An unofficial ONNX export of Kev-0.8B v1.0 (commit bf75a6a8), a decision model that reads a document plus typed questions (choice, noul, score) and returns calibrated probabilities over each question's options. It generates no text.
dc7cdfe2.lm_head and the MTP head are excluded. The graph output is last_hidden_state [batch, seq, 1024].PointerHead from head.pt) is shipped separately as head.bin + head.json.| File | Contents |
|---|---|
onnx/model.onnx, onnx/model.onnx.data |
Backbone (3.1 GB). Weights are 64 KiB-aligned so onnxruntime can memory-map them. Keep both names together. |
tokenizer.json, tokenizer_config.json |
The base model's tokenizer, which kev itself uses (kev's load_tokenizer(meta.base)). |
head.bin |
float32 little-endian: q.weight [256, 1024], q.bias [256], k.weight [256, 1024], k.bias [256], back to back. |
head.json |
Shapes and float offsets of those tensors, head_dim 256, hidden_size 1024, temperature 2.3510958125672174. |
REPRODUCE_scripts.txt, requirements.txt |
Every script used to build these files, in run order, with pinned versions. |
The graph uses onnxruntime contrib ops (LinearAttention, CausalConvWithState, GroupQueryAttention, MatMulNBits). It runs on onnxruntime 1.32 (Python) and onnxruntime-node 1.30.0.
The backbone is Qwen3.5's hybrid Gated DeltaNet architecture, so kev scores each question as its own causal row: [<state> state… <q> instructions… (<opt> option… </opt>)×K <decide>].
<|fim_prefix|>=248060 (state), <|fim_middle|>=248061 (q), <|box_start|>=248049 (opt), <|box_end|>=248050 (/opt), <|fim_suffix|>=248062 (decide).<|name|> to <¦name¦>.noul becomes the options ["no", "yes"] (answer = p(yes)). score gives one option per level in order (answer = Σ i·p_i). choice gives "name: description".input_ids [1, L]attention_mask all ones [1, L]position_ids [3, 1, L], the row 0..L-1 repeated 3 timespast_key_values.* [1, 2, 0, 256], and past.*.conv [1, 6144, 3] / past.*.recurrent [1, 16, 128, 128] all zeroslast_hidden_state at <decide> and h_j at each option's </opt>:q = Wq·h_d + bqk_j = Wk·h_j + bkz_j = (k_j·q) / sqrt(256) / temperaturep = softmax(z)These are checked against kev's own PyTorch code (kev @ 2074cc4, bf16 on CPU; fp32 did not fit on the 3 GB test machine) on 3 short requests (one choice, one noul, one score):
This is a small smoke test, not a re-run of kev's evaluation suites.
head.bin 2 MBonnx/model.onnx 306.8 kBjaredpalmer/kev-0.8b