Text Classification
MLX
Safetensors
English
Japanese
gemma4
d1a
decision-model
calibration
system-one
chat
multimodal
vision
audio
video
pull-requests
apple-silicon
quantized
4-bit precision
qat
on-device
experimental
Instructions to use mlx-community/d1a-e4b-it-qat-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- MLX
How to use mlx-community/d1a-e4b-it-qat-4bit with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] hf download mlx-community/d1a-e4b-it-qat-4bit --local-dir d1a-e4b-it-qat-4bit
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
Commit History
Yesterday
-
Card: transfer-v4 calibration as a known limit (d1a#361)
Verified signature808b145
4 days ago
-
Media read the per-layer pad row as stored before quantization (jonpol01/d1a#394)
Verified signaturemedia_pad_row.safetensors holds the pad token's per-layer embedding row as the QAT checkpoint stores it; d1a_config.json names it; the card says what it changes.f5b444b
7 days ago
-
Routing temperature 1.41 for use_case routing (d1a#382)
Verified signatureuse_case_temperatures.routing = 1.4142 with its fit record. Fit set uninformative (accuracy 0.99); the method's point estimate, confirmed directionally out of sample on factory-development (ECE 0.049 -> 0.035); not claimed optimal. Every other request keeps 1.66; no top choice changes.431a7e8
18 days ago
-
Card: the scorecard
Verified signaturedee1cd4
19 days ago
-
Model card
Verified signatured0aca51 -
D1A-E4B-it QAT, MLX 4-bit: the files of JohnP1/d1a-e4b-it-qat-mlx-q4@508ba4af
Verified signature341acb0 -
initial commit
Verified signaturedb62bd3