diff options
| -rw-r--r-- | Makefile | 38 | ||||
| -rw-r--r-- | README.md | 2 | ||||
| -rw-r--r-- | main.c | 4 | ||||
| -rw-r--r-- | pplx_embed.c | 4 | ||||
| -rw-r--r-- | pplx_embed_internal.h | 2 | ||||
| -rw-r--r-- | pplx_embed_mlx.c | 2 | ||||
| -rw-r--r-- | pplx_server.c | 4 | ||||
| -rw-r--r-- | qwen_kernels.c (renamed from qwen_asr_kernels.c) | 4 | ||||
| -rw-r--r-- | qwen_kernels.h (renamed from qwen_asr_kernels.h) | 8 | ||||
| -rw-r--r-- | qwen_kernels_avx.c (renamed from qwen_asr_kernels_avx.c) | 4 | ||||
| -rw-r--r-- | qwen_kernels_generic.c (renamed from qwen_asr_kernels_generic.c) | 4 | ||||
| -rw-r--r-- | qwen_kernels_impl.h (renamed from qwen_asr_kernels_impl.h) | 8 | ||||
| -rw-r--r-- | qwen_kernels_neon.c (renamed from qwen_asr_kernels_neon.c) | 4 | ||||
| -rw-r--r-- | qwen_safetensors.c (renamed from qwen_asr_safetensors.c) | 2 | ||||
| -rw-r--r-- | qwen_safetensors.h (renamed from qwen_asr_safetensors.h) | 8 | ||||
| -rw-r--r-- | qwen_tokenizer.c (renamed from qwen_asr_tokenizer.c) | 4 | ||||
| -rw-r--r-- | qwen_tokenizer.h (renamed from qwen_asr_tokenizer.h) | 8 | ||||
| -rw-r--r-- | server_main.c | 2 | ||||
| -rw-r--r-- | tests/test_kernels.c | 2 |
19 files changed, 57 insertions, 57 deletions
@@ -19,12 +19,12 @@ UNAME_S := $(shell uname -s) SRCS = pplx_embed.c \ pplx_distributed.c \ pplx_server.c \ - qwen_asr_kernels.c \ - qwen_asr_kernels_generic.c \ - qwen_asr_kernels_neon.c \ - qwen_asr_kernels_avx.c \ - qwen_asr_tokenizer.c \ - qwen_asr_safetensors.c \ + qwen_kernels.c \ + qwen_kernels_generic.c \ + qwen_kernels_neon.c \ + qwen_kernels_avx.c \ + qwen_tokenizer.c \ + qwen_safetensors.c \ deps/ae/ae.c \ deps/ae/anet.c \ deps/ae/monotonic.c @@ -38,8 +38,8 @@ TARGET = pplx_embed SERVER_TARGET = pplx-embed-server LIB = libpplxembed.a -KERNEL_SRCS = qwen_asr_kernels.c qwen_asr_kernels_generic.c \ - qwen_asr_kernels_neon.c qwen_asr_kernels_avx.c +KERNEL_SRCS = qwen_kernels.c qwen_kernels_generic.c \ + qwen_kernels_neon.c qwen_kernels_avx.c ifeq ($(shell uname -s),Darwin) SHARED_LIB = libpplxembed.dylib @@ -192,40 +192,40 @@ $(SERVER_TARGET): $(LIB) server_main.o # ============================================================================= # Compile rules # ============================================================================= -pplx_embed.o: pplx_embed.c pplx_embed.h qwen_asr_kernels.h qwen_asr_safetensors.h +pplx_embed.o: pplx_embed.c pplx_embed.h qwen_kernels.h qwen_safetensors.h $(CC) $(CFLAGS) $(CJSON_CFLAGS) -Ideps/ae -c -o $@ $< pplx_distributed.o: pplx_distributed.c pplx_distributed.h pplx_embed.h pplx_embed_mlx.h $(CC) $(CFLAGS) -c -o $@ $< -pplx_server.o: pplx_server.c pplx_server.h pplx_embed.h pplx_embed_mlx.h qwen_asr_tokenizer.h deps/ae/ae.h deps/ae/anet.h +pplx_server.o: pplx_server.c pplx_server.h pplx_embed.h pplx_embed_mlx.h qwen_tokenizer.h deps/ae/ae.h deps/ae/anet.h $(CC) $(CFLAGS) $(CJSON_CFLAGS) -Ideps/ae -c -o $@ $< -pplx_embed_mlx.o: pplx_embed_mlx.c pplx_embed_mlx.h pplx_embed.h qwen_asr_safetensors.h +pplx_embed_mlx.o: pplx_embed_mlx.c pplx_embed_mlx.h pplx_embed.h qwen_safetensors.h $(CC) $(CFLAGS) -c -o $@ $< -main.o: main.c pplx_embed.h pplx_distributed.h qwen_asr_kernels.h qwen_asr_tokenizer.h +main.o: main.c pplx_embed.h pplx_distributed.h qwen_kernels.h qwen_tokenizer.h $(CC) $(CFLAGS) $(CJSON_CFLAGS) -Ideps/ae -c -o $@ $< -server_main.o: server_main.c pplx_embed.h pplx_server.h qwen_asr_kernels.h +server_main.o: server_main.c pplx_embed.h pplx_server.h qwen_kernels.h $(CC) $(CFLAGS) $(CJSON_CFLAGS) -Ideps/ae -c -o $@ $< -qwen_asr_kernels.o: qwen_asr_kernels.c qwen_asr_kernels.h qwen_asr_kernels_impl.h +qwen_kernels.o: qwen_kernels.c qwen_kernels.h qwen_kernels_impl.h $(CC) $(CFLAGS) -c -o $@ $< -qwen_asr_kernels_generic.o: qwen_asr_kernels_generic.c qwen_asr_kernels_impl.h +qwen_kernels_generic.o: qwen_kernels_generic.c qwen_kernels_impl.h $(CC) $(CFLAGS) -c -o $@ $< -qwen_asr_kernels_neon.o: qwen_asr_kernels_neon.c qwen_asr_kernels_impl.h +qwen_kernels_neon.o: qwen_kernels_neon.c qwen_kernels_impl.h $(CC) $(CFLAGS) -c -o $@ $< -qwen_asr_kernels_avx.o: qwen_asr_kernels_avx.c qwen_asr_kernels_impl.h +qwen_kernels_avx.o: qwen_kernels_avx.c qwen_kernels_impl.h $(CC) $(CFLAGS) -c -o $@ $< -qwen_asr_tokenizer.o: qwen_asr_tokenizer.c qwen_asr_tokenizer.h qwen_asr_kernels.h +qwen_tokenizer.o: qwen_tokenizer.c qwen_tokenizer.h qwen_kernels.h $(CC) $(CFLAGS) -c -o $@ $< -qwen_asr_safetensors.o: qwen_asr_safetensors.c qwen_asr_safetensors.h +qwen_safetensors.o: qwen_safetensors.c qwen_safetensors.h $(CC) $(CFLAGS) -c -o $@ $< deps/ae/%.o: deps/ae/%.c deps/ae/ae.h deps/ae/anet.h deps/ae/monotonic.h @@ -81,7 +81,7 @@ Output is a line-delimited stream of JSON objects with the embedding vector, dim ```c #include "pplx_embed.h" -#include "qwen_asr_tokenizer.h" +#include "qwen_tokenizer.h" /* Load model */ pplx_ctx_t *ctx = pplx_load("/path/to/model"); @@ -2,8 +2,8 @@ #include "pplx_embed.h" #include "pplx_distributed.h" -#include "qwen_asr_kernels.h" -#include "qwen_asr_tokenizer.h" +#include "qwen_kernels.h" +#include "qwen_tokenizer.h" #ifdef USE_MLX #include "pplx_embed_mlx.h" diff --git a/pplx_embed.c b/pplx_embed.c index 4dc358e..996be9c 100644 --- a/pplx_embed.c +++ b/pplx_embed.c @@ -1,7 +1,7 @@ #include "pplx_embed.h" #include "pplx_embed_internal.h" -#include "qwen_asr_kernels.h" -#include "qwen_asr_safetensors.h" +#include "qwen_kernels.h" +#include "qwen_safetensors.h" #include <stdio.h> #include <stdlib.h> diff --git a/pplx_embed_internal.h b/pplx_embed_internal.h index fa028e9..1e7e4d0 100644 --- a/pplx_embed_internal.h +++ b/pplx_embed_internal.h @@ -2,7 +2,7 @@ #define PPLX_EMBED_INTERNAL_H #include "pplx_embed.h" -#include "qwen_asr_safetensors.h" +#include "qwen_safetensors.h" typedef struct { safetensor_dtype_t dtype; diff --git a/pplx_embed_mlx.c b/pplx_embed_mlx.c index 88129d5..602571d 100644 --- a/pplx_embed_mlx.c +++ b/pplx_embed_mlx.c @@ -1,6 +1,6 @@ #include "pplx_embed_mlx.h" #include "pplx_embed.h" -#include "qwen_asr_safetensors.h" +#include "qwen_safetensors.h" #include <mlx/c/mlx.h> diff --git a/pplx_server.c b/pplx_server.c index f8ad5e1..d831062 100644 --- a/pplx_server.c +++ b/pplx_server.c @@ -4,8 +4,8 @@ #include "pplx_server.h" #include "pplx_embed.h" -#include "qwen_asr_safetensors.h" -#include "qwen_asr_tokenizer.h" +#include "qwen_safetensors.h" +#include "qwen_tokenizer.h" #ifdef USE_MLX #include "pplx_embed_mlx.h" diff --git a/qwen_asr_kernels.c b/qwen_kernels.c index 24c8162..5cf38cd 100644 --- a/qwen_asr_kernels.c +++ b/qwen_kernels.c @@ -1,5 +1,5 @@ -#include "qwen_asr_kernels.h" -#include "qwen_asr_kernels_impl.h" +#include "qwen_kernels.h" +#include "qwen_kernels_impl.h" #include <math.h> #include <string.h> #include <stdlib.h> diff --git a/qwen_asr_kernels.h b/qwen_kernels.h index 63b704d..e92f944 100644 --- a/qwen_asr_kernels.h +++ b/qwen_kernels.h @@ -1,12 +1,12 @@ /* - * qwen_asr_kernels.h - Math kernels for Qwen3-ASR inference + * qwen_kernels.h - Math kernels for Qwen3-ASR inference * * Low-level math operations. All operate on float32 tensors in row-major order. * Adapted from voxtral-realtime project. */ -#ifndef QWEN_ASR_KERNELS_H -#define QWEN_ASR_KERNELS_H +#ifndef QWEN_KERNELS_H +#define QWEN_KERNELS_H #include <stddef.h> #include <stdint.h> @@ -207,4 +207,4 @@ int qwen_get_num_cpus(void); /* Global verbose flag */ extern int qwen_verbose; -#endif /* QWEN_ASR_KERNELS_H */ +#endif /* QWEN_KERNELS_H */ diff --git a/qwen_asr_kernels_avx.c b/qwen_kernels_avx.c index 3fa2194..f47725f 100644 --- a/qwen_asr_kernels_avx.c +++ b/qwen_kernels_avx.c @@ -1,5 +1,5 @@ /* - * qwen_asr_kernels_avx.c - x86 SIMD hot kernels (AVX2+FMA, with AVX-512 when available) + * qwen_kernels_avx.c - x86 SIMD hot kernels (AVX2+FMA, with AVX-512 when available) * * bf16 to f32 conversion: load 16 uint16, zero-extend to 32-bit, shift left 16. * Uses AVX-512F+BW for 16-wide bf16 matvec/argmax (dominant for seq_len==1), and @@ -9,7 +9,7 @@ * overhead and improve out-of-order execution on memory-bound workloads. */ -#include "qwen_asr_kernels_impl.h" +#include "qwen_kernels_impl.h" #if defined(__AVX2__) && defined(__FMA__) diff --git a/qwen_asr_kernels_generic.c b/qwen_kernels_generic.c index d322d8e..cc25d3a 100644 --- a/qwen_asr_kernels_generic.c +++ b/qwen_kernels_generic.c @@ -1,8 +1,8 @@ /* - * qwen_asr_kernels_generic.c - architecture-generic hot kernels + * qwen_kernels_generic.c - architecture-generic hot kernels */ -#include "qwen_asr_kernels_impl.h" +#include "qwen_kernels_impl.h" #include <string.h> diff --git a/qwen_asr_kernels_impl.h b/qwen_kernels_impl.h index 31b9cbb..4a351a3 100644 --- a/qwen_asr_kernels_impl.h +++ b/qwen_kernels_impl.h @@ -1,9 +1,9 @@ /* - * qwen_asr_kernels_impl.h - internal architecture dispatch for hot kernels + * qwen_kernels_impl.h - internal architecture dispatch for hot kernels */ -#ifndef QWEN_ASR_KERNELS_IMPL_H -#define QWEN_ASR_KERNELS_IMPL_H +#ifndef QWEN_KERNELS_IMPL_H +#define QWEN_KERNELS_IMPL_H #include <stdint.h> @@ -62,4 +62,4 @@ void qwen_vec_scale_add_avx(float *dst, const float *src, float correction, int #define qwen_vec_scale_add_impl qwen_vec_scale_add_generic #endif -#endif /* QWEN_ASR_KERNELS_IMPL_H */ +#endif /* QWEN_KERNELS_IMPL_H */ diff --git a/qwen_asr_kernels_neon.c b/qwen_kernels_neon.c index 90d8ed9..0060e88 100644 --- a/qwen_asr_kernels_neon.c +++ b/qwen_kernels_neon.c @@ -1,8 +1,8 @@ /* - * qwen_asr_kernels_neon.c - ARM NEON hot kernels + * qwen_kernels_neon.c - ARM NEON hot kernels */ -#include "qwen_asr_kernels_impl.h" +#include "qwen_kernels_impl.h" #ifdef __ARM_NEON diff --git a/qwen_asr_safetensors.c b/qwen_safetensors.c index 726ee9c..37a5d3e 100644 --- a/qwen_asr_safetensors.c +++ b/qwen_safetensors.c @@ -1,4 +1,4 @@ -#include "qwen_asr_safetensors.h" +#include "qwen_safetensors.h" #include <stdio.h> #include <stdlib.h> #include <string.h> diff --git a/qwen_asr_safetensors.h b/qwen_safetensors.h index 7bd1846..a93e720 100644 --- a/qwen_asr_safetensors.h +++ b/qwen_safetensors.h @@ -1,10 +1,10 @@ /* - * qwen_asr_safetensors.h - Safetensors file format reader (multi-shard support) + * qwen_safetensors.h - Safetensors file format reader (multi-shard support) * Adapted from voxtral-realtime project. */ -#ifndef QWEN_ASR_SAFETENSORS_H -#define QWEN_ASR_SAFETENSORS_H +#ifndef QWEN_SAFETENSORS_H +#define QWEN_SAFETENSORS_H #include <stddef.h> #include <stdint.h> @@ -76,4 +76,4 @@ int64_t safetensor_numel(const safetensor_t *t); void safetensor_print(const safetensor_t *t); void safetensors_print_all(const safetensors_file_t *sf); -#endif /* QWEN_ASR_SAFETENSORS_H */ +#endif /* QWEN_SAFETENSORS_H */ diff --git a/qwen_asr_tokenizer.c b/qwen_tokenizer.c index fa85017..38ecb4e 100644 --- a/qwen_asr_tokenizer.c +++ b/qwen_tokenizer.c @@ -1,5 +1,5 @@ -#include "qwen_asr_tokenizer.h" -#include "qwen_asr_kernels.h" +#include "qwen_tokenizer.h" +#include "qwen_kernels.h" #include <limits.h> #include <stdint.h> #include <stdio.h> diff --git a/qwen_asr_tokenizer.h b/qwen_tokenizer.h index 742d0d1..5d80237 100644 --- a/qwen_asr_tokenizer.h +++ b/qwen_tokenizer.h @@ -1,13 +1,13 @@ /* - * qwen_asr_tokenizer.h - Qwen2/Qwen3 BPE tokenizer (GPT-2 byte-level) + * qwen_tokenizer.h - Qwen2/Qwen3 BPE tokenizer (GPT-2 byte-level) * * Supports: * - decode token ID -> UTF-8 text * - encode UTF-8 text -> token IDs (vocab.json + merges.txt) */ -#ifndef QWEN_ASR_TOKENIZER_H -#define QWEN_ASR_TOKENIZER_H +#ifndef QWEN_TOKENIZER_H +#define QWEN_TOKENIZER_H typedef struct { char **id_to_text; /* [vocab_size] decoded text strings */ @@ -58,4 +58,4 @@ int *qwen_tokenizer_encode_with_workspace(const qwen_tokenizer_t *tok, /* Free tokenizer */ void qwen_tokenizer_free(qwen_tokenizer_t *tok); -#endif /* QWEN_ASR_TOKENIZER_H */ +#endif /* QWEN_TOKENIZER_H */ diff --git a/server_main.c b/server_main.c index b4c7a34..ace0100 100644 --- a/server_main.c +++ b/server_main.c @@ -2,7 +2,7 @@ #include "pplx_embed.h" #include "pplx_server.h" -#include "qwen_asr_kernels.h" +#include "qwen_kernels.h" #ifdef USE_CUDA #include "pplx_embed_cuda.h" diff --git a/tests/test_kernels.c b/tests/test_kernels.c index 113c731..72a205a 100644 --- a/tests/test_kernels.c +++ b/tests/test_kernels.c @@ -2,7 +2,7 @@ * Runs via `make test` (generic and BLAS variants) and * scripts/check_kernel_golden.py. No model files required. */ -#include "qwen_asr_kernels.h" +#include "qwen_kernels.h" #include <math.h> #include <stdint.h> |
