vad : print out the rest of the model weights

Author: danbev

src/whisper.cpp

@@ -5055,6 +5055,135 @@ whisper_vad_context * whisper_vad_init_from_file_with_params_no_state(
                 WHISPER_LOG_INFO("%s: encoder_0_bias: [%d]: %f\n", __func__, i, read_b[i]);
             }
         }
+
+        {
+            // Print as F16
+            struct ggml_tensor * tensor = model.encoder_1_weight;
+            std::vector<uint16_t> raw_data(ggml_nbytes(tensor) / sizeof(uint16_t));
+            ggml_backend_tensor_get(tensor, raw_data.data(), 0, ggml_nbytes(tensor));
+
+            // Convert first 10 values from F16 to F32 for display
+            for (int i = 0; i < 10; i++) {
+                float converted_value = ggml_fp16_to_fp32(raw_data[i]);
+                WHISPER_LOG_INFO("%s: model.encoder.1.reparam_conv: [%d]: %f (raw: 0x%04x)\n",
+                            __func__, i, converted_value, raw_data[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.encoder_1_bias;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: encoder_1_bias: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+
+        {
+            // Print as F16
+            struct ggml_tensor * tensor = model.encoder_2_weight;
+            std::vector<uint16_t> raw_data(ggml_nbytes(tensor) / sizeof(uint16_t));
+            ggml_backend_tensor_get(tensor, raw_data.data(), 0, ggml_nbytes(tensor));
+
+            // Convert first 10 values from F16 to F32 for display
+            for (int i = 0; i < 10; i++) {
+                float converted_value = ggml_fp16_to_fp32(raw_data[i]);
+                WHISPER_LOG_INFO("%s: model.encoder.2.reparam_conv: [%d]: %f (raw: 0x%04x)\n",
+                            __func__, i, converted_value, raw_data[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.encoder_2_bias;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: encoder_2_bias: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+
+        {
+            // Print as F16
+            struct ggml_tensor * tensor = model.encoder_3_weight;
+            std::vector<uint16_t> raw_data(ggml_nbytes(tensor) / sizeof(uint16_t));
+            ggml_backend_tensor_get(tensor, raw_data.data(), 0, ggml_nbytes(tensor));
+
+            // Convert first 10 values from F16 to F32 for display
+            for (int i = 0; i < 10; i++) {
+                float converted_value = ggml_fp16_to_fp32(raw_data[i]);
+                WHISPER_LOG_INFO("%s: model.encoder.2.reparam_conv: [%d]: %f (raw: 0x%04x)\n",
+                            __func__, i, converted_value, raw_data[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.encoder_3_bias;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: encoder_3_bias: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.lstm_weight_ih;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: lstm_weight_ih: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.lstm_bias_ih;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: lstm_bias_ih: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.lstm_weight_hh;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: lstm_weight_hh: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.lstm_bias_hh;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: lstm_bias_hh: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.final_conv_weight;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: final_conv_weight: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
+        {
+            // Print as F32
+            struct ggml_tensor * tensor = model.final_conv_bias;
+            std::vector<float> read_b(ggml_nbytes(tensor));
+            ggml_backend_tensor_get(tensor, read_b.data(), 0, ggml_nbytes(tensor));
+            for (int i = 0; i < 10; i++) {
+                WHISPER_LOG_INFO("%s: final_conv_bias: [%d]: %f\n", __func__, i, read_b[i]);
+            }
+        }
     }
 
     return vctx;