/* * embed_mlx.h - MLX GPU backend for pplx-embed inference * * Uses Apple's mlx-c to run the transformer on Metal. */ #ifndef PPLX_EMBED_MLX_H #define PPLX_EMBED_MLX_H /* Public-API export annotation. The libraries are built with * -fvisibility=hidden, so only declarations carrying PPLX_API are exported * from the shared library; everything else stays internal. */ #ifndef PPLX_API #if defined(__GNUC__) #define PPLX_API __attribute__((visibility("default"))) #else #define PPLX_API #endif #endif #include #include #include "embed.h" /* Opaque MLX model context */ typedef struct pplx_mlx_ctx pplx_mlx_ctx_t; typedef struct pplx_mlx_late_ctx pplx_mlx_late_ctx_t; typedef struct pplx_mlx_late_vectors pplx_mlx_late_vectors_t; typedef struct { int quantize_bits; /* 0 disables; currently 8 */ int quantize_group_size; /* default 64 when quantize_bits is set */ } pplx_mlx_options_t; /* * Load model into MLX arrays from safetensors + config.json. * Returns NULL on error. */ PPLX_API pplx_mlx_ctx_t *pplx_mlx_load(const char *model_dir); PPLX_API pplx_mlx_ctx_t *pplx_mlx_load_with_options(const char *model_dir, const pplx_mlx_options_t *opts); /* * Load one contiguous transformer layer range for distributed execution. * Uses the same half-open range and endpoint ownership rules as * pplx_model_load_slice(). */ PPLX_API pplx_mlx_ctx_t *pplx_mlx_load_slice(const char *model_dir, int layer_start, int layer_end); PPLX_API pplx_mlx_ctx_t *pplx_mlx_load_slice_with_options( const char *model_dir, int layer_start, int layer_end, const pplx_mlx_options_t *opts); PPLX_API void pplx_mlx_free(pplx_mlx_ctx_t *ctx); /* * Late-interaction MLX path. This is separate from pooled embeddings because * late snapshots produce one projected token vector per input token for MaxSim * scoring, not one pooled document vector. */ PPLX_API pplx_mlx_late_ctx_t *pplx_mlx_late_load(const char *model_dir); PPLX_API pplx_mlx_late_ctx_t *pplx_mlx_late_load_with_options( const char *model_dir, const pplx_mlx_options_t *opts); PPLX_API void pplx_mlx_late_free(pplx_mlx_late_ctx_t *ctx); PPLX_API const pplx_config_t *pplx_mlx_late_config(const pplx_mlx_late_ctx_t *ctx); PPLX_API int pplx_mlx_late_token_dim(const pplx_mlx_late_ctx_t *ctx); /* * Encode token_ids into out_vectors[n_tokens, token_dim]. * * If normalize is non-zero, every token vector is L2-normalized on device * before copying back to CPU memory. */ PPLX_API int pplx_mlx_late_encode_tokens(pplx_mlx_late_ctx_t *ctx, const int *token_ids, int n_tokens, int normalize, float *out_vectors); /* * Device-resident late-interaction token vectors. * * The model context owns the MLX stream; the returned vector handle must not * outlive ctx. Copying is optional: MaxSim can consume these handles directly * and copy back only final candidate scores. */ PPLX_API pplx_mlx_late_vectors_t *pplx_mlx_late_encode_tokens_device( pplx_mlx_late_ctx_t *ctx, const int *token_ids, int n_tokens, int normalize); PPLX_API void pplx_mlx_late_vectors_free(pplx_mlx_late_vectors_t *vecs); PPLX_API int pplx_mlx_late_vectors_token_count(const pplx_mlx_late_vectors_t *vecs); PPLX_API int pplx_mlx_late_vectors_dim(const pplx_mlx_late_vectors_t *vecs); PPLX_API int pplx_mlx_late_vectors_copy(const pplx_mlx_late_vectors_t *vecs, float *out_vectors); PPLX_API pplx_mlx_late_vectors_t *pplx_mlx_late_vectors_concat( pplx_mlx_late_ctx_t *ctx, const pplx_mlx_late_vectors_t *const *items, int count); PPLX_API pplx_mlx_late_vectors_t *pplx_mlx_late_vectors_select( pplx_mlx_late_ctx_t *ctx, const pplx_mlx_late_vectors_t *vecs, const int *token_indices, int count); PPLX_API int pplx_mlx_late_maxsim_batch_device( pplx_mlx_late_ctx_t *ctx, const pplx_mlx_late_vectors_t *query, const pplx_mlx_late_vectors_t *docs, const int *doc_offsets, int docs_count, float *scores); /* * Compute embedding for token_ids[0..n_tokens-1]. * Returns malloc'd float[hidden_size] (caller frees). NULL on error. */ PPLX_API float *pplx_mlx_embed(pplx_mlx_ctx_t *ctx, const int *token_ids, int n_tokens); /* * Compute one embedding into caller-provided out[hidden_size]. * Returns 0 on success, -1 on error. */ PPLX_API int pplx_mlx_embed_into(pplx_mlx_ctx_t *ctx, const int *token_ids, int n_tokens, float *out_embedding); /* * Compute a true padded dense batch on MLX. * * out_embeddings is caller-provided [batch, hidden_size]. * Padded tokens are masked out of attention keys and mean pooling. * Returns 0 on success, -1 on error. */ PPLX_API int pplx_mlx_embed_batch(pplx_mlx_ctx_t *ctx, const pplx_input_t *inputs, int batch, float *out_embeddings); /* * MLX equivalent of pplx_model_forward_slice_batch(). * * Hidden states cross the API boundary as packed float32 rows even though MLX * executes a padded dense batch internally. */ PPLX_API int pplx_mlx_forward_slice_batch(pplx_mlx_ctx_t *ctx, const pplx_input_t *inputs, int batch, const float *input_states, int layer_start, int layer_end, int apply_final_norm, float *out_states); /* * Run one contextual sequence and pool selected token spans. * Returns 0 on success, -1 on error. */ PPLX_API int pplx_mlx_embed_spans(pplx_mlx_ctx_t *ctx, const int *token_ids, int n_tokens, const pplx_span_t *spans, int n_spans, float *out_embeddings); /* * Run a padded dense contextual document batch and pool every selected span. * out_embeddings contains spans in document order. */ PPLX_API int pplx_mlx_embed_spans_batch(pplx_mlx_ctx_t *ctx, const pplx_context_input_t *inputs, int batch, float *out_embeddings); /* Get the config (so embed_cli.c can read hidden_size etc.) */ PPLX_API const pplx_config_t *pplx_mlx_config(const pplx_mlx_ctx_t *ctx); #endif /* PPLX_EMBED_MLX_H */