opencl: add SOFTPLUS op support (#18726)

* preset: allow named remote preset

* nits: fix docs

* cont docs

common/arg.cpp

@@ -281,12 +281,20 @@ static std::string clean_file_name(const std::string & fname) {
 static bool common_params_handle_remote_preset(common_params & params, llama_example ex) {
     GGML_ASSERT(!params.model.hf_repo.empty());
 
+    // the returned hf_repo is without tag
+    auto [hf_repo, hf_tag] = common_download_split_repo_tag(params.model.hf_repo);
+
+    // "latest" tag (default if not specified) is translated to "default" preset
+    if (hf_tag == "latest") {
+        hf_tag = "default";
+    }
+
     const bool offline = params.offline;
     std::string model_endpoint = get_model_endpoint();
-    auto preset_url = model_endpoint + params.model.hf_repo + "/resolve/main/preset.ini";
+    auto preset_url = model_endpoint + hf_repo + "/resolve/main/preset.ini";
 
     // prepare local path for caching
-    auto preset_fname = clean_file_name(params.model.hf_repo + "_preset.ini");
+    auto preset_fname = clean_file_name(hf_repo + "_preset.ini");
     auto preset_path = fs_get_cache_file(preset_fname);
     const int status = common_download_file_single(preset_url, preset_path, params.hf_token, offline);
     const bool has_preset = status >= 200 && status < 400;
@@ -295,14 +303,15 @@ static bool common_params_handle_remote_preset(common_params & params, llama_exa
     if (has_preset) {
         LOG_INF("applying remote preset from %s\n", preset_url.c_str());
         common_preset_context ctx(ex, /* only_remote_allowed */ true);
-        common_preset global; // unused for now
+        common_preset global;
         auto remote_presets = ctx.load_from_ini(preset_path, global);
-        if (remote_presets.find(COMMON_PRESET_DEFAULT_NAME) != remote_presets.end()) {
-            common_preset & preset = remote_presets.at(COMMON_PRESET_DEFAULT_NAME);
+        remote_presets = ctx.cascade(global, remote_presets);
+        if (remote_presets.find(hf_tag) != remote_presets.end()) {
+            common_preset preset = remote_presets.at(hf_tag);
             LOG_INF("\n%s", preset.to_ini().c_str()); // to_ini already added trailing newline
             preset.apply_to_params(params);
         } else {
-            throw std::runtime_error("Remote preset.ini does not contain [" + std::string(COMMON_PRESET_DEFAULT_NAME) + "] section");
+            throw std::runtime_error("Remote preset.ini does not contain [" + std::string(hf_tag) + "] section");
         }
     } else {
         LOG_INF("%s", "no remote preset found, skipping\n");

common/download.cpp

@@ -161,6 +161,16 @@ static bool is_http_status_ok(int status) {
     return status >= 200 && status < 400;
 }
 
+std::pair<std::string, std::string> common_download_split_repo_tag(const std::string & hf_repo_with_tag) {
+    auto parts = string_split<std::string>(hf_repo_with_tag, ':');
+    std::string tag = parts.size() > 1 ? parts.back() : "latest";
+    std::string hf_repo = parts[0];
+    if (string_split<std::string>(hf_repo, '/').size() != 2) {
+        throw std::invalid_argument("error: invalid HF repo format, expected <user>/<model>[:quant]\n");
+    }
+    return {hf_repo, tag};
+}
+
 #ifdef LLAMA_USE_CURL
 
 //
@@ -922,12 +932,8 @@ common_hf_file_res common_get_hf_file(const std::string & hf_repo_with_tag,
                                       const std::string & bearer_token,
                                       bool offline,
                                       const common_header_list & custom_headers) {
-    auto parts = string_split<std::string>(hf_repo_with_tag, ':');
-    std::string tag = parts.size() > 1 ? parts.back() : "latest";
-    std::string hf_repo = parts[0];
-    if (string_split<std::string>(hf_repo, '/').size() != 2) {
-        throw std::invalid_argument("error: invalid HF repo format, expected <user>/<model>[:quant]\n");
-    }
+    // the returned hf_repo is without tag
+    auto [hf_repo, tag] = common_download_split_repo_tag(hf_repo_with_tag);
 
     std::string url = get_model_endpoint() + "v2/" + hf_repo + "/manifests/" + tag;
 

common/download.h

@@ -17,6 +17,12 @@ struct common_remote_params {
 // get remote file content, returns <http_code, raw_response_body>
 std::pair<long, std::vector<char>> common_remote_get_content(const std::string & url, const common_remote_params & params);
 
