aboutsummaryrefslogtreecommitdiff
path: root/src/embed.c
diff options
context:
space:
mode:
authorOnur Gündüz <ogunduz@gmail.com>2026-06-16 16:27:47 +0200
committerOnur Gündüz <ogunduz@gmail.com>2026-06-16 16:27:47 +0200
commitd1ba1ebc06ef0d62ca932388fae83e6992594719 (patch)
treef962207b779016593258f990348216999a58766a /src/embed.c
parentbe20dfda0f836808dccfc14db244592e41ba28bc (diff)
downloadffwd-d1ba1ebc06ef0d62ca932388fae83e6992594719.tar.gz
ffwd-d1ba1ebc06ef0d62ca932388fae83e6992594719.zip
chore: move sources to src/ with proper names, drop qwen_ prefix
Diffstat (limited to 'src/embed.c')
-rw-r--r--src/embed.c1814
1 files changed, 1814 insertions, 0 deletions
diff --git a/src/embed.c b/src/embed.c
new file mode 100644
index 0000000..f482c95
--- /dev/null
+++ b/src/embed.c
@@ -0,0 +1,1814 @@
+#include "embed.h"
+#include "config.h"
+#include "internal.h"
+#include "kernels.h"
+#include "safetensors.h"
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+#include <limits.h>
+#include <float.h>
+
+struct embed_late_model {
+ embed_model_t *base;
+ safetensors_file_t *dense_sf;
+ embed_weight_ref_t projection; /* [token_dim, hidden] */
+ int token_dim;
+};
+
+struct embed_late_workspace {
+ const embed_late_model_t *model;
+ embed_workspace_t *base_ws;
+ float *states;
+ int states_seq_cap;
+};
+
+/* ========================================================================
+ * Globals
+ * ======================================================================== */
+
+int embed_verbose = 0;
+
+#define EMBED_BF16_QKV_FUSE_MAX_SEQ 16
+#define EMBED_BF16_PAIR_FUSE_MAX_SEQ 4
+#define EMBED_MIN_WORKSPACE_SEQ_CAP 16
+#define EMBED_WORKSPACE_SEQ_GRANULARITY 16
+
+/* ========================================================================
+ * Size helpers
+ * ======================================================================== */
+
+static int mul_size(size_t a, size_t b, size_t *out) {
+ if (a != 0 && b > SIZE_MAX / a)
+ return -1;
+ *out = a * b;
+ return 0;
+}
+
+static int add_size(size_t a, size_t b, size_t *out) {
+ if (b > SIZE_MAX - a)
+ return -1;
+ *out = a + b;
+ return 0;
+}
+
+static int grow_cap(int current, int needed, int *out) {
+ if (needed <= 0)
+ return -1;
+
+ int cap = needed;
+ if (cap < EMBED_MIN_WORKSPACE_SEQ_CAP)
+ cap = EMBED_MIN_WORKSPACE_SEQ_CAP;
+
+ int rem = cap % EMBED_WORKSPACE_SEQ_GRANULARITY;
+ if (rem != 0) {
+ int add = EMBED_WORKSPACE_SEQ_GRANULARITY - rem;
+ if (cap > INT_MAX - add)
+ return -1;
+ cap += add;
+ }
+
+ if (cap < current)
+ cap = current;
+
+ *out = cap;
+ return 0;
+}
+
+static int realloc_floats(float **ptr, size_t count) {
+ size_t bytes;
+ if (mul_size(count, sizeof(float), &bytes) != 0)
+ return -1;
+
+ void *p = realloc(*ptr, bytes);
+ if (!p && bytes != 0)
+ return -1;
+ *ptr = (float *)p;
+ return 0;
+}
+
+static float *malloc_floats(size_t count) {
+ size_t bytes;
+ if (mul_size(count, sizeof(float), &bytes) != 0)
+ return NULL;
+ return (float *)malloc(bytes);
+}
+
+static int model_dir_has_file(const char *model_dir, const char *rel) {
+ char path[1024];
+ int n = snprintf(path, sizeof(path), "%s/%s", model_dir, rel);
+ if (n < 0 || (size_t)n >= sizeof(path))
+ return 0;
+ FILE *f = fopen(path, "rb");
+ if (!f)
+ return 0;
+ fclose(f);
+ return 1;
+}
+
+static int model_dir_has_late_projection(const char *model_dir) {
+ return model_dir_has_file(model_dir, "1_Dense/config.json") &&
+ model_dir_has_file(model_dir, "1_Dense/model.safetensors");
+}
+
+static int realloc_floats_2d(float **ptr, int rows, int cols) {
+ if (rows < 0 || cols < 0)
+ return -1;
+
+ size_t count;
+ if (mul_size((size_t)rows, (size_t)cols, &count) != 0)
+ return -1;
+ return realloc_floats(ptr, count);
+}
+
+static int ensure_late_state_buffer(embed_late_workspace_t *ws, int n_tokens) {
+ if (!ws || !ws->model || n_tokens <= 0)
+ return -1;
+ if (n_tokens <= ws->states_seq_cap)
+ return 0;
+ int hidden = ws->model->base->config.hidden_size;
+ if (realloc_floats_2d(&ws->states, n_tokens, hidden) != 0)
+ return -1;
+ ws->states_seq_cap = n_tokens;
+ return 0;
+}
+
+static int realloc_ints(int **ptr, size_t count) {
+ size_t bytes;
+ if (mul_size(count, sizeof(int), &bytes) != 0)
+ return -1;
+
+ void *p = realloc(*ptr, bytes);
+ if (!p && bytes != 0)
+ return -1;
+ *ptr = (int *)p;
+ return 0;
+}
+
+/* ========================================================================
+ * Weight loading (direct mmap pointers into safetensors)
+ * ======================================================================== */
+
+static size_t dtype_size(safetensor_dtype_t dtype) {
+ switch (dtype) {
+ case DTYPE_F32:
+ return sizeof(float);
+ case DTYPE_BF16:
+ return sizeof(uint16_t);
+ default:
+ return 0;
+ }
+}
+
+static const char *dtype_name(safetensor_dtype_t dtype) {
+ switch (dtype) {
+ case DTYPE_F32:
+ return "F32";
+ case DTYPE_F16:
+ return "F16";
+ case DTYPE_BF16:
+ return "BF16";
+ case DTYPE_I32:
+ return "I32";
+ case DTYPE_I64:
+ return "I64";
+ case DTYPE_BOOL:
+ return "BOOL";
+ default:
+ return "UNKNOWN";
+ }
+}
+
+static int tensor_has_supported_shape(const safetensors_file_t *sf,
+ const safetensor_t *t,
+ const char *name,
+ const int64_t *shape,
+ int ndim,
+ int bf16_ok) {
+ if (t->dtype != DTYPE_F32 && !(bf16_ok && t->dtype == DTYPE_BF16)) {
+ fprintf(stderr, "embed: unsupported dtype for %s: got %s, expected %s\n", name,
+ dtype_name(t->dtype), bf16_ok ? "F32 or BF16" : "F32");
+ return 0;
+ }
+ if (t->ndim != ndim) {
+ fprintf(stderr, "embed: bad rank for %s: got %d, expected %d\n", name, t->ndim, ndim);
+ return 0;
+ }
+
+ size_t numel = 1;
+ for (int i = 0; i < ndim; i++) {
+ if (t->shape[i] != shape[i]) {
+ fprintf(stderr,
+ "embed: bad shape for %s at dim %d: "
+ "got %lld, expected %lld\n",
+ name, i, (long long)t->shape[i], (long long)shape[i]);
+ return 0;
+ }
+ if (shape[i] < 0 || (uint64_t)shape[i] > SIZE_MAX / numel) {
+ fprintf(stderr, "embed: shape too large for %s\n", name);
+ return 0;
+ }
+ numel *= (size_t)shape[i];
+ }
+
+ size_t elem_size = dtype_size(t->dtype);
+ if (elem_size == 0) {
+ fprintf(stderr, "embed: unsupported dtype for %s: got %s\n", name, dtype_name(t->dtype));
+ return 0;
+ }
+
+ size_t bytes;
+ if (mul_size(numel, elem_size, &bytes) != 0) {
+ fprintf(stderr, "embed: tensor data too large for %s\n", name);
+ return 0;
+ }
+ if (t->data_size != bytes) {
+ fprintf(stderr, "embed: bad data size for %s: got %zu, expected %zu\n", name, t->data_size,
+ bytes);
+ return 0;
+ }
+
+ size_t data_start = 8 + sf->header_size;
+ if (data_start > sf->file_size || t->data_offset > sf->file_size - data_start ||
+ t->data_size > sf->file_size - data_start - t->data_offset) {
+ fprintf(stderr, "embed: tensor data out of bounds for %s\n", name);
+ return 0;
+ }
+ return 1;
+}
+
+static embed_weight_ref_t
+load_weight_direct(multi_safetensors_t *ms, const char *name, const int64_t *shape, int ndim) {
+ embed_weight_ref_t ref = {DTYPE_UNKNOWN, NULL};
+ safetensors_file_t *sf = NULL;
+ const safetensor_t *t = multi_safetensors_find(ms, name, &sf);
