Refactor Vulkan backend to allow multiple contexts (#7961)

* Refactor Vulkan backend to allow multiple contexts

* Fix too many shader groups called validation error in llama3 on AMD and Intel GPUs

* Fix Vulkan debug build error

.editorconfig

@@ -28,4 +28,5 @@ indent_size = 2
 indent_style = tab
 
 [examples/cvector-generator/*.txt]
+trim_trailing_whitespace = unset
 insert_final_newline = unset

common/common.cpp

@@ -1989,8 +1989,8 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
     options.push_back({ "cvector",     "       --completions-file FNAME",
                                                                         "completions file (default: '%s')", params.cvector_completions_file.c_str() });
     options.push_back({ "cvector",     "       --completions N",        "number of lines of completions file to use (default: %d)", params.n_completions });
-    options.push_back({ "cvector",     "       --batch-pca N",          "batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch });
-    options.push_back({ "cvector",     "       --iter-pca N",           "number of iterations used for PCA (default: %d)", params.n_pca_iterations });
+    options.push_back({ "cvector",     "       --pca-batch N",          "batch size used for PCA. Larger batch runs faster, but uses more memory (default: %d)", params.n_pca_batch });
+    options.push_back({ "cvector",     "       --pca-iter N",           "number of iterations used for PCA (default: %d)", params.n_pca_iterations });
 
     printf("usage: %s [options]\n", argv[0]);
 

convert-hf-to-gguf.py

@@ -967,7 +967,11 @@ class XverseModel(Model):
         from transformers import AutoTokenizer
         tokenizer = AutoTokenizer.from_pretrained(dir_model)
         vocab_size = hparams.get("vocab_size", len(tokenizer.vocab))
-        assert max(tokenizer.vocab.values()) < vocab_size
+        # Since we are checking the maximum index, we need to ensure it's strictly less than vocab_size,
+        # because vocab_size is the count of items, and indexes start at 0.
+        max_vocab_index = max(tokenizer.get_vocab().values())
+        if max_vocab_index >= vocab_size:
+            raise ValueError("Vocabulary size exceeds expected maximum size.")
 
         reverse_vocab: dict[int, str] = {id_: encoded_tok for encoded_tok, id_ in tokenizer.vocab.items()}
         added_vocab = tokenizer.get_added_vocab()

examples/cvector-generator/README.md

@@ -17,7 +17,7 @@ Related PRs:
 ./cvector-generator -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99
 
 # With advanced options
-./cvector-generator -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99 --completions 128 --pca-iter 2000 --batch-pca 100
+./cvector-generator -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99 --completions 128 --pca-iter 2000 --pca-batch 100
 
 # To see help message
 ./cvector-generator -h

examples/cvector-generator/cvector-generator.cpp

@@ -40,7 +40,7 @@ static void print_usage(int argc, char ** argv, const gpt_params & params) {
     printf("\nexample usage:\n");
     printf("\n    CPU only:   %s -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf\n", argv[0]);
     printf("\n    with GPU:   %s -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99\n", argv[0]);
-    printf("\n    advanced:   %s -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99 --completions 128 --pca-iter 2000 --batch-pca 100\n", argv[0]);
+    printf("\n    advanced:   %s -m ./dolphin-2.0-mistral-7b.Q4_K_M.gguf -ngl 99 --completions 128 --pca-iter 2000 --pca-batch 100\n", argv[0]);
     printf("\n");
 }
 
@@ -377,8 +377,8 @@ static int prepare_entries(gpt_params & params, train_context & ctx_train) {
     // create templated prompts
     std::vector<std::string> completions = ctrlvec_load_prompt_file(params.cvector_completions_file, false);
     auto format_template = [](std::string persona, std::string suffix) {
-        // entry in positive/negative.txt must already be formatted i.e. "[INST] Act as if you're extremely happy. [/INST]"
-        return persona + " " + suffix;
+        // entry in positive/negative.txt must already be formatted i.e. "[INST] Act as if you're extremely happy. [/INST] "
+        return persona + suffix;
     };
     for (size_t i = 0; i < positive_prompts.size(); ++i) {
         for (int j = 0; j < std::min((int) completions.size(), params.n_completions); ++j) {

examples/cvector-generator/negative.txt

@@ -1 +1 @@
-[INST] Act like a person who is extremely sad. [/INST]
+[INST] Act like a person who is extremely sad. [/INST] 

examples/cvector-generator/positive.txt

@@ -1 +1 @@
-[INST] Act like a person who is extremely happy. [/INST]
+[INST] Act like a person who is extremely happy. [/INST] 

vulkan-shaders/mul_mat_vec.comp

@@ -13,7 +13,7 @@ layout (constant_id = 0) const uint BLOCK_SIZE = 32;
 shared FLOAT_TYPE tmp[BLOCK_SIZE];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
     const uint tid = gl_LocalInvocationID.x;
 
     uint a_offset, b_offset, d_offset;

vulkan-shaders/mul_mat_vec_q2_k.comp

@@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
 shared FLOAT_TYPE tmp[32];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
 
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);

vulkan-shaders/mul_mat_vec_q3_k.comp

@@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
 shared FLOAT_TYPE tmp[32];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
 
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);

vulkan-shaders/mul_mat_vec_q4_k.comp

@@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
 shared FLOAT_TYPE tmp[32];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
 
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);

vulkan-shaders/mul_mat_vec_q5_k.comp

@@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
 shared FLOAT_TYPE tmp[32];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
 
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);

vulkan-shaders/mul_mat_vec_q6_k.comp

@@ -7,7 +7,7 @@ layout(local_size_x = 32, local_size_y = 1, local_size_z = 1) in;
 shared FLOAT_TYPE tmp[32];
 
 void main() {
-    const uint row = gl_WorkGroupID.x;
+    const uint row = gl_WorkGroupID.x + gl_NumWorkGroups.x * gl_WorkGroupID.z;
 
     uint a_offset, b_offset, d_offset;
     get_offsets(a_offset, b_offset, d_offset);