+// split HF repo with tag into <repo, tag>
+// for example: "user/model:tag" -> <"user/model", "tag">
+// if tag is not present, default to "latest"
+// example: "user/model" -> <"user/model", "latest">
+std::pair<std::string, std::string> common_download_split_repo_tag(const std::string & hf_repo_with_tag);
+
 struct common_cached_model_info {
     std::string manifest_path;
     std::string user;

common/preset.cpp

@@ -32,8 +32,10 @@ static std::set<std::string> get_remote_preset_whitelist(const std::map<std::str
         "batch-size",
         "ubatch-size",
         "cache-reuse",
+        "chat-template-kwargs",
+        "mmap",
         // note: sampling params are automatically allowed by default
-        // negated args will be added automatically
+        // negated args will be added automatically if the positive arg is specified above
     };
 
     std::set<std::string> allowed_keys;
@@ -318,6 +320,11 @@ common_presets common_preset_context::load_from_ini(const std::string & path, co
         }
         LOG_DBG("loading preset: %s\n", preset.name.c_str());
         for (const auto & [key, value] : section.second) {
+            if (key == "version") {
+                // skip version key (reserved for future use)
+                continue;
+            }
+
             LOG_DBG("option: %s = %s\n", key.c_str(), value.c_str());
             if (filter_allowed_keys && allowed_keys.find(key) == allowed_keys.end()) {
                 throw std::runtime_error(string_format(
@@ -334,7 +341,10 @@ common_presets common_preset_context::load_from_ini(const std::string & path, co
                 }
                 LOG_DBG("accepted option: %s = %s\n", key.c_str(), preset.options[opt].c_str());
             } else {
-                // TODO: maybe warn about unknown key?
+                throw std::runtime_error(string_format(
+                    "option '%s' not recognized in preset '%s'",
+                    key.c_str(), preset.name.c_str()
+                ));
             }
         }
 

docs/ops.md

@@ -57,7 +57,6 @@ Legend:
 |                         GET_ROWS | ❌ | 🟡 | ✅ | 🟡 | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ❌ | ❌ |
 |                    GET_ROWS_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                       GROUP_NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
-|               GROUP_NORM_MUL_ADD | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                      HARDSIGMOID | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
 |                        HARDSWISH | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
 |                           IM2COL | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ | ❌ |
@@ -71,10 +70,9 @@ Legend:
 |                       MUL_MAT_ID | ❌ | 🟡 | ✅ | ✅ | ✅ | 🟡 | 🟡 | ✅ | ❌ | ❌ | ❌ |
 |                              NEG | ❌ | ✅ | ✅ | 🟡 | 🟡 | ❌ | ✅ | 🟡 | ✅ | ❌ | ❌ |
 |                             NORM | ❌ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | 🟡 | ❌ | ❌ | ❌ |
-|                     NORM_MUL_ADD | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                   OPT_STEP_ADAMW | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
 |                     OPT_STEP_SGD | ❌ | ❌ | ✅ | ✅ | ✅ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
-|                         OUT_PROD | 🟡 | ❌ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | ❌ |
+|                         OUT_PROD | 🟡 | ❌ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ | 🟡 |
 |                              PAD | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | 🟡 | ✅ | ❌ | ❌ | ❌ |
 |                   PAD_REFLECT_1D | ❌ | ✅ | ✅ | ✅ | ✅ | ❌ | ✅ | ❌ | ❌ | ❌ | ❌ |
 |                          POOL_2D | ❌ | 🟡 | ✅ | ✅ | ✅ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ |
@@ -99,7 +97,6 @@ Legend:
 |                             SILU | ❌ | ✅ | ✅ | 🟡 | 🟡 | 🟡 | ✅ | 🟡 | ✅ | ❌ | ❌ |
 |                        SILU_BACK | ❌ | ❌ | ✅ | ✅ | ❌ | ❌ | ❌ | ✅ | ❌ | ❌ | ❌ |
 |                              SIN | ❌ | ✅ | ✅ | ✅ | 🟡 | ❌ | ✅ | 🟡 | ❌ | ❌ | ❌ |
-|                          SOFTCAP | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ | ❌ |
 |                         SOFTPLUS | ❌ | ❌ | ✅ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ❌ | ❌ | ❌ |
 |                         SOFT_MAX | ❌ | 🟡 | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ❌ | ❌ |
 |                    SOFT_MAX_BACK | ❌ | ❌ | 🟡 | 🟡 | ❌ | ❌ | 🟡 | ✅ | ❌ | ❌ | ❌ |