+ if (!t) {
+ fprintf(stderr, "embed: tensor not found: %s\n", name);
+ return ref;
+ }
+ if (!tensor_has_supported_shape(sf, t, name, shape, ndim, 1))
+ return ref;
+
+ ref.dtype = t->dtype;
+ ref.data = safetensors_data(sf, t);
+ return ref;
+}
+
+static float *
+load_norm_f32(multi_safetensors_t *ms, const char *name, const int64_t *shape, int ndim) {
+ safetensors_file_t *sf = NULL;
+ const safetensor_t *t = multi_safetensors_find(ms, name, &sf);
+ if (!t) {
+ fprintf(stderr, "embed: tensor not found: %s\n", name);
+ return NULL;
+ }
+ if (!tensor_has_supported_shape(sf, t, name, shape, ndim, 1))
+ return NULL;
+ return safetensors_get_f32(sf, t);
+}
+
+static void bf16_row_to_f32(float *dst, const uint16_t *src, int n) {
+ for (int i = 0; i < n; i++) {
+ uint32_t u = ((uint32_t)src[i]) << 16;
+ memcpy(dst + i, &u, sizeof(float));
+ }
+}
+
+static void copy_weight_row(float *dst, const embed_weight_ref_t *w, size_t row, int row_width) {
+ if (w->dtype == DTYPE_F32) {
+ const float *src = (const float *)w->data + row * (size_t)row_width;
+ memcpy(dst, src, (size_t)row_width * sizeof(float));
+ } else {
+ const uint16_t *src = (const uint16_t *)w->data + row * (size_t)row_width;
+ bf16_row_to_f32(dst, src, row_width);
+ }
+}
+
+static int ensure_bf16_widen(embed_workspace_t *ws, size_t count) {
+ if (ws->bf16_widen_count >= count)
+ return 0;
+ float *p = (float *)realloc(ws->bf16_widen, count * sizeof(float));
+ if (!p)
+ return -1;
+ ws->bf16_widen = p;
+ ws->bf16_widen_count = count;
+ return 0;
+}
+
+static void linear_nobias_weight(embed_workspace_t *ws,
+ float *y,
+ const float *x,
+ const embed_weight_ref_t *w,
+ int seq_len,
+ int in_dim,
+ int out_dim) {
+ if (w->dtype == DTYPE_F32) {
+ embed_linear_nobias(y, x, (const float *)w->data, seq_len, in_dim, out_dim);
+ return;
+ }
+ if (seq_len > 16) {
+ /* Widen the weight matrix once into workspace scratch and use the
+ * BLAS-backed F32 path; below that the fused BF16 row kernels win. */
+ size_t count = (size_t)out_dim * (size_t)in_dim;
+ if (ensure_bf16_widen(ws, count) == 0) {
+ embed_bf16_to_f32_buf(ws->bf16_widen, (const uint16_t *)w->data, count);
+ embed_linear_nobias(y, x, ws->bf16_widen, seq_len, in_dim, out_dim);
+ return;
+ }
+ }
+ embed_linear_nobias_bf16(y, x, (const uint16_t *)w->data, seq_len, in_dim, out_dim);
+}
+
+static void linear_qkv_weight(embed_workspace_t *ws,
+ float *q,
+ float *k,
+ float *v,
+ const float *x,
+ const embed_weight_ref_t *wq,
+ const embed_weight_ref_t *wk,
+ const embed_weight_ref_t *wv,
+ int seq_len,
+ int in_dim,
+ int q_dim,
+ int kv_dim) {
+ if (wq->dtype == DTYPE_BF16 && wk->dtype == DTYPE_BF16 && wv->dtype == DTYPE_BF16 &&
+ seq_len <= EMBED_BF16_QKV_FUSE_MAX_SEQ) {
+ embed_linear_nobias_bf16_qkv(q, k, v, x, (const uint16_t *)wq->data,
+ (const uint16_t *)wk->data, (const uint16_t *)wv->data, seq_len,
+ in_dim, q_dim, kv_dim);
+ return;
+ }
+
+ linear_nobias_weight(ws, q, x, wq, seq_len, in_dim, q_dim);
+ linear_nobias_weight(ws, k, x, wk, seq_len, in_dim, kv_dim);
+ linear_nobias_weight(ws, v, x, wv, seq_len, in_dim, kv_dim);
+}
+
+static void add_bias_rows(float *x, const float *bias, int rows, int dim) {
+ if (!bias)
+ return;
+ for (int row = 0; row < rows; row++) {
+ float *dst = x + (size_t)row * dim;
+ for (int d = 0; d < dim; d++)
+ dst[d] += bias[d];
+ }
+}
+
+static void linear_pair_weight(embed_workspace_t *ws,
+ float *a,
+ float *b,
+ const float *x,
+ const embed_weight_ref_t *wa,
+ const embed_weight_ref_t *wb,
+ int seq_len,
+ int in_dim,
+ int a_dim,
+ int b_dim) {
+ if (wa->dtype == DTYPE_BF16 && wb->dtype == DTYPE_BF16 &&
+ seq_len <= EMBED_BF16_PAIR_FUSE_MAX_SEQ) {
+ embed_linear_nobias_bf16_pair(a, b, x, (const uint16_t *)wa->data,
+ (const uint16_t *)wb->data, seq_len, in_dim, a_dim, b_dim);
+ return;
+ }
+
+ linear_nobias_weight(ws, a, x, wa, seq_len, in_dim, a_dim);
+ linear_nobias_weight(ws, b, x, wb, seq_len, in_dim, b_dim);
+}
+
+/* ========================================================================
+ * Working buffer management
+ * ======================================================================== */
+
+static int ensure_buffers(embed_workspace_t *ws, const embed_config_t *c, int seq) {
+ if (seq <= ws->buf_seq_cap)
+ return 0;
+
+ int cap;
+ if (grow_cap(ws->buf_seq_cap, seq, &cap) != 0)
+ return -1;
+
+#define R(ptr, n) \
+ do { \
+ if (realloc_floats_2d(&(ptr), cap, (n)) != 0) \
+ return -1; \
+ } while (0)
+
+ R(ws->x, c->hidden_size);
+ R(ws->x_norm, c->hidden_size);
+ R(ws->q, c->q_dim);
+ R(ws->k, c->kv_dim);
+ R(ws->v, c->kv_dim);
+ R(ws->attn_out, c->q_dim);
+ R(ws->proj_out, c->hidden_size);
+ R(ws->ffn_gate, c->intermediate_size);
+ R(ws->ffn_up, c->intermediate_size);
+
+#undef R
+
+ ws->buf_seq_cap = cap;
+ return 0;
+}
+
+static void ensure_attention_scores(embed_workspace_t *ws, const int *offsets, int batch) {
+ size_t bytes = embed_bidirectional_gqa_attention_packed_scratch_bytes(offsets, batch);
+ if (bytes <= ws->attn_scores_bytes)
+ return;
+
+ void *p = realloc(ws->attn_scores, bytes);
+ if (!p)
+ return;
+ ws->attn_scores = (float *)p;
+ ws->attn_scores_bytes = bytes;
+}
+
+/* ========================================================================
+ * RoPE cache
+ * ======================================================================== */
+
+static int ensure_rope_cache(embed_workspace_t *ws, const embed_config_t *cfg, int n_pos) {
+ if (n_pos <= ws->rope_cache_cap)
+ return 0;
+
+ int cap = ws->rope_cache_cap > 0 ? ws->rope_cache_cap : 512;
+ while (cap < n_pos) {
+ if (cap > INT_MAX / 2)
+ return -1;
+ cap *= 2;
+ }
+
+ int head_dim = cfg->head_dim;
+ int half = head_dim / 2;
+ float theta = cfg->rope_theta;
+
+ size_t count;
+ if (mul_size((size_t)cap, (size_t)head_dim, &count) != 0)
+ return -1;
+ if (realloc_floats(&ws->rope_cos, count) != 0)
+ return -1;
+ if (realloc_floats(&ws->rope_sin, count) != 0)
+ return -1;
+
+ for (int pos = ws->rope_cache_cap; pos < cap; pos++) {
+ float fp = (float)pos;
+ float *c = ws->rope_cos + (size_t)pos * head_dim;
+ float *s = ws->rope_sin + (size_t)pos * head_dim;
+ for (int d = 0; d < half; d++) {
+ float inv = 1.0f / powf(theta, (float)(2 * d) / (float)head_dim);
+ float angle = fp * inv;
+ float cv = cosf(angle), sv = sinf(angle);
+ c[d] = cv;
+ c[half + d] = cv;
+ s[d] = sv;
+ s[half + d] = sv;
+ }
+ }
+
+ ws->rope_cache_cap = cap;
+ return 0;
+}
+
+static int ensure_offsets(embed_workspace_t *ws, int batch) {
+ if (!ws || batch <= 0 || batch == INT_MAX)
+ return -1;
+
+ int needed = batch + 1;
+ if (needed <= ws->offsets_cap)
+ return 0;
+
+ if (realloc_ints(&ws->offsets, (size_t)needed) != 0)
+ return -1;
+ ws->offsets_cap = needed;
+ return 0;
+}
+
+/* ========================================================================
+ * Model / workspace load and free
+ * ======================================================================== */
+
+/* BERT-family weight loader. The tensor names differ entirely from the Qwen
