d1-omni-600M ExecuTorch INT4 (Localia gate)
Text-only decision gate exported from LiquidAI/d1-omni-600M for on-device command approval.
Files
| Path | Backend | Notes |
|---|---|---|
android/d1_omni_gate_xnnpack_int4.pte |
XNNPACK | Weight-only INT4 (group 32) + fp32 decision head |
ios/d1_omni_gate_mlx_int4.pte |
MLX | Same quant recipe for Apple Silicon |
*/tokenizer.json |
— | Tokenizer assets |
I/O
- Inputs:
input_ids[1,512] long,pad[1,512] bool,marker_pos[1,4] long,marker_mask[1,4] bool,qtype[1] long(0=choice,1=score,2=noul) - Output:
logits[1,4] float(masked unused options ≈ -1e4)
License
LFM Open License v1.0 (same as upstream).
- Downloads last month
- 32
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support
Model tree for kloodia/d1-omni-600m-executorch-int4
Base model
LiquidAI/LFM2.5-350M-Base Finetuned
LiquidAI/LFM2.5-Encoder-350M Finetuned
LiquidAI/d1-omni-600M