docs/ops/BLAS.csv

@@ -965,6 +965,7 @@
 "BLAS","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[12,12,1,2560],ne_kernel=[3,3,1,2560],s0=1,s1=1,p0=1,p1=1,d0=1,d1=1,is_2D=1","support","0","no","BLAS"
 "BLAS","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[12,12,2,2560],ne_kernel=[3,3,2,2560],s0=1,s1=1,p0=1,p1=1,d0=1,d1=1,is_2D=1","support","0","no","BLAS"
 "BLAS","IM2COL","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[5,5,1,32],ne_kernel=[3,4,1,32],s0=1,s1=1,p0=0,p1=0,d0=1,d1=1,is_2D=1","support","0","no","BLAS"
+"BLAS","IM2COL","type_input=f32,type_kernel=f32,dst_type=f32,ne_input=[2,2,1536,729],ne_kernel=[2,2,1536,4096],s0=1,s1=1,p0=0,p1=0,d0=1,d1=1,is_2D=1","support","0","no","BLAS"
 "BLAS","IM2COL_3D","type_input=f32,type_kernel=f32,dst_type=f32,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","BLAS"
 "BLAS","IM2COL_3D","type_input=f32,type_kernel=f16,dst_type=f32,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","BLAS"
 "BLAS","IM2COL_3D","type_input=f32,type_kernel=f16,dst_type=f16,ne_input=[10,10,10,9],ne_kernel=[3,3,3,1],IC=3,s0=1,s1=1,s2=1,p0=1,p1=1,p2=1,d0=1,d1=1,d2=1,v=0","support","0","no","BLAS"
@@ -4964,6 +4965,7 @@
 "BLAS","CONV_TRANSPOSE_1D","ne_input=[2,1,1,1],ne_kernel=[3,1,1,1],s0=1,p0=0,d0=1","support","0","no","BLAS"
 "BLAS","CONV_TRANSPOSE_2D","ne_input=[3,2,3,1],ne_kernel=[2,2,1,3],stride=1","support","0","no","BLAS"
 "BLAS","CONV_TRANSPOSE_2D","ne_input=[10,10,9,1],ne_kernel=[3,3,1,9],stride=2","support","0","no","BLAS"
+"BLAS","CONV_TRANSPOSE_2D","ne_input=[129,63,35,1],ne_kernel=[3,3,48,35],stride=1","support","0","no","BLAS"
 "BLAS","COUNT_EQUAL","type=f32,ne=[4,500,1,1]","support","0","no","BLAS"
 "BLAS","COUNT_EQUAL","type=f32,ne=[4,5000,1,1]","support","0","no","BLAS"
 "BLAS","ARGMAX","type=f32,ne=[32,1,1,1]","support","0","no","BLAS"
@@ -5715,15 +5717,15 @@
 "BLAS","L2_NORM","type=f32,ne=[64,5,4,3]","support","0","no","BLAS"
 "BLAS","RMS_NORM","type=f32,ne=[64,5,4,3],v=0,eps=0.000001,inplace=1","support","0","no","BLAS"
 "BLAS","L2_NORM","type=f32,ne=[64,5,4,3]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,1024,1,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[8,1024,1,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,1024,4,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,1536,1,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[8,1536,1,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,1536,4,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,2048,1,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[8,2048,1,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
-"BLAS","SSM_CONV","type=f32,ne_a=[4,2048,4,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,1024,1,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[6,1024,1,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,1024,4,1],ne_b=[3,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,1536,1,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[6,1536,1,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,1536,4,1],ne_b=[3,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,2048,1,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[6,2048,1,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[3,2048,4,1],ne_b=[3,2048,1,1]","support","0","no","BLAS"
 "BLAS","SSM_CONV","type=f32,ne_a=[4,1024,1,1],ne_b=[4,1024,1,1]","support","0","no","BLAS"
 "BLAS","SSM_CONV","type=f32,ne_a=[8,1024,1,1],ne_b=[4,1024,1,1]","support","0","no","BLAS"
 "BLAS","SSM_CONV","type=f32,ne_a=[4,1024,4,1],ne_b=[4,1024,1,1]","support","0","no","BLAS"
@@ -5733,6 +5735,15 @@
 "BLAS","SSM_CONV","type=f32,ne_a=[4,2048,1,1],ne_b=[4,2048,1,1]","support","0","no","BLAS"
 "BLAS","SSM_CONV","type=f32,ne_a=[8,2048,1,1],ne_b=[4,2048,1,1]","support","0","no","BLAS"
 "BLAS","SSM_CONV","type=f32,ne_a=[4,2048,4,1],ne_b=[4,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,1024,1,1],ne_b=[9,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[18,1024,1,1],ne_b=[9,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,1024,4,1],ne_b=[9,1024,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,1536,1,1],ne_b=[9,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[18,1536,1,1],ne_b=[9,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,1536,4,1],ne_b=[9,1536,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,2048,1,1],ne_b=[9,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[18,2048,1,1],ne_b=[9,2048,1,1]","support","0","no","BLAS"
+"BLAS","SSM_CONV","type=f32,ne_a=[9,2048,4,1],ne_b=[9,2048,1,1]","support","0","no","BLAS"
 "BLAS","SSM_SCAN","type=f32,d_state=16,head_dim=1,n_head=1024,n_group=1,n_seq_tokens=32,n_seqs=4","support","0","no","BLAS"
 "BLAS","SSM_SCAN","type=f32,d_state=128,head_dim=64,n_head=16,n_group=2,n_seq_tokens=32,n_seqs=4","support","0","no","BLAS"
 "BLAS","SSM_SCAN","type=f32,d_state=256,head_dim=64,n_head=8,n_group=2,n_seq_tokens=32,n_seqs=4","support","0","no","BLAS"
@@ -6592,6 +6603,30 @@
 "BLAS","MUL_MAT","type_a=f16,type_b=f32,m=1056,n=1,k=67,bs=[1,1],nr=[4,1],per=[0,2,1,3],k_v=0,o=1","support","0","no","BLAS"
 "BLAS","MUL_MAT","type_a=f32,type_b=f32,m=64,n=77,k=77,bs=[12,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","1","yes","BLAS"
 "BLAS","MUL_MAT","type_a=q4_0,type_b=f32,m=576,n=512,k=576,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","1","yes","BLAS"
