Salvatore Sanfilippo
df6eedd972
Share the Metal session graph with CUDA for text and vision, MTP and exact sampling, steering, checkpoints, and disk-only BF16 n-grams. Add bounded concurrent Linux n-gram reads, vectorized dense and expert kernels, register-held recurrent prefill state, and efficient small speculative batches. Keep Qwen out of DeepSeek batch dispatch and exclude disk n-grams from resident memory accounting. Use normal file-backed weight loading in diagnostics and link n-gram support into the CUDA quality scorer. Include independent kernel and memory-safety tests, mixed-prefill and recovery checks, long-context server and agent coverage, and user documentation with measured DGX Spark performance.
20 lines
666 B
C
20 lines
666 B
C
#ifndef DS4_QWEN4_VISION_H
|
|
#define DS4_QWEN4_VISION_H
|
|
|
|
#include <stdint.h>
|
|
|
|
#define DS4_QWEN4_VISION_LAYERS 27
|
|
typedef struct {
|
|
uint64_t ln1_w, ln1_b, qkv_w, qkv_b, out_w, out_b, ln2_w, ln2_b, up_w, up_b, down_w, down_b;
|
|
uint32_t qkv_type, out_type, up_type, down_type;
|
|
} ds4_qwen4_vision_layer_weights;
|
|
typedef struct {
|
|
uint64_t patch_w0, patch_w1, patch_b, pos_embd, post_ln_w, post_ln_b, mm0_w, mm0_b, mm2_w, mm2_b;
|
|
uint32_t mm0_type, mm2_type, patch_type;
|
|
uint32_t n_embd, n_ff, n_head, n_patch, n_merge, n_pos_side, n_out;
|
|
float eps;
|
|
ds4_qwen4_vision_layer_weights layer[DS4_QWEN4_VISION_LAYERS];
|
|
} ds4_qwen4_vision_weights;
|
|
|
|
#endif
|