diff options
| author | Onur Gündüz <ogunduz@gmail.com> | 2026-06-20 20:10:56 +0200 |
|---|---|---|
| committer | Onur Gündüz <ogunduz@gmail.com> | 2026-06-21 05:57:19 +0200 |
| commit | 6b969d25ce136eaf695222bd1a10bff1ecac0220 (patch) | |
| tree | c23afb50c28b9deaddae5c6526a74ee11cf2d865 /tools/server/server.h | |
| parent | d4bdb8dae753462f621507d5fc3582e0e53da35c (diff) | |
| download | ffwd-6b969d25ce136eaf695222bd1a10bff1ecac0220.tar.gz ffwd-6b969d25ce136eaf695222bd1a10bff1ecac0220.zip | |
all: split Makefile and reorg library/tools
Diffstat (limited to 'tools/server/server.h')
| -rw-r--r-- | tools/server/server.h | 57 |
1 files changed, 57 insertions, 0 deletions
diff --git a/tools/server/server.h b/tools/server/server.h new file mode 100644 index 0000000..b8ec7e4 --- /dev/null +++ b/tools/server/server.h @@ -0,0 +1,57 @@ +#ifndef FFWD_SERVER_H +#define FFWD_SERVER_H + +/* Public-API export annotation. The libraries are built with + * -fvisibility=hidden, so only declarations carrying FFWD_API are exported + * from the shared library; everything else stays internal. */ +#ifndef FFWD_API +# if defined(__GNUC__) +# define FFWD_API __attribute__((visibility("default"))) +# else +# define FFWD_API +# endif +#endif + +#include <stdbool.h> + +typedef enum { + FFWD_SERVER_MODEL_STANDARD = 0, + FFWD_SERVER_MODEL_CONTEXTUAL = 1, + FFWD_SERVER_MODEL_LATE = 2, +} ffwd_server_model_kind_t; + +typedef enum { + FFWD_SERVER_API_DEFAULT = 0, /* currently OpenAI-compatible float output */ + FFWD_SERVER_API_OPENAI = 1, + FFWD_SERVER_API_PERPLEXITY = 2, +} ffwd_server_embedding_api_t; + +typedef struct { + const char *id; /* served label; requests match this string exactly */ + const char *path; + ffwd_server_model_kind_t kind; + ffwd_server_embedding_api_t api; + int min_dim; /* 0 means the loaded model's full hidden size */ + const char *query_instruct; /* NULL means use config_sentence_transformers.json if present */ +} ffwd_server_model_spec_t; + +typedef struct { + const ffwd_server_model_spec_t *models; + int n_models; + const char *host; + int port; + int batch_size; + int max_batch_tokens; + int batch_wait_us; + /* GPU build tuning; ignored by backends that do not use it. */ + int gpu_quantize_bits; + int gpu_quantize_group_size; + double memory_utilization; /* fraction of physical memory the GPU preflight + may plan for; 0 = default 0.90, above 1.0 + allows overcommit */ + const char *api_key; +} ffwd_server_config_t; + +FFWD_API int ffwd_run_server(const ffwd_server_config_t *cfg); + +#endif |
