Add cross-attention accessor functions for AlignAtt streaming

include/whisper.h

@@ -339,6 +339,27 @@ extern "C" {
                                int   n_past,
                                int   n_threads);
 
+    // Same as whisper_decode_with_state, but saves alignment head cross-attention data.
+    // Requires context created with dtw_token_timestamps=true and flash_attn=false.
+    WHISPER_API int whisper_decode_with_state_and_aheads(
+            struct whisper_context * ctx,
+              struct whisper_state * state,
+               const whisper_token * tokens,
+                               int   n_tokens,
+                               int   n_past,
+                               int   n_threads);
+
+    // Get cross-attention data from alignment heads after a decode call with aheads enabled.
+    // Returns pointer to float array of shape [n_tokens x n_audio_ctx x n_heads].
+    // Copies data from GPU/backend to CPU on each call.
+    // Returns NULL if DTW is not enabled or no attention data is available.
+    // The pointer is valid until the next call to this function or whisper_free_state.
+    WHISPER_API const float * whisper_state_get_aheads_cross_qks(
+              struct whisper_state * state,
+                               int * n_tokens,
+                               int * n_audio_ctx,
+                               int * n_heads);
+
     // Convert the provided text into tokens.
     // The tokens pointer must be large enough to hold the resulting tokens.
     // Returns the number of tokens on success, no more than n_max_tokens

src/whisper.cpp

@@ -3954,6 +3954,38 @@ int whisper_decode(struct whisper_context * ctx, const whisper_token * tokens, i
     return whisper_decode_with_state(ctx, ctx->state, tokens, n_tokens, n_past, n_threads);
 }
 
+int whisper_decode_with_state_and_aheads(struct whisper_context * ctx, struct whisper_state * state, const whisper_token * tokens, int n_tokens, int n_past, int n_threads) {
+    whisper_batch_prep_legacy(state->batch, tokens, n_tokens, n_past, 0);
+    whisper_kv_cache_seq_rm(state->kv_self, 0, n_past, -1);
+
+    if (!whisper_decode_internal(*ctx, *state, state->batch, n_threads, true, nullptr, nullptr)) {
+        WHISPER_LOG_ERROR("%s: failed to eval\n", __func__);
+        return 1;
+    }
+
+    return 0;
+}
+
+const float * whisper_state_get_aheads_cross_qks(struct whisper_state * state, int * out_n_tokens, int * out_n_audio_ctx, int * out_n_heads) {
+    if (state->aheads_cross_QKs == nullptr) {
+        return nullptr;
+    }
+
+    const int n_tokens    = state->aheads_cross_QKs->ne[0];
+    const int n_audio_ctx = state->aheads_cross_QKs->ne[1];
+    const int n_heads     = state->aheads_cross_QKs->ne[2];
+
+    auto & data = state->aheads_cross_QKs_data;
+    data.resize(n_tokens * n_audio_ctx * n_heads);
+    ggml_backend_tensor_get(state->aheads_cross_QKs, data.data(), 0, sizeof(float) * n_tokens * n_audio_ctx * n_heads);
+
+    if (out_n_tokens)    *out_n_tokens    = n_tokens;
+    if (out_n_audio_ctx) *out_n_audio_ctx = n_audio_ctx;
+    if (out_n_heads)     *out_n_heads     = n_heads;
+
+    return data.data();
+}
+
 int whisper_tokenize(struct whisper_context * ctx, const char * text, whisper_token * tokens, int n_max_tokens) {
     const auto res = tokenize(ctx->vocab, text);