+"BLAS","MUL_MAT","type_a=q4_0,type_b=f32,m=1,n=2048,k=8192,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=f32,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=f16,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=bf16,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q4_0,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q4_1,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q5_0,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q5_1,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q8_0,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=mxfp4,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q2_K,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q3_K,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q4_K,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q5_K,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=q6_K,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq2_xxs,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq2_xs,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq2_s,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq3_xxs,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq1_s,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq1_m,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq4_nl,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq3_s,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
+"BLAS","MUL_MAT","type_a=iq4_xs,type_b=f32,m=1,n=64,k=256,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=0,o=1","support","0","no","BLAS"
 "BLAS","MUL_MAT","type_a=f16,type_b=f32,m=1056,n=1,k=128,bs=[1,1],nr=[1,1],per=[0,2,1,3],k_v=0,o=1","support","0","no","BLAS"
 "BLAS","MUL_MAT","type_a=f16,type_b=f32,m=128,n=1,k=1056,bs=[1,1],nr=[1,1],per=[0,1,2,3],k_v=2112,o=1","support","0","no","BLAS"
 "BLAS","MUL_MAT","type_a=bf16,type_b=f32,m=1056,n=1,k=128,bs=[1,1],nr=[1,1],per=[0,2,1,3],k_v=0,o=1","support","0","no","BLAS"
@@ -8916,6 +8951,11 @@
 "BLAS","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=0,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=0.000000,inplace=0","support","0","no","BLAS"
 "BLAS","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=1,m_prec=f32,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","0","no","BLAS"
 "BLAS","SOFT_MAX","type=f32,ne=[32,2,32,1],mask=1,sinks=1,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","0","no","BLAS"
+"BLAS","SOFT_MAX","type=f32,ne=[200001,2,3,1],mask=1,sinks=1,m_prec=f32,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","0","no","BLAS"
+"BLAS","SOFT_MAX","type=f32,ne=[200001,2,3,1],mask=1,sinks=1,m_prec=f16,nr23=[1,1],scale=0.100000,max_bias=8.000000,inplace=0","support","0","no","BLAS"
+"BLAS","SOFT_MAX","type=f32,ne=[200000,1,1,1],mask=0,sinks=0,m_prec=f32,nr23=[1,1],scale=1.000000,max_bias=0.000000,inplace=0","support","0","no","BLAS"
+"BLAS","SOFT_MAX","type=f32,ne=[200000,4,1,1],mask=0,sinks=0,m_prec=f32,nr23=[1,1],scale=1.000000,max_bias=0.000000,inplace=0","support","0","no","BLAS"
+"BLAS","SOFT_MAX","type=f32,ne=[643251,3,1,1],mask=0,sinks=0,m_prec=f32,nr23=[1,1],scale=1.000000,max_bias=0.000000,inplace=0","support","0","no","BLAS"
 "BLAS","SOFT_MAX_BACK","type=f32,ne=[16,16,1,1],scale=1.000000,max_bias=0.000000","support","0","no","BLAS"
 "BLAS","SOFT_MAX_BACK","type=f32,ne=[15,15,1,1],scale=1.000000,max_bias=0.000000","support","0","no","BLAS"
 "BLAS","SOFT_MAX_BACK","type=f32,ne=[16,16,2,3],scale=1.000000,max_bias=0.000000","support","0","no","BLAS"
@@ -8968,6 +9008,7 @@
 "BLAS","ROPE","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -8977,6 +9018,7 @@
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -8987,11 +9029,13 @@
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
@@ -9001,6 +9045,7 @@
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
@@ -9011,11 +9056,13 @@
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
@@ -9025,6 +9072,7 @@
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
@@ -9035,11 +9083,13 @@
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
@@ -9049,6 +9099,7 @@
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
@@ -9059,6 +9110,7 @@
 "BLAS","ROPE","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f16,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE","type=f16,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -9184,6 +9236,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -9193,6 +9246,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -9203,11 +9257,13 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
@@ -9217,6 +9273,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
@@ -9227,11 +9284,13 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
@@ -9241,6 +9300,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
@@ -9251,11 +9311,13 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,40,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,52,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,64,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,1,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,71,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,8,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
@@ -9265,6 +9327,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=20,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,2,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,32,4,1],n_dims=32,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=128,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,12,2,1],n_dims=20,mode=8,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
@@ -9275,6 +9338,7 @@
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,28,2,1],n_dims=32,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[80,16,2,1],n_dims=80,mode=24,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[128,16,2,1],n_dims=128,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
+"BLAS","ROPE_BACK","type=f32,ne_a=[16,16,8192,1],n_dims=16,mode=40,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f32,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=1,v=1,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f16,ne_a=[128,32,2,1],n_dims=128,mode=0,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
