Hybrid MLX quant of ornith-ai/Ornith-1.5-35B-A3B, pinned to revision 10fbf86fed7ecee4a061f8b499a618f46001cac1.
mtp/weights.safetensorsmtplx serve \
--model Shiftedx/ornith-1.5-35b-a3b-attention8-bf16recurrence-vision-mtplx \
--profile sustained \
--depth 3 \
--mtp \
--reasoning off
Tested on an M4 Max with 64 GB unified memory.
| Gate | Result |
|---|---|
| Executable Python tasks | 8/8 |
| Text exact-output smoke | pass |
| Vision smoke | pass |
| MTP tensor gate | 785/785 exact |
The final three-prompt, 512-token MTPLX sweep selected D3 at 109.0 decode tok/s. D1, D2, and D3 all passed the quality gate. Fans remained on automatic, so treat the speed as a host-specific measurement rather than a universal benchmark.
Vision requires a compatible MLX-VLM or MTPLX runtime. Review generated code and tool calls before execution. The parent model and this derivative use the upstream MIT license; see the parent repository for original training and architecture details.
model-00014-of-00017.safetensors 2.1 GBmodel-00015-of-00017.safetensors 2.1 GBmodel-00013-of-00017.safetensors 2.0 GBmtp/weights.safetensors 1.7 GBmodel-00009-of-00017.safetensors 1.6 GBmodel-00006-of-00017.safetensors 1.6 GBmodel-00012-of-00017.safetensors 1.6 GBmodel-00003-of-00017.safetensors 1.6 GBmodel-00001-of-00017.safetensors 1.5 GBmodel-00007-of-00017.safetensors 1.5 GBmodel-00010-of-00017.safetensors 1.5 GBmodel-00004-of-00017.safetensors 1.5 GBmodel-00017-of-00017.safetensors 1.4 GBmodel-00008-of-00017.safetensors 1.3 GBmodel-00011-of-00017.safetensors 1.3 GBmodel-00005-of-00017.safetensors 1.3 GBmodel-00002-of-00017.safetensors 1.3 GBmodel-vision-00001-of-00001.safetensors 893 MBmodel-00016-of-00017.safetensors 579 MB有文件?提交种子。
提交种子