oQ5e mixed-precision MLX conversion of Sage S2.1 Flash for Apple Silicon.
Effective weight storage: 5.61 BPW · 225.36 GB · group size 64.
Sage S2.1 Flash is a scientific multimodal and agentic model from Shanghai Artificial Intelligence Laboratory / InternLM, built on the GLM-5.3-Flash architecture. This conversion retains the full expert set, vision encoder, multi-token-prediction head and original tokenizer, processor and chat template. No additional training, abliteration or expert pruning was applied.
internlm/Sage-S2.1-Flash@76f4513db7464485605a002a06a6190de0460574.0.8.0rc1 native oQe streaming converter.Conversion details are in CONVERSION.json.
5.61 effective bits per weight (BPW), including affine scales/biases, unquantized tensors and safetensors headers.
| Component | Weight precision | Original parameters | Stored GB |
|---|---|---|---|
| Routed MoE experts | 5-bit | 311,653,564,416 | 214.261826 |
| Attention projections | 8-bit | 6,088,556,544 | 6.469091 |
| Shared experts | 8-bit | 1,082,130,432 | 1.149764 |
| Dense MLP projections | 8-bit | 452,984,832 | 0.481296 |
| Token embeddings | 8-bit | 634,388,480 | 0.674038 |
| Output head | 8-bit | 634,388,480 | 0.674038 |
| Vision encoder | BF16 | 563,627,008 | 1.127254 |
| Linear-attention convolutions | BF16 | 3,342,336 | 0.006685 |
| Indexer pooling/compression | 8-bit / BF16 | 89,662,464 | 0.101173 |
| Text/MTP normalization | BF16 | 422,144 | 0.000844 |
| MTP fusion projection | BF16 | 33,554,432 | 0.067109 |
| MoE routers/correction biases | FP32 | 50,737,248 | 0.202949 |
| Hyperconnections | FP32 | 35,391,870 | 0.141567 |
| Recurrent decay controls | FP32 | 280,704 | 0.001123 |
| Safetensors headers | — | — | 0.000403 |
| Total | 5.61 effective BPW | 321,323,031,390 | 225.359160 |
Stored GB uses decimal units. Component totals include MTP and exclude the separate scientific projectors. The nine original scientific projectors add 0.168 GB and remain in FP32.
Per-block precision map| Block | Routed expert matrices | Other quantized matrices | Unquantized tensors | Stored GB |
|---|---|---|---|---|
| Text 0 | — (dense MLP) | 8-bit | BF16 / FP32 | 0.310052 |
| Text 1 | — (dense MLP) | 8-bit | BF16 / FP32 | 0.310052 |
| Text 2 | — (dense MLP) | 8-bit | BF16 / FP32 | 0.310052 |
| Text 3 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 4 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 5 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 6 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 7 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 8 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 9 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 10 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 11 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 12 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 13 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 14 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 15 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 16 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 17 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 18 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 19 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 20 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 21 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 22 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 23 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 24 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 25 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 26 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 27 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 28 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 29 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 30 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 31 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 32 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 33 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 34 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 35 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 36 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 37 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 38 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 39 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 40 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 41 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 42 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| Text 43 | 5-bit | 8-bit | BF16 / FP32 | 5.150670 |
| Text 44 | 5-bit | 8-bit | BF16 / FP32 | 5.163911 |
| MTP 0 | 5-bit | 8-bit | BF16 / FP32 | 5.214657 |
Complete per-module/tensor precision data (CSV) and exact accounting (JSON) are included.
Use oMLX 0.8.0rc1 for text, images, tool calling and native MTP decoding. Keep the included Sage tokenizer, processor and chat template together.
Scientific inputs require the matching SciPerception encoders and integration in addition to the included projector files.
Each variant is converted directly from the same pinned creator BF16/FP32 checkpoint, not requantized from another variant.
Model: Shanghai Artificial Intelligence Laboratory / InternLM. GLM architecture: Z.ai. Conversion tooling: oMLX / oQe. Community conversion: Hygaard.
Licensed under MIT. See NOTICE for attribution and modification notes.
model-00012-of-00038.safetensors 6.6 GBmodel-00025-of-00038.safetensors 6.6 GBmodel-00015-of-00038.safetensors 6.6 GBmodel-00021-of-00038.safetensors 6.6 GBmodel-00024-of-00038.safetensors 6.6 GBmodel-00008-of-00038.safetensors 6.6 GBmodel-00020-of-00038.safetensors 6.6 GBmodel-00034-of-00038.safetensors 6.6 GBmodel-00011-of-00038.safetensors 6.6 GBmodel-00017-of-00038.safetensors 6.6 GBmodel-00030-of-00038.safetensors 6.6 GBmodel-00014-of-00038.safetensors 6.6 GBmodel-00018-of-00038.safetensors 6.6 GBmodel-00027-of-00038.safetensors 6.6 GBmodel-00028-of-00038.safetensors 6.6 GBmodel-00009-of-00038.safetensors 6.6 GBmodel-00033-of-00038.safetensors 6.6 GBmodel-00004-of-00038.safetensors 6.6 GBmodel-00007-of-00038.safetensors 6.6 GBmodel-00022-of-00038.safetensors 6.6 GBmodel-00035-of-00038.safetensors 6.6 GBmodel-00031-of-00038.safetensors 6.6 GBmodel-00005-of-00038.safetensors 6.6 GBmodel-00002-of-00038.safetensors 6.2 GBmodel-00037-of-00038.safetensors 5.1 GBmodel-00001-of-00038.safetensors 5.0 GBmodel-00023-of-00038.safetensors 5.0 GBmodel-00010-of-00038.safetensors 5.0 GBmodel-00026-of-00038.safetensors 5.0 GBmodel-00013-of-00038.safetensors 5.0 GBmodel-00016-of-00038.safetensors 5.0 GBmodel-00029-of-00038.safetensors 5.0 GBmodel-00036-of-00038.safetensors 5.0 GBmodel-00019-of-00038.safetensors 5.0 GBmodel-00032-of-00038.safetensors 5.0 GBmodel-00003-of-00038.safetensors 5.0 GBmodel-00006-of-00038.safetensors 5.0 GBmodel-00038-of-00038.safetensors 1.2 GBprojectors/sage-s2.1-flash/protein-seq/mpbind/projector.safetensors 28 MBprojectors/sage-s2.1-flash/rna/ncrna/projector.safetensors 25 MBprojectors/sage-s2.1-flash/ecg/phu01/projector.safetensors 20 MBprojectors/sage-s2.1-flash/eeg/neu06/projector.safetensors 19 MBprojectors/sage-s2.1-flash/xrd/simxrd/projector.safetensors 19 MBprojectors/sage-s2.1-flash/protein-struct/psr/projector.safetensors 19 MBprojectors/sage-s2.1-flash/cif/jarvis/projector.safetensors 17 MBprojectors/sage-s2.1-flash/spectra/qm9s/projector.safetensors 17 MBprojectors/sage-s2.1-flash/dna/splice/projector.safetensors 4 MBinternlm/Sage-S2.1-Flash