+ * block (encoder.layer.N.*, embeddings.*), so the family gets its own entry
+ * point. Reused embed_layer_t slots carry their BERT meaning: wo is the
+ * attention output dense, input_norm/post_attn_norm are the two block LayerNorm
+ * weights, up_proj/down_proj are the two FFN matrices (see embed_internal.h). */
+static int
+load_bert_weights(embed_model_t *model, multi_safetensors_t *ms, int layer_start, int layer_end) {
+ const embed_config_t *cfg = &model->config;
+ embed_weights_t *w = &model->weights;
+ char name[256];
+
+ /* sentence-transformers checkpoints store the BERT submodule unprefixed;
+ * raw HF checkpoints prefix every tensor with "bert.". */
+ const char *p;
+ if (multi_safetensors_find(ms, "embeddings.word_embeddings.weight", NULL))
+ p = "";
+ else if (multi_safetensors_find(ms, "bert.embeddings.word_embeddings.weight", NULL))
+ p = "bert.";
+ else {
+ fprintf(stderr, "embed_model_load: BERT word embeddings not found\n");
+ return -1;
+ }
+
+ int hidden = cfg->hidden_size, inter = cfg->intermediate_size;
+
+ if (layer_start == 0) {
+ const int64_t emb_shape[2] = {cfg->vocab_size, hidden};
+ snprintf(name, sizeof(name), "%sembeddings.word_embeddings.weight", p);
+ w->embed_tokens = load_weight_direct(ms, name, emb_shape, 2);
+ const int64_t pos_shape[2] = {cfg->max_position_embeddings, hidden};
+ snprintf(name, sizeof(name), "%sembeddings.position_embeddings.weight", p);
+ w->position_embeddings = load_norm_f32(ms, name, pos_shape, 2);
+ const int64_t type_shape[2] = {2, hidden};
+ snprintf(name, sizeof(name), "%sembeddings.token_type_embeddings.weight", p);
+ w->token_type_embeddings = load_norm_f32(ms, name, type_shape, 2);
+ const int64_t ln_shape[1] = {hidden};
+ snprintf(name, sizeof(name), "%sembeddings.LayerNorm.weight", p);
+ w->embed_ln_w = load_norm_f32(ms, name, ln_shape, 1);
+ snprintf(name, sizeof(name), "%sembeddings.LayerNorm.bias", p);
+ w->embed_ln_b = load_norm_f32(ms, name, ln_shape, 1);
+ if (!w->embed_tokens.data || !w->position_embeddings || !w->token_type_embeddings ||
+ !w->embed_ln_w || !w->embed_ln_b)
+ return -1;
+ }
+
+ w->layers = (embed_layer_t *)calloc(cfg->n_layers, sizeof(embed_layer_t));
+ if (!w->layers)
+ return -1;
+
+ for (int i = layer_start; i < layer_end; i++) {
+ embed_layer_t *l = &w->layers[i];
+
+#define BERT_W(field, suffix, d0, d1) \
+ do { \
+ const int64_t e[2] = {(d0), (d1)}; \
+ snprintf(name, sizeof(name), "%sencoder.layer.%d." suffix, p, i); \
+ l->field = load_weight_direct(ms, name, e, 2); \
+ if (!l->field.data) \
+ return -1; \
+ } while (0)
+#define BERT_V(field, suffix, d0) \
+ do { \
+ const int64_t e[1] = {(d0)}; \
+ snprintf(name, sizeof(name), "%sencoder.layer.%d." suffix, p, i); \
+ l->field = load_norm_f32(ms, name, e, 1); \
+ if (!l->field) \
+ return -1; \
+ } while (0)
+
+ BERT_W(wq, "attention.self.query.weight", hidden, hidden);
+ BERT_V(q_bias, "attention.self.query.bias", hidden);
+ BERT_W(wk, "attention.self.key.weight", hidden, hidden);
+ BERT_V(k_bias, "attention.self.key.bias", hidden);
+ BERT_W(wv, "attention.self.value.weight", hidden, hidden);
+ BERT_V(v_bias, "attention.self.value.bias", hidden);
+ BERT_W(wo, "attention.output.dense.weight", hidden, hidden);
+ BERT_V(o_bias, "attention.output.dense.bias", hidden);
+ BERT_V(input_norm, "attention.output.LayerNorm.weight", hidden);
+ BERT_V(attn_ln_bias, "attention.output.LayerNorm.bias", hidden);
+ BERT_W(up_proj, "intermediate.dense.weight", inter, hidden);
+ BERT_V(ffn_inter_bias, "intermediate.dense.bias", inter);
+ BERT_W(down_proj, "output.dense.weight", hidden, inter);
+ BERT_V(ffn_out_bias, "output.dense.bias", hidden);
+ BERT_V(post_attn_norm, "output.LayerNorm.weight", hidden);
+ BERT_V(ffn_ln_bias, "output.LayerNorm.bias", hidden);
+
+#undef BERT_W
+#undef BERT_V
+ }
+ /* BERT has no final norm: w->norm stays NULL, and pooling skips it. */
+ return 0;
+}
+
+static embed_model_t *
+model_load_range_ex(const char *model_dir, int layer_start, int layer_end, int allow_late) {
+ if (!allow_late && model_dir_has_late_projection(model_dir)) {
+ fprintf(stderr, "embed_model_load: late-interaction models require "
+ "token-level MaxSim support and are not valid pooled "
+ "embedding models yet\n");
+ return NULL;
+ }
+
+ embed_model_t *model = (embed_model_t *)calloc(1, sizeof(embed_model_t));
+ if (!model)
+ return NULL;
+
+ /* Parse config.json */
+ if (embed_config_parse(&model->config, model_dir) != 0) {
+ free(model);
+ return NULL;
+ }
+
+ const embed_config_t *cfg = &model->config;
+ if (layer_end == -1)
+ layer_end = cfg->n_layers;
+ if (layer_start < 0 || layer_start >= layer_end || layer_end > cfg->n_layers) {
+ fprintf(stderr,
+ "embed_model_load: invalid layer range [%d, %d) "
+ "for %d-layer model\n",
+ layer_start, layer_end, cfg->n_layers);
+ free(model);
+ return NULL;
+ }
+ model->layer_start = layer_start;
+ model->layer_end = layer_end;
+ model->attention = cfg->attention_mode == EMBED_ATTENTION_CAUSAL
+ ? embed_causal_gqa_attention_packed_with_scratch
+ : embed_bidirectional_gqa_attention_packed_with_scratch;
+
+ /* Open safetensors (handles single file or multi-shard) */
+ multi_safetensors_t *ms = multi_safetensors_open(model_dir);
+ if (!ms) {
+ fprintf(stderr, "embed_model_load: failed to open safetensors in %s\n", model_dir);
+ free(model);
+ return NULL;
+ }
+ model->safetensors = ms;
+
+ if (cfg->family == EMBED_FAMILY_BERT) {
+ if (load_bert_weights(model, ms, layer_start, layer_end) != 0)
+ goto fail;
+ if (embed_verbose >= 1)
+ fprintf(stderr, "embed_model_load: BERT layers [%d, %d) loaded (%d-dim)\n", layer_start,
+ layer_end, cfg->hidden_size);
+ return model;
+ }
+
+ const char *weight_prefix = multi_safetensors_weight_prefix(ms, "embed_tokens.weight");
+ if (!weight_prefix) {
+ fprintf(stderr, "embed_model_load: neither embed_tokens.weight nor "
+ "model.embed_tokens.weight was found\n");
+ goto fail;
+ }
+
+ if (embed_verbose >= 1)
+ fprintf(stderr, "embed_model_load: loading weights from %s\n", model_dir);
+
+ char name[256];
+
+ /* Only the first stage needs token embeddings. */
+ embed_weights_t *w = &model->weights;
+ if (layer_start == 0) {
+ const int64_t embed_shape[2] = {cfg->vocab_size, cfg->hidden_size};
+ snprintf(name, sizeof(name), "%sembed_tokens.weight", weight_prefix);
+ w->embed_tokens = load_weight_direct(ms, name, embed_shape, 2);
+ if (!w->embed_tokens.data)
+ goto fail;
+ }
+
+ /* Allocate layer array */
+ w->layers = (embed_layer_t *)calloc(cfg->n_layers, sizeof(embed_layer_t));
+ if (!w->layers)
+ goto fail;
+
+ /* Per-layer weights */
+ for (int i = layer_start; i < layer_end; i++) {
+ embed_layer_t *l = &w->layers[i];
+
+#define LOAD_NORM(field, fmt, d0) \
+ do { \
+ const int64_t expect[1] = {(d0)}; \
+ snprintf(name, sizeof(name), fmt, weight_prefix, i); \
+ l->field = load_norm_f32(ms, name, expect, 1); \
+ if (!l->field) \
+ goto fail; \
+ } while (0)
+
+#define LOAD2(field, fmt, d0, d1) \
+ do { \
+ const int64_t expect[2] = {(d0), (d1)}; \
