Build the DS4 chat prompt: BOS, optional system text, user prompt, assistant * marker, and either or depending on the requested mode. Max * thinking is only a prompt prefix: the model still enters through . */
| 22283 | metal_graph_qr(g), |
| 22284 | metal_graph_kv_raw(g), |
| 22285 | metal_graph_comp_kv_cur(g), |
| 22286 | metal_graph_comp_sc_cur(g), |
| 22287 | metal_graph_comp_kv_cur(g), |
| 22288 | metal_graph_comp_sc_cur(g), |
| 22289 | g->layer_attn_state_kv[il], |
| 22290 | g->layer_attn_state_score[il], |
| 22291 | g->layer_attn_state_kv[il], |
| 22292 | g->layer_attn_state_score[il], |
| 22293 | model->map, |
| 22294 | model->size, |
| 22295 | layer->attn_q_a->abs_offset, |
| 22296 | layer->attn_kv->abs_offset, |
| 22297 | layer->attn_compressor_kv->abs_offset, |
| 22298 | layer->attn_compressor_gate->abs_offset, |
| 22299 | layer->attn_compressor_kv->abs_offset, |
| 22300 | layer->attn_compressor_gate->abs_offset, |
| 22301 | layer->attn_compressor_ape->abs_offset, |
| 22302 | layer->attn_compressor_ape->type, |
| 22303 | layer->attn_compressor_ape->abs_offset, |
| 22304 | layer->attn_compressor_ape->type, |
| 22305 | DS4_N_EMBD, |
| 22306 | q_rank, |
| 22307 | DS4_N_HEAD_DIM, |
| 22308 | DS4_N_HEAD_DIM, |
| 22309 | 0u, |
no test coverage detected