MCPcopy Create free account
hub / github.com/antirez/ds4 / payload_read_tensor_span

Function payload_read_tensor_span

ds4.c:23473–23494  ·  view source on GitHub ↗

Source from the content-addressed store, hash-verified

23471 DS4_ROPE_YARN_BETA_SLOW,
23472 DS4_RMS_EPS) != 0;
23473 }
23474 if (!decode_q_norm_rope_fused) {
23475 if (ok) ok = ds4_gpu_head_rms_norm_tensor(metal_graph_q(g), 1, tp_heads, DS4_N_HEAD_DIM, DS4_RMS_EPS) != 0;
23476 if (ok) {
23477 metal_graph_debug_dump_tensor("Qnorm", metal_graph_q(g), q_dim, il, pos);
23478 }
23479 if (ok) ok = ds4_gpu_rope_tail_tensor(metal_graph_q(g), 1, tp_heads, DS4_N_HEAD_DIM,
23480 DS4_N_ROT, pos,
23481 compressed ? (uint32_t)DS4_ROPE_ORIG_CTX : 0,
23482 false, freq_base, freq_scale, ext_factor, attn_factor,
23483 DS4_ROPE_YARN_BETA_FAST, DS4_ROPE_YARN_BETA_SLOW) != 0;
23484 }
23485 DS4_METAL_PROFILE_DECODE_STAGE("q_path");
23486 if (ok) {
23487 metal_graph_debug_dump_tensor("Qcur", metal_graph_q(g), q_dim, il, pos);
23488 }
23489 if (!qkv_rms_fused) {
23490 if (ok) ok = metal_graph_matmul_dense_quant_tensor(metal_graph_kv_raw(g),
23491 model,
23492 layer->attn_kv,
23493 DS4_N_EMBD,
23494 DS4_N_HEAD_DIM,
23495 metal_graph_attn_norm(g),
23496 1);
23497 if (ok) {

Callers 2

ds4_session_load_payloadFunction · 0.85

Calls 2

payload_set_errFunction · 0.85
payload_read_bytesFunction · 0.85

Tested by

no test coverage detected