+ snprintf(name, sizeof(name), fmt, weight_prefix, i); \
+ l->field = load_weight_direct(ms, name, expect, 2); \
+ if (!l->field.data) \
+ goto fail; \
+ } while (0)
+
+ LOAD2(wq, "%slayers.%d.self_attn.q_proj.weight", cfg->q_dim, cfg->hidden_size);
+ LOAD2(wk, "%slayers.%d.self_attn.k_proj.weight", cfg->kv_dim, cfg->hidden_size);
+ LOAD2(wv, "%slayers.%d.self_attn.v_proj.weight", cfg->kv_dim, cfg->hidden_size);
+ LOAD2(wo, "%slayers.%d.self_attn.o_proj.weight", cfg->hidden_size, cfg->q_dim);
+ if (cfg->qkv_bias) {
+ LOAD_NORM(q_bias, "%slayers.%d.self_attn.q_proj.bias", cfg->q_dim);
+ LOAD_NORM(k_bias, "%slayers.%d.self_attn.k_proj.bias", cfg->kv_dim);
+ LOAD_NORM(v_bias, "%slayers.%d.self_attn.v_proj.bias", cfg->kv_dim);
+ }
+ if (cfg->qk_norm) {
+ LOAD_NORM(q_norm, "%slayers.%d.self_attn.q_norm.weight", cfg->head_dim);
+ LOAD_NORM(k_norm, "%slayers.%d.self_attn.k_norm.weight", cfg->head_dim);
+ }
+ LOAD_NORM(input_norm, "%slayers.%d.input_layernorm.weight", cfg->hidden_size);
+ LOAD_NORM(post_attn_norm, "%slayers.%d.post_attention_layernorm.weight", cfg->hidden_size);
+ LOAD2(gate_proj, "%slayers.%d.mlp.gate_proj.weight", cfg->intermediate_size,
+ cfg->hidden_size);
+ LOAD2(up_proj, "%slayers.%d.mlp.up_proj.weight", cfg->intermediate_size, cfg->hidden_size);
+ LOAD2(down_proj, "%slayers.%d.mlp.down_proj.weight", cfg->hidden_size,
+ cfg->intermediate_size);
+
+#undef LOAD2
+#undef LOAD_NORM
+
+ if (embed_verbose >= 2)
+ fprintf(stderr, " layer %d loaded\n", i);
+ }
+
+ /* Only the final stage applies the output RMSNorm. */
+ if (layer_end == cfg->n_layers) {
+ const int64_t norm_shape[1] = {cfg->hidden_size};
+ snprintf(name, sizeof(name), "%snorm.weight", weight_prefix);
+ w->norm = load_norm_f32(ms, name, norm_shape, 1);
+ if (!w->norm)
+ goto fail;
+ }
+
+ if (embed_verbose >= 1)
+ fprintf(stderr,
+ "embed_model_load: layers [%d, %d) loaded "
+ "(%d-dim embeddings)\n",
+ layer_start, layer_end, cfg->hidden_size);
+
+ return model;
+
+fail:
+ embed_model_free(model);
+ return NULL;
+}
+
+embed_model_t *embed_model_load(const char *model_dir) {
+ return model_load_range_ex(model_dir, 0, -1, 0);
+}
+
+void embed_model_free(embed_model_t *model) {
+ if (!model)
+ return;
+ if (model->weights.layers) {
+ for (int i = 0; i < model->config.n_layers; i++) {
+ embed_layer_t *l = &model->weights.layers[i];
+ free((void *)l->q_bias);
+ free((void *)l->k_bias);
+ free((void *)l->v_bias);
+ free((void *)l->q_norm);
+ free((void *)l->k_norm);
+ free((void *)l->input_norm);
+ free((void *)l->post_attn_norm);
+ free((void *)l->o_bias);
+ free((void *)l->attn_ln_bias);
+ free((void *)l->ffn_inter_bias);
+ free((void *)l->ffn_out_bias);
+ free((void *)l->ffn_ln_bias);
+ }
+ }
+ free((void *)model->weights.norm);
+ free((void *)model->weights.position_embeddings);
+ free((void *)model->weights.token_type_embeddings);
+ free((void *)model->weights.embed_ln_w);
+ free((void *)model->weights.embed_ln_b);
+ if (model->safetensors)
+ multi_safetensors_close((multi_safetensors_t *)model->safetensors);
+ free(model->weights.layers);
+ free(model);
+}
+
+embed_workspace_t *embed_workspace_new(const embed_model_t *model) {
+ if (!model)
+ return NULL;
+ embed_workspace_t *ws = (embed_workspace_t *)calloc(1, sizeof(embed_workspace_t));
+ if (!ws)
+ return NULL;
+ ws->model = model;
+ return ws;
+}
+
+void embed_workspace_free(embed_workspace_t *ws) {
+ if (!ws)
+ return;
+ free(ws->x);
+ free(ws->x_norm);
+ free(ws->q);
+ free(ws->k);
+ free(ws->v);
+ free(ws->attn_out);
+ free(ws->proj_out);
+ free(ws->ffn_gate);
+ free(ws->ffn_up);
+ free(ws->attn_scores);
+ free(ws->offsets);
+ free(ws->rope_cos);
+ free(ws->rope_sin);
+ free(ws->bf16_widen);
+ free(ws);
+}
+
+const embed_config_t *embed_model_config(const embed_model_t *model) {
+ return model ? &model->config : NULL;
+}
+
+static const safetensor_t *find_single_safetensor(const safetensors_file_t *sf, const char *name) {
+ if (!sf || !name)
+ return NULL;
+ for (int i = 0; i < sf->num_tensors; i++) {
+ if (strcmp(sf->tensors[i].name, name) == 0)
+ return &sf->tensors[i];
+ }
+ return NULL;
+}
+
+embed_late_model_t *embed_late_model_load(const char *model_dir) {
+ if (!model_dir_has_late_projection(model_dir)) {
+ fprintf(stderr, "embed_late_model_load: missing 1_Dense projection\n");
+ return NULL;
+ }
+
+ embed_late_model_t *late = (embed_late_model_t *)calloc(1, sizeof(*late));
+ if (!late)
+ return NULL;
+
+ late->base = model_load_range_ex(model_dir, 0, -1, 1);
+ if (!late->base)
+ goto fail;
+
+ char path[1024];
+ int n = snprintf(path, sizeof(path), "%s/1_Dense/model.safetensors", model_dir);
+ if (n < 0 || (size_t)n >= sizeof(path)) {
+ fprintf(stderr, "embed_late_model_load: model path too long\n");
+ goto fail;
+ }
+
+ late->dense_sf = safetensors_open(path);
+ if (!late->dense_sf) {
+ fprintf(stderr, "embed_late_model_load: cannot open %s\n", path);
+ goto fail;
+ }
+
+ const safetensor_t *t = find_single_safetensor(late->dense_sf, "linear.weight");
+ if (!t) {
+ fprintf(stderr, "embed_late_model_load: missing linear.weight\n");
+ goto fail;
+ }
+ if (t->ndim != 2 || t->shape[0] <= 0 || t->shape[0] > INT_MAX) {
+ fprintf(stderr, "embed_late_model_load: bad projection shape\n");
+ goto fail;
+ }
+
+ int64_t shape[2] = {t->shape[0], late->base->config.hidden_size};
+ if (!tensor_has_supported_shape(late->dense_sf, t, "linear.weight", shape, 2, 1))
+ goto fail;
+
+ late->projection.dtype = t->dtype;
+ late->projection.data = safetensors_data(late->dense_sf, t);
+ late->token_dim = (int)t->shape[0];
+ return late;
+
+fail:
+ embed_late_model_free(late);
+ return NULL;
+}
+
+void embed_late_model_free(embed_late_model_t *model) {
+ if (!model)
+ return;
+ embed_model_free(model->base);
+ safetensors_close(model->dense_sf);
+ free(model);
+}
+
+embed_late_workspace_t *embed_late_workspace_new(const embed_late_model_t *model) {
+ if (!model || !model->base)
+ return NULL;
+ embed_late_workspace_t *ws = (embed_late_workspace_t *)calloc(1, sizeof(*ws));
+ if (!ws)
+ return NULL;
+ ws->model = model;
+ ws->base_ws = embed_workspace_new(model->base);
+ if (!ws->base_ws) {
+ free(ws);
+ return NULL;
+ }
+ return ws;
+}
+
+void embed_late_workspace_free(embed_late_workspace_t *ws) {
+ if (!ws)
+ return;
+ embed_workspace_free(ws->base_ws);
+ free(ws->states);
+ free(ws);
+}
+
+const embed_config_t *embed_late_model_config(const embed_late_model_t *model) {
+ return model && model->base ? &model->base->config : NULL;
+}
+
+int embed_late_model_token_dim(const embed_late_model_t *model) {
+ return model ? model->token_dim : 0;
+}
+
+/* Internal (embed_internal.h): expose the base model and projection so GPU
+ * backends can upload them to device without seeing struct embed_late_model. */
+embed_model_t *embed_late_model_base(const embed_late_model_t *model) {
+ return model ? model->base : NULL;
+}
+
+const embed_weight_ref_t *embed_late_model_projection(const embed_late_model_t *model) {
+ return model ? &model->projection : NULL;
+}
+
