jacksonchen1103/sn92-gen-mg

REAX SN92 generalist entry (merge)

A weight-space merge of four public REAX SN92 generalist entries that share the base baselabs/mammoth-1 (61cec44c281c693992e3d6ebeb44dd3402dfa186). No training of our own: all learned changes come from the parents below, whose authors deserve the credit for them.

Parents

Model	Revision	Weight of its change
danyloooah/sn92-generalist-m1-v3a	ac7fe8d04867802926ac9b05d16592a7b53bc26a	0.5
gurujustin/sn92-gen-m1b	1ea7105575281e7434b438fa4f0591ea4b87f74c	0.7
mufasa1228/sn92-gen-r2	5dcb31e1d72efe7b9bc868b2fe173d12e8aaed77	0.4
edinlong83/sn92-generalist-h5	764dad666e3018c4724c02cc43d6e882f8dd431b	0.3

Recipe

Task arithmetic: for every tensor the base also has, base + 0.5 * (parent_1 - base) + 0.7 * (parent_2 - base) + 0.4 * (parent_3 - base) + 0.3 * (parent_4 - base) with base = baselabs/mammoth-1@61cec44c281c693992e3d6ebeb44dd3402dfa186 (its names carry a leading model.), i.e. parent i keeps weight_i of its own change from the base (danyloooah/sn92-generalist-m1-v3a: 0.5, gurujustin/sn92-gen-m1b: 0.7, mufasa1228/sn92-gen-r2: 0.4, edinlong83/sn92-generalist-h5: 0.3). The answer readout (readout.safetensors, not in the base) is the average of the parents' readouts with the same weights, renormalised to sum 1, after multiplying each parent's readout by T_primary / T_parent (decision_config temperatures: danyloooah/sn92-generalist-m1-v3a 2, gurujustin/sn92-gen-m1b 1.38235, mufasa1228/sn92-gen-r2 2, edinlong83/sn92-generalist-h5 2), so that every parent keeps its own logit scale under the copied temperature. Computed one tensor at a time in float32 and rounded back to bfloat16 (round to nearest even). All parents have identical tensor names, shapes and dtypes (Qwen3_5Model joint-readout package). config.json, decision_config.json, joint_config.json and the tokenizer are copied unchanged from danyloooah/sn92-generalist-m1-v3a.

Licence

MIT. See LICENSE (copyright lines of every parent retained) and NOTICE (each parent's notice retained verbatim). This model is not affiliated with, endorsed by, or sponsored by Base Labs / Baseten, the Qwen team, Perplexity AI or the parents' authors.