Upload folder using huggingface_hub about 2 months ago
video_preprocessor_config.json

385 Bytes
Upload folder using huggingface_hub about 2 months ago
vocab.json

6.72 MB
Upload folder using huggingface_hub about 2 months ago