+int embed_late_model_encode_tokens(const embed_late_model_t *model,
+ embed_late_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens,
+ int normalize,
+ float *out_vectors) {
+ if (!model || !ws || ws->model != model || !token_ids || n_tokens <= 0 || !out_vectors ||
+ model->token_dim <= 0)
+ return -1;
+
+ int hidden = model->base->config.hidden_size;
+ int dim = model->token_dim;
+ if (ensure_late_state_buffer(ws, n_tokens) != 0)
+ return -1;
+ if (embed_model_forward_into(model->base, ws->base_ws, token_ids, n_tokens, ws->states) != 0)
+ return -1;
+
+ linear_nobias_weight(ws->base_ws, out_vectors, ws->states, &model->projection, n_tokens, hidden,
+ dim);
+ if (normalize) {
+ for (int i = 0; i < n_tokens; i++) {
+ if (embed_l2_normalize(out_vectors + (size_t)i * dim, dim) != 0)
+ return -1;
+ }
+ }
+ return 0;
+}
+
+float embed_late_maxsim(const float *query_vectors,
+ int query_tokens,
+ const float *doc_vectors,
+ int doc_tokens,
+ int dim) {
+ if (!query_vectors || !doc_vectors || query_tokens <= 0 || doc_tokens <= 0 || dim <= 0)
+ return 0.0f;
+
+ float score = 0.0f;
+ for (int qi = 0; qi < query_tokens; qi++) {
+ const float *q = query_vectors + (size_t)qi * dim;
+ float best = -FLT_MAX;
+ for (int di = 0; di < doc_tokens; di++) {
+ const float *d = doc_vectors + (size_t)di * dim;
+ float dot = embed_dot_f32(q, d, dim);
+ if (dot > best)
+ best = dot;
+ }
+ score += best;
+ }
+ return score;
+}
+
+int embed_late_maxsim_batch(const float *query_vectors,
+ int query_tokens,
+ const float *doc_vectors,
+ const int *doc_offsets,
+ int docs,
+ int dim,
+ float *scores) {
+ if (!query_vectors || !doc_vectors || !doc_offsets || !scores || query_tokens <= 0 ||
+ docs <= 0 || dim <= 0 || doc_offsets[0] != 0)
+ return -1;
+
+ for (int i = 0; i < docs; i++) {
+ if (doc_offsets[i] < 0 || doc_offsets[i + 1] <= doc_offsets[i])
+ return -1;
+ }
+
+#ifdef USE_BLAS
+ /* One similarity GEMM per group of whole candidates:
+ * S[query_tokens, group_tokens] = Q @ D_group^T, then per-document
+ * row-max plus sum. The group token budget bounds the scratch matrix;
+ * one oversized document still runs alone in a group of its own. */
+ enum { LATE_GROUP_TOKEN_BUDGET = 4096 };
+ int total = doc_offsets[docs];
+ int budget = LATE_GROUP_TOKEN_BUDGET;
+ size_t cap = (size_t)query_tokens * (size_t)(total < budget ? total : budget);
+ int largest = 0;
+ for (int i = 0; i < docs; i++) {
+ int len = doc_offsets[i + 1] - doc_offsets[i];
+ if (len > largest)
+ largest = len;
+ }
+ if ((size_t)query_tokens * (size_t)largest > cap)
+ cap = (size_t)query_tokens * (size_t)largest;
+ float *sim = (float *)malloc(cap * sizeof(float));
+ if (sim) {
+ int doc = 0;
+ while (doc < docs) {
+ int first = doc;
+ int start = doc_offsets[first];
+ doc++;
+ while (doc < docs && doc_offsets[doc + 1] - start <= budget &&
+ doc_offsets[doc + 1] - start <= (int)(cap / (size_t)query_tokens))
+ doc++;
+ int group_tokens = doc_offsets[doc] - start;
+
+ embed_matmul_t(sim, query_vectors, doc_vectors + (size_t)start * dim, query_tokens, dim,
+ group_tokens);
+
+ for (int i = first; i < doc; i++) {
+ int s0 = doc_offsets[i] - start;
+ int s1 = doc_offsets[i + 1] - start;
+ float score = 0.0f;
+ for (int qi = 0; qi < query_tokens; qi++) {
+ const float *row = sim + (size_t)qi * group_tokens;
+ float best = row[s0];
+ for (int t = s0 + 1; t < s1; t++)
+ if (row[t] > best)
+ best = row[t];
+ score += best;
+ }
+ scores[i] = score;
+ }
+ }
+ free(sim);
+ return 0;
+ }
+ /* Allocation failure: fall through to the scalar path. */
+#endif
+
+ for (int i = 0; i < docs; i++) {
+ int start = doc_offsets[i];
+ int end = doc_offsets[i + 1];
+ scores[i] = embed_late_maxsim(query_vectors, query_tokens,
+ doc_vectors + (size_t)start * dim, end - start, dim);
+ }
+ return 0;
+}
+
+size_t embed_workspace_nbytes(const embed_workspace_t *ws) {
+ if (!ws)
+ return 0;
+
+ size_t total = sizeof(*ws);
+ const embed_model_t *model = ws->model;
+ if (!model)
+ return total;
+
+ const embed_config_t *c = &model->config;
+ size_t row_floats = 0;
+ const int dims[] = {
+ c->hidden_size, /* x */
+ c->hidden_size, /* x_norm */
+ c->q_dim, /* q */
+ c->kv_dim, /* k */
+ c->kv_dim, /* v */
+ c->q_dim, /* attn_out */
+ c->hidden_size, /* proj_out */
+ c->intermediate_size, /* ffn_gate */
+ c->intermediate_size, /* ffn_up */
+ };
+
+ for (size_t i = 0; i < sizeof(dims) / sizeof(dims[0]); i++) {
+ if (dims[i] < 0 || add_size(row_floats, (size_t)dims[i], &row_floats) != 0)
+ return SIZE_MAX;
+ }
+
+ size_t buf_floats = 0;
+ if (ws->buf_seq_cap > 0 && mul_size((size_t)ws->buf_seq_cap, row_floats, &buf_floats) != 0)
+ return SIZE_MAX;
+
+ size_t rope_floats = 0;
+ if (ws->rope_cache_cap > 0) {
+ size_t rope_rows = 0;
+ if (mul_size((size_t)ws->rope_cache_cap, (size_t)c->head_dim, &rope_rows) != 0 ||
+ mul_size(rope_rows, 2, &rope_floats) != 0)
+ return SIZE_MAX;
+ }
+
+ size_t all_floats = 0, bytes = 0;
+ if (add_size(buf_floats, rope_floats, &all_floats) != 0 ||
+ mul_size(all_floats, sizeof(float), &bytes) != 0 || add_size(total, bytes, &total) != 0)
+ return SIZE_MAX;
+
+ size_t offset_bytes = 0;
+ if (ws->offsets_cap > 0 && mul_size((size_t)ws->offsets_cap, sizeof(int), &offset_bytes) != 0)
+ return SIZE_MAX;
+ if (add_size(total, offset_bytes, &total) != 0)
+ return SIZE_MAX;
+ if (add_size(total, ws->attn_scores_bytes, &total) != 0)
+ return SIZE_MAX;
+
+ size_t widen_bytes = 0;
+ if (ws->bf16_widen_count > 0 &&
+ mul_size(ws->bf16_widen_count, sizeof(float), &widen_bytes) != 0)
+ return SIZE_MAX;
+ if (add_size(total, widen_bytes, &total) != 0)
+ return SIZE_MAX;
+
+ return total;
+}
+
+/* ========================================================================
+ * Packed forward pass
+ * ======================================================================== */
+
+static int build_offsets(const embed_input_t *inputs,
+ int batch,
+ int require_ids,
+ int *offsets,
+ int *total_out,
+ int *max_out) {
+ if (!inputs || batch <= 0 || !offsets || !total_out || !max_out)
+ return -1;
+
+ int total = 0;
+ int max_seq = 0;
+ for (int b = 0; b < batch; b++) {
+ if ((require_ids && !inputs[b].ids) || inputs[b].n_tokens <= 0)
+ return -1;
+ if (total > INT32_MAX - inputs[b].n_tokens)
+ return -1;
+ offsets[b] = total;
+ total += inputs[b].n_tokens;
+ if (inputs[b].n_tokens > max_seq)
+ max_seq = inputs[b].n_tokens;
+ }
+ offsets[batch] = total;
+ *total_out = total;
+ *max_out = max_seq;
+ return 0;
+}
+
+static int forward_packed_slice_inplace(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const embed_input_t *inputs,
+ int batch,
+ const int *offsets,
+ int total_seq,
+ int max_seq,
+ const float *input_states,
+ int layer_start,
+ int layer_end,
+ int apply_final_norm) {
+ if (!model || !ws || ws->model != model || !inputs || batch <= 0 || !offsets ||
+ total_seq <= 0 || max_seq <= 0)
+ return -1;
+
+ const embed_config_t *cfg = &model->config;
+ const embed_weights_t *w = &model->weights;
+ if (layer_start < 0 || layer_start > layer_end || layer_end > cfg->n_layers ||
