extern/whisper.cpp
1
0
Fork 1
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2025-05-04 16:14:34 +02:00
Code Issues Packages Projects Releases Wiki Activity

whisper : remove extra backend instance (huh?)

Browse Source
This commit is contained in:
Georgi Gerganov 2024-05-14 17:26:22 +03:00
parent 2877b026cf
commit 4caa64b73e
No known key found for this signature in database
GPG Key ID: 449E073F9DC10735
1 changed files with 4 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
whisper.cpp
View File

@ -819,8 +819,6 @@ struct whisper_state {
whisper_decoder decoders[WHISPER_MAX_DECODERS]; whisper_decoder decoders[WHISPER_MAX_DECODERS];
ggml_backend_t backend = nullptr;
// ggml-alloc: // ggml-alloc:
// - stores meta info about the intermediate tensors into the `meta` buffers // - stores meta info about the intermediate tensors into the `meta` buffers
// - stores the actual tensor data into the `data` buffers // - stores the actual tensor data into the `data` buffers
@ -2240,7 +2238,7 @@ static bool whisper_encode_internal(
} }
if (!whisper_encode_external(wstate)) { if (!whisper_encode_external(wstate)) {
if (!ggml_graph_compute_helper(wstate.backend, gf, n_threads)) { if (!ggml_graph_compute_helper(wctx.backend, gf, n_threads)) {
return false; return false;
} }
} else { } else {
@ -2263,7 +2261,7 @@ static bool whisper_encode_internal(
return false; return false;
} }
if (!ggml_graph_compute_helper(wstate.backend, gf, n_threads)) { if (!ggml_graph_compute_helper(wctx.backend, gf, n_threads)) {
return false; return false;
} }
} }
@ -2279,7 +2277,7 @@ static bool whisper_encode_internal(
return false; return false;
} }
if (!ggml_graph_compute_helper(wstate.backend, gf, n_threads)) { if (!ggml_graph_compute_helper(wctx.backend, gf, n_threads)) {
return false; return false;
} }
} }
@ -2744,7 +2742,7 @@ static bool whisper_decode_internal(
logits = gf->nodes[gf->n_nodes - 1]; logits = gf->nodes[gf->n_nodes - 1];
if (!ggml_graph_compute_helper(wstate.backend, gf, n_threads)) { if (!ggml_graph_compute_helper(wctx.backend, gf, n_threads)) {
return false; return false;
} }
} }
@ -3191,13 +3189,6 @@ struct whisper_state * whisper_init_state(whisper_context * ctx) {
whisper_state * state = new whisper_state; whisper_state * state = new whisper_state;
state->backend = whisper_backend_init(ctx->params);
if (!state->backend) {
WHISPER_LOG_ERROR("%s: whisper_backend_init() failed\n", __func__);
whisper_free_state(state);
return nullptr;
}
// at this point, we don't know yet how many decoders will be used, so we overallocate 3x ctx // at this point, we don't know yet how many decoders will be used, so we overallocate 3x ctx
// in theory, there can be a case where this is not enough, but in practice it should always be enough // in theory, there can be a case where this is not enough, but in practice it should always be enough
const int factor = 3; const int factor = 3;
@ -3623,8 +3614,6 @@ void whisper_free_state(struct whisper_state * state) {
ggml_gallocr_free(state->alloc_cross.alloc); ggml_gallocr_free(state->alloc_cross.alloc);
ggml_gallocr_free(state->alloc_decode.alloc); ggml_gallocr_free(state->alloc_decode.alloc);
ggml_backend_free(state->backend);
// [EXPERIMENTAL] Token-level timestamps with DTW // [EXPERIMENTAL] Token-level timestamps with DTW
aheads_masks_free(state->aheads_masks); aheads_masks_free(state->aheads_masks);