+ layer_start < model->layer_start || layer_end > model->layer_end ||
+ (layer_start == 0 && input_states) || (layer_start != 0 && !input_states) ||
+ (layer_start == 0 && !w->embed_tokens.data) ||
+ (apply_final_norm && (layer_end != cfg->n_layers || !w->norm)))
+ return -1;
+
+ int hidden = cfg->hidden_size;
+ int n_heads = cfg->n_heads;
+ int n_kv_heads = cfg->n_kv_heads;
+ int head_dim = cfg->head_dim;
+ int q_dim = cfg->q_dim;
+ int kv_dim = cfg->kv_dim;
+ int inter = cfg->intermediate_size;
+ float eps = cfg->rms_norm_eps;
+
+ if (ensure_buffers(ws, cfg, total_seq) != 0)
+ return -1;
+ if (ensure_rope_cache(ws, cfg, max_seq) != 0)
+ return -1;
+ ensure_attention_scores(ws, offsets, batch);
+
+ const float *rope_cos = ws->rope_cos;
+ const float *rope_sin = ws->rope_sin;
+
+ float *x = ws->x;
+ float *x_norm = ws->x_norm;
+ float *q_buf = ws->q;
+ float *k_buf = ws->k;
+ float *v_buf = ws->v;
+ float *attn_out = ws->attn_out;
+ float *proj_out = ws->proj_out;
+ float *ffn_gate = ws->ffn_gate;
+ float *ffn_up = ws->ffn_up;
+
+ /* A first stage performs embedding lookup. Later stages resume from a
+ * packed hidden-state tensor received from the preceding stage. */
+ if (layer_start == 0) {
+ for (int b = 0; b < batch; b++) {
+ int start = offsets[b];
+ for (int i = 0; i < inputs[b].n_tokens; i++) {
+ int id = inputs[b].ids[i];
+ if (id < 0 || id >= cfg->vocab_size) {
+ fprintf(stderr, "embed: invalid token id %d at input %d position %d\n", id, b,
+ i);
+ return -1;
+ }
+ copy_weight_row(x + (size_t)(start + i) * hidden, &w->embed_tokens, (size_t)id,
+ hidden);
+ }
+ }
+ } else {
+ memcpy(x, input_states, (size_t)total_seq * hidden * sizeof(float));
+ }
+
+ float scale = 1.0f / sqrtf((float)head_dim);
+
+ /* 2. Transformer layers. Linear/MLP ops run over all packed tokens;
+ * attention is explicitly block-diagonal by sequence offsets. */
+ for (int layer = layer_start; layer < layer_end; layer++) {
+ const embed_layer_t *l = &w->layers[layer];
+
+ embed_rms_norm(x_norm, x, l->input_norm, total_seq, hidden, eps);
+
+ linear_qkv_weight(ws, q_buf, k_buf, v_buf, x_norm, &l->wq, &l->wk, &l->wv, total_seq,
+ hidden, q_dim, kv_dim);
+
+ add_bias_rows(q_buf, l->q_bias, total_seq, q_dim);
+ add_bias_rows(k_buf, l->k_bias, total_seq, kv_dim);
+ add_bias_rows(v_buf, l->v_bias, total_seq, kv_dim);
+ if (cfg->qk_norm) {
+ embed_rms_norm_per_head(q_buf, l->q_norm, total_seq, n_heads, head_dim, eps);
+ embed_rms_norm_per_head(k_buf, l->k_norm, total_seq, n_kv_heads, head_dim, eps);
+ }
+
+ for (int b = 0; b < batch; b++) {
+ int start = offsets[b];
+ int len = offsets[b + 1] - start;
+ embed_apply_rope_neox(q_buf + (size_t)start * q_dim, rope_cos, rope_sin, len, n_heads,
+ head_dim);
+ embed_apply_rope_neox(k_buf + (size_t)start * kv_dim, rope_cos, rope_sin, len,
+ n_kv_heads, head_dim);
+ }
+
+ model->attention(attn_out, q_buf, k_buf, v_buf, offsets, batch, n_heads, n_kv_heads,
+ head_dim, scale, ws->attn_scores, ws->attn_scores_bytes);
+
+ linear_nobias_weight(ws, proj_out, attn_out, &l->wo, total_seq, q_dim, hidden);
+ embed_add_inplace(x, proj_out, total_seq * hidden);
+
+ embed_rms_norm(x_norm, x, l->post_attn_norm, total_seq, hidden, eps);
+
+ linear_pair_weight(ws, ffn_gate, ffn_up, x_norm, &l->gate_proj, &l->up_proj, total_seq,
+ hidden, inter, inter);
+ embed_silu_mul_inplace(ffn_gate, ffn_up, total_seq * inter);
+ linear_nobias_weight(ws, proj_out, ffn_gate, &l->down_proj, total_seq, inter, hidden);
+ embed_add_inplace(x, proj_out, total_seq * hidden);
+ }
+
+ if (apply_final_norm)
+ embed_rms_norm(x, x, w->norm, total_seq, hidden, eps);
+ return 0;
+}
+
+static int forward_packed_inplace(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const embed_input_t *inputs,
+ int batch,
+ const int *offsets,
+ int total_seq,
+ int max_seq,
+ int apply_final_norm) {
+ return forward_packed_slice_inplace(model, ws, inputs, batch, offsets, total_seq, max_seq, NULL,
+ 0, model->config.n_layers, apply_final_norm);
+}
+
+/* BERT-family forward: token + learned-position + token-type embeddings and an
+ * embedding LayerNorm, then post-norm encoder layers (bias on every projection,
+ * GeLU feed-forward, no RoPE, no per-head norm). Bidirectional attention is
+ * block-diagonal over the packed batch, so each sequence's positions restart at
+ * 0. Leaves the final hidden states in ws->x; pooling reads them as-is because
+ * the last layer's LayerNorm is already the model's final normalization. */
+static int forward_packed_bert(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const embed_input_t *inputs,
+ int batch,
+ const int *offsets,
+ int total_seq,
+ int max_seq) {
+ if (!model || !ws || ws->model != model || !inputs || batch <= 0 || !offsets ||
+ total_seq <= 0 || max_seq <= 0)
+ return -1;
+
+ const embed_config_t *cfg = &model->config;
+ const embed_weights_t *w = &model->weights;
+ int hidden = cfg->hidden_size;
+ int n_heads = cfg->n_heads;
+ int head_dim = cfg->head_dim;
+ int inter = cfg->intermediate_size;
+ float eps = cfg->layer_norm_eps;
+
+ if (!w->embed_tokens.data || !w->position_embeddings || !w->token_type_embeddings ||
+ !w->embed_ln_w || !w->embed_ln_b)
+ return -1;
+ if (max_seq > cfg->max_position_embeddings) {
+ fprintf(stderr, "embed: sequence length %d exceeds max_position_embeddings %d\n", max_seq,
+ cfg->max_position_embeddings);
+ return -1;
+ }
+ if (ensure_buffers(ws, cfg, total_seq) != 0)
+ return -1;
+ ensure_attention_scores(ws, offsets, batch);
+
+ float *x = ws->x;
+ float *q_buf = ws->q, *k_buf = ws->k, *v_buf = ws->v;
+ float *attn_out = ws->attn_out, *proj_out = ws->proj_out, *ffn_up = ws->ffn_up;
+
+ /* Embeddings: token + learned absolute position + token-type[0], then the
+ * embedding LayerNorm over every packed token. */
+ for (int b = 0; b < batch; b++) {
+ int start = offsets[b];
+ for (int i = 0; i < inputs[b].n_tokens; i++) {
+ int id = inputs[b].ids[i];
+ if (id < 0 || id >= cfg->vocab_size) {
+ fprintf(stderr, "embed: invalid token id %d at input %d position %d\n", id, b, i);
+ return -1;
+ }
+ float *row = x + (size_t)(start + i) * hidden;
+ copy_weight_row(row, &w->embed_tokens, (size_t)id, hidden);
+ embed_add_inplace(row, w->position_embeddings + (size_t)i * hidden, hidden);
+ embed_add_inplace(row, w->token_type_embeddings, hidden);
+ }
+ }
+ embed_layer_norm(x, x, w->embed_ln_w, w->embed_ln_b, total_seq, hidden, eps);
+
+ float scale = 1.0f / sqrtf((float)head_dim);
+
+ for (int layer = 0; layer < cfg->n_layers; layer++) {
+ const embed_layer_t *l = &w->layers[layer];
+
+ /* Self-attention (post-norm): projections read x directly. */
+ linear_qkv_weight(ws, q_buf, k_buf, v_buf, x, &l->wq, &l->wk, &l->wv, total_seq, hidden,
+ hidden, hidden);
+ add_bias_rows(q_buf, l->q_bias, total_seq, hidden);
+ add_bias_rows(k_buf, l->k_bias, total_seq, hidden);
+ add_bias_rows(v_buf, l->v_bias, total_seq, hidden);
+
+ model->attention(attn_out, q_buf, k_buf, v_buf, offsets, batch, n_heads, n_heads, head_dim,
+ scale, ws->attn_scores, ws->attn_scores_bytes);
+
+ linear_nobias_weight(ws, proj_out, attn_out, &l->wo, total_seq, hidden, hidden);
+ add_bias_rows(proj_out, l->o_bias, total_seq, hidden);
+ embed_add_inplace(x, proj_out, total_seq * hidden);
+ embed_layer_norm(x, x, l->input_norm, l->attn_ln_bias, total_seq, hidden, eps);
+
+ /* Feed-forward (post-norm): dense -> GeLU -> dense. */
+ linear_nobias_weight(ws, ffn_up, x, &l->up_proj, total_seq, hidden, inter);
+ add_bias_rows(ffn_up, l->ffn_inter_bias, total_seq, inter);
+ if (cfg->ffn_act == EMBED_ACT_GELU_TANH)
+ embed_gelu_tanh_inplace(ffn_up, total_seq * inter);
+ else
+ embed_gelu_inplace(ffn_up, total_seq * inter);
+ linear_nobias_weight(ws, proj_out, ffn_up, &l->down_proj, total_seq, inter, hidden);
+ add_bias_rows(proj_out, l->ffn_out_bias, total_seq, hidden);
+ embed_add_inplace(x, proj_out, total_seq * hidden);
+ embed_layer_norm(x, x, l->post_attn_norm, l->ffn_ln_bias, total_seq, hidden, eps);
+ }
+ return 0;
+}
+
+static int pool_embeddings(const embed_model_t *model,
+ const embed_workspace_t *ws,
+ const int *offsets,
+ int batch,
+ float *out_embeddings) {
+ const embed_config_t *cfg = &model->config;
+ int hidden = cfg->hidden_size;
+ float eps = cfg->rms_norm_eps;
+ /* Qwen pools pre-final-norm states and applies the final RMSNorm here; BERT
+ * pools already-normed states (no final norm), signalled by a NULL norm. */
+ const float *norm_weight = model->weights.norm;
+ const float *x = ws->x;
+ if (hidden <= 0 || !x)
+ return -1;
+
+ for (int b = 0; b < batch; b++) {
+ int start = offsets[b];
+ int end = offsets[b + 1];
+ int len = end - start;
+ float *emb = out_embeddings + (size_t)b * hidden;
+
+ if (cfg->pooling_mode == EMBED_POOL_LAST_TOKEN || cfg->pooling_mode == EMBED_POOL_CLS) {
+ /* Single-token pooling: CLS takes the first token, last-token the
+ * final one. */
+ int idx = cfg->pooling_mode == EMBED_POOL_CLS ? start : end - 1;
+ const float *row = x + (size_t)idx * hidden;
+ if (norm_weight) {
+ float sum_sq = 0.0f;
+ for (int d = 0; d < hidden; d++)
+ sum_sq += row[d] * row[d];
+ float rms_inv = 1.0f / sqrtf(sum_sq / (float)hidden + eps);
+ for (int d = 0; d < hidden; d++)
+ emb[d] = row[d] * rms_inv * norm_weight[d];
+ } else {
+ memcpy(emb, row, (size_t)hidden * sizeof(float));
+ }
+ } else {
+ memset(emb, 0, (size_t)hidden * sizeof(float));
+ for (int i = start; i < end; i++) {
+ const float *row = x + (size_t)i * hidden;
+ if (norm_weight) {
+ float sum_sq = 0.0f;
+ for (int d = 0; d < hidden; d++)
+ sum_sq += row[d] * row[d];
+ float rms_inv = 1.0f / sqrtf(sum_sq / (float)hidden + eps);
+ for (int d = 0; d < hidden; d++)
+ emb[d] += row[d] * rms_inv * norm_weight[d];
+ } else {
+ for (int d = 0; d < hidden; d++)
+ emb[d] += row[d];
+ }
+ }
+ float inv_len = 1.0f / (float)len;
+ for (int d = 0; d < hidden; d++)
+ emb[d] *= inv_len;
+ }
+
+ if (cfg->normalize_embeddings && embed_l2_normalize(emb, hidden) != 0)
+ return -1;
+ }
+ return 0;
+}
+
+/* ========================================================================
+ * Public forward/embed APIs
+ * ======================================================================== */
+
+int embed_model_forward_into(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens,
+ float *out_states) {
+ if (!model || !ws || !token_ids || n_tokens <= 0 || !out_states)
+ return -1;
+
+ size_t count, bytes;
+ if (mul_size((size_t)n_tokens, (size_t)model->config.hidden_size, &count) != 0 ||
+ mul_size(count, sizeof(float), &bytes) != 0)
+ return -1;
+
+ embed_input_t input = {token_ids, n_tokens};
+ int offsets[2] = {0, n_tokens};
+ if (forward_packed_inplace(model, ws, &input, 1, offsets, n_tokens, n_tokens, 1) != 0)
+ return -1;
+
+ memcpy(out_states, ws->x, bytes);
+ return 0;
+}
+
+int embed_pool_spans(const embed_config_t *cfg,
+ const float *states,
+ int n_tokens,
+ const embed_span_t *spans,
+ int n_spans,
+ float *out_embeddings) {
+ if (!cfg || !states || n_tokens <= 0 || !spans || n_spans <= 0 || !out_embeddings)
+ return -1;
+
+ int hidden = cfg->hidden_size;
+ if (hidden <= 0)
+ return -1;
+
+ for (int s = 0; s < n_spans; s++) {
+ int start = spans[s].start;
+ int len = spans[s].n_tokens;
+ if (start < 0 || len <= 0 || start > n_tokens || len > n_tokens - start)
+ return -1;
+
+ float *emb = out_embeddings + (size_t)s * hidden;
+ memset(emb, 0, (size_t)hidden * sizeof(float));
+
+ for (int t = 0; t < len; t++) {
+ const float *row = states + (size_t)(start + t) * hidden;
+ for (int d = 0; d < hidden; d++)
+ emb[d] += row[d];
+ }
+
+ float inv_len = 1.0f / (float)len;
+ for (int d = 0; d < hidden; d++) {
+ emb[d] *= inv_len;
+ }
+ }
+ return 0;
+}
+
+int embed_model_encode_spans_batch(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const embed_context_input_t *inputs,
+ int batch,
+ float *out_embeddings) {
+ if (!model || !ws || !inputs || batch <= 0 || !out_embeddings)
+ return -1;
+
+ if (ensure_offsets(ws, batch) != 0)
+ return -1;
+ if ((size_t)batch > SIZE_MAX / sizeof(embed_input_t))
+ return -1;
+ embed_input_t *packed_inputs = malloc((size_t)batch * sizeof(*packed_inputs));
+ if (!packed_inputs)
+ return -1;
+ int *offsets = ws->offsets;
+ int total_seq = 0;
+ int max_seq = 0;
+ int total_spans = 0;
+ for (int b = 0; b < batch; b++) {
+ const embed_context_input_t *input = &inputs[b];
+ int n_tokens = input->input.n_tokens;
+ if (!input->input.ids || n_tokens <= 0 || !input->spans || input->n_spans <= 0 ||
+ total_seq > INT_MAX - n_tokens || total_spans > INT_MAX - input->n_spans) {
+ free(packed_inputs);
+ return -1;
+ }
+ packed_inputs[b] = input->input;
+ offsets[b] = total_seq;
+ total_seq += n_tokens;
+ total_spans += input->n_spans;
+ if (n_tokens > max_seq)
+ max_seq = n_tokens;
+ for (int s = 0; s < input->n_spans; s++) {
+ int start = input->spans[s].start;
+ int len = input->spans[s].n_tokens;
+ if (start < 0 || len <= 0 || start > n_tokens || len > n_tokens - start) {
+ free(packed_inputs);
+ return -1;
+ }
+ }
+ }
+ offsets[batch] = total_seq;
+
+ if (forward_packed_inplace(model, ws, packed_inputs, batch, offsets, total_seq, max_seq, 1) !=
+ 0) {
+ free(packed_inputs);
+ return -1;
+ }
+ free(packed_inputs);
+
+ int span_offset = 0;
+ int hidden = model->config.hidden_size;
+ for (int b = 0; b < batch; b++) {
+ const embed_context_input_t *input = &inputs[b];
+ if (embed_pool_spans(&model->config, ws->x + (size_t)offsets[b] * hidden,
+ input->input.n_tokens, input->spans, input->n_spans,
+ out_embeddings + (size_t)span_offset * hidden) != 0)
+ return -1;
+ span_offset += input->n_spans;
+ }
+ return 0;
+}
+
+int embed_model_encode_spans(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens,
+ const embed_span_t *spans,
+ int n_spans,
+ float *out_embeddings) {
+ embed_context_input_t input = {
+ .input = {token_ids, n_tokens},
+ .spans = spans,
+ .n_spans = n_spans,
+ };
+ return embed_model_encode_spans_batch(model, ws, &input, 1, out_embeddings);
+}
+
+float *embed_model_forward(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens) {
+ if (!model || !ws || !token_ids || n_tokens <= 0)
+ return NULL;
+
+ size_t n;
+ if (mul_size((size_t)n_tokens, (size_t)model->config.hidden_size, &n) != 0)
+ return NULL;
+ float *out = malloc_floats(n);
+ if (!out)
+ return NULL;
+
+ if (embed_model_forward_into(model, ws, token_ids, n_tokens, out) != 0) {
+ free(out);
+ return NULL;
+ }
+ return out;
+}
+
+int embed_model_encode_batch(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const embed_input_t *inputs,
+ int batch,
+ float *out_embeddings) {
+ if (!model || !ws || !inputs || batch <= 0 || !out_embeddings)
+ return -1;
+
+ if (ensure_offsets(ws, batch) != 0)
+ return -1;
+ int *offsets = ws->offsets;
+
+ int total_seq = 0;
+ int max_seq = 0;
+ if (build_offsets(inputs, batch, 1, offsets, &total_seq, &max_seq) != 0)
+ return -1;
+
+ int rc = model->config.family == EMBED_FAMILY_BERT
+ ? forward_packed_bert(model, ws, inputs, batch, offsets, total_seq, max_seq)
+ : forward_packed_inplace(model, ws, inputs, batch, offsets, total_seq, max_seq, 0);
+ if (rc == 0)
+ rc = pool_embeddings(model, ws, offsets, batch, out_embeddings);
+
+ return rc;
+}
+
+/* Late-interaction batch encoder. Lives here, beside the packed/ragged batch
+ * machinery it reuses, rather than next to embed_late_model_encode_tokens. */
+int embed_late_model_encode_docs(const embed_late_model_t *model,
+ embed_late_workspace_t *ws,
+ const int *const *doc_ids,
+ const int *n_tokens,
+ const int *const *keep,
+ const int *n_keep,
+ int n_docs,
+ int normalize,
+ float *out_vectors,
+ int *out_offsets) {
+ if (!model || !ws || ws->model != model || !doc_ids || !n_tokens || !keep || !n_keep ||
+ n_docs <= 0 || !out_vectors || !out_offsets || model->token_dim <= 0)
+ return -1;
+
+ int hidden = model->base->config.hidden_size;
+ int dim = model->token_dim;
+
+ /* Pack all candidates for one block-diagonal forward. inputs/offsets index
+ * the full token stream; out_offsets is the prefix sum of kept tokens. */
+ embed_input_t *inputs = (embed_input_t *)malloc((size_t)n_docs * sizeof(*inputs));
+ int *offsets = (int *)malloc((size_t)(n_docs + 1) * sizeof(*offsets));
+ if (!inputs || !offsets) {
+ free(inputs);
+ free(offsets);
+ return -1;
+ }
+
+ int total_keep = 0;
+ out_offsets[0] = 0;
+ for (int d = 0; d < n_docs; d++) {
+ if (!doc_ids[d] || n_tokens[d] <= 0 || !keep[d] || n_keep[d] <= 0 ||
+ n_keep[d] > n_tokens[d] || total_keep > INT_MAX - n_keep[d]) {
+ free(inputs);
+ free(offsets);
+ return -1;
+ }
+ for (int k = 0; k < n_keep[d]; k++) {
+ if (keep[d][k] < 0 || keep[d][k] >= n_tokens[d]) {
+ free(inputs);
+ free(offsets);
+ return -1;
+ }
+ }
+ inputs[d].ids = doc_ids[d];
+ inputs[d].n_tokens = n_tokens[d];
+ total_keep += n_keep[d];
+ out_offsets[d + 1] = total_keep;
+ }
+
+ int total_seq = 0, max_seq = 0;
+ int rc = build_offsets(inputs, n_docs, 1, offsets, &total_seq, &max_seq);
+ if (rc == 0)
+ rc = forward_packed_inplace(model->base, ws->base_ws, inputs, n_docs, offsets, total_seq,
+ max_seq, 1);
+ free(inputs);
+ if (rc != 0) {
+ free(offsets);
+ return -1;
+ }
+
+ /* Gather each document's kept hidden rows into ws->states (its natural
+ * [seq, hidden] shape), then project the packed [total_keep, hidden] block
+ * to [total_keep, dim] in one GEMM, matching the per-document path. */
+ if (ensure_late_state_buffer(ws, total_keep) != 0) {
+ free(offsets);
+ return -1;
+ }
+ const float *states = ws->base_ws->x;
+ int pos = 0;
+ for (int d = 0; d < n_docs; d++) {
+ int start = offsets[d];
+ for (int k = 0; k < n_keep[d]; k++) {
+ memcpy(ws->states + (size_t)pos * hidden,
+ states + (size_t)(start + keep[d][k]) * hidden, (size_t)hidden * sizeof(float));
+ pos++;
+ }
+ }
+ free(offsets);
+
+ linear_nobias_weight(ws->base_ws, out_vectors, ws->states, &model->projection, total_keep,
+ hidden, dim);
+ if (normalize) {
+ for (int i = 0; i < total_keep; i++) {
+ if (embed_l2_normalize(out_vectors + (size_t)i * dim, dim) != 0)
+ return -1;
+ }
+ }
+ return 0;
+}
+
+int embed_pool_batch(const embed_config_t *cfg,
+ const float *states,
+ const int *seq_lengths,
+ int batch,
+ float *out_embeddings) {
+ if (!cfg || !states || !seq_lengths || batch <= 0 || !out_embeddings || cfg->hidden_size <= 0)
+ return -1;
+
+ int offset = 0;
+ for (int b = 0; b < batch; b++) {
+ int len = seq_lengths[b];
+ if (len <= 0 || offset > INT_MAX - len)
+ return -1;
+
+ float *emb = out_embeddings + (size_t)b * cfg->hidden_size;
+ if (cfg->pooling_mode == EMBED_POOL_LAST_TOKEN || cfg->pooling_mode == EMBED_POOL_CLS) {
+ /* CLS takes the first token, last-token the final one; these states
+ * already carry the model's final RMSNorm. */
+ int idx = cfg->pooling_mode == EMBED_POOL_CLS ? offset : offset + len - 1;
+ const float *row = states + (size_t)idx * cfg->hidden_size;
+ memcpy(emb, row, (size_t)cfg->hidden_size * sizeof(float));
+ } else {
+ memset(emb, 0, (size_t)cfg->hidden_size * sizeof(float));
+ for (int t = 0; t < len; t++) {
+ const float *row = states + (size_t)(offset + t) * cfg->hidden_size;
+ for (int d = 0; d < cfg->hidden_size; d++)
+ emb[d] += row[d];
+ }
+ float inv_len = 1.0f / (float)len;
+ for (int d = 0; d < cfg->hidden_size; d++)
+ emb[d] *= inv_len;
+ }
+ if (cfg->normalize_embeddings && embed_l2_normalize(emb, cfg->hidden_size) != 0)
+ return -1;
+ offset += len;
+ }
+ return 0;
+}
+
+int embed_model_encode_into(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens,
+ float *out_embedding) {
+ embed_input_t input = {token_ids, n_tokens};
+ return embed_model_encode_batch(model, ws, &input, 1, out_embedding);
+}
+
+float *embed_model_encode(const embed_model_t *model,
+ embed_workspace_t *ws,
+ const int *token_ids,
+ int n_tokens) {
+ if (!model || !ws || !token_ids || n_tokens <= 0)
+ return NULL;
+
+ int hidden = model->config.hidden_size;
+ if (hidden <= 0)
+ return NULL;
+ float *emb = malloc_floats((size_t)hidden);
+ if (!emb)
+ return NULL;
+
+ if (embed_model_encode_into(model, ws, token_ids, n_tokens, emb) != 0) {
+ free(emb);
+ return NULL;
+ }
+ return emb;
+}
+
+/* ========================================================================
+ * Vector helpers
+ * ======================================================================== */
+
+int embed_l2_normalize(float *vec, int dim) {
+ if (!vec || dim <= 0)
+ return -1;
+
+ float norm_sq = 0.0f;
+ for (int i = 0; i < dim; i++)
+ norm_sq += vec[i] * vec[i];
+
+ if (!(norm_sq > 0.0f) || !isfinite(norm_sq))
+ return -1;
+
+ float inv_norm = 1.0f / sqrtf(norm_sq);
+ for (int i = 0; i < dim; i++)
+ vec[i] *= inv_norm;
+ return 0;
+}
+
+float embed_cosine_similarity(const float *a, const float *b, int dim) {
+ if (!a || !b || dim <= 0)
+ return 0.0f;
+
+ float dot = 0.0f;
+ float norm_a_sq = 0.0f;
+ float norm_b_sq = 0.0f;
+ for (int i = 0; i < dim; i++) {
+ dot += a[i] * b[i];
+ norm_a_sq += a[i] * a[i];
+ norm_b_sq += b[i] * b[i];
+ }
+
+ if (!(norm_a_sq > 0.0f) || !(norm_b_sq > 0.0f) || !isfinite(norm_a_sq) || !isfinite(norm_b_sq))
+ return 0.0f;
+ return dot / sqrtf(norm_a_sq * norm_b_sq);
+}