 "BLAS","ROPE_BACK","type=f16,ne_a=[64,128,2,1],n_dims=64,mode=2,n_ctx=512,fs=1.000000,ef=0.000000,af=1.000000,ff=0,v=0,inplace=0","support","0","no","BLAS"
@@ -9542,333 +9606,333 @@
 "BLAS","ARGSORT","type=f32,ne=[2048,2,1,3],order=1","support","0","no","BLAS"
 "BLAS","ARGSORT","type=f32,ne=[2049,2,1,3],order=1","support","0","no","BLAS"
 "BLAS","ARGSORT","type=f32,ne=[2,8,8192,1],order=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[12,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[13,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[13,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=100","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=500","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=1023","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=9999","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=1","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=2","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=3","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=7","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=15","support","0","no","BLAS"
-"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=15","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[12,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[13,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[13,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[15,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[19,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[27,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[43,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[64,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[75,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[128,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[139,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[256,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[267,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[512,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[523,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1035,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2059,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4096,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[4107,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8192,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[8203,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16395,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32768,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[32779,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65536,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[65547,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131072,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[131083,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262144,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[262155,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=100,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=500,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=1023,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524288,1,1,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[524299,1,2,1],k=9999,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=1,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=2,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=3,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=7,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16,10,10,10],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[60,10,10,10],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1023,2,1,3],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1024,2,1,3],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[1025,2,1,3],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[16384,1,1,1],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2047,2,1,3],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2048,2,1,3],k=15,ties=0","support","0","no","BLAS"
+"BLAS","TOP_K","type=f32,ne=[2049,2,1,3],k=15,ties=0","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=nearest,transpose=0","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=nearest,transpose=1","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=nearest,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=nearest,flags=none","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=nearest","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=nearest","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bilinear,transpose=0","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bilinear,transpose=1","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bilinear,flags=none","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bilinear","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bicubic,transpose=0","support","0","no","BLAS"
 "BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bicubic,transpose=1","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bicubic,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bicubic,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=513,transpose=0","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=513,transpose=1","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bilinear,flags=none","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear,flags=align_corners","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[1,4,3,2],ne_tgt=[2,8,3,2],mode=bilinear,flags=align_corners","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[4,1,3,2],ne_tgt=[1,1,3,2],mode=bilinear,flags=align_corners","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bicubic,flags=align_corners","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[1,4,3,2],ne_tgt=[2,8,3,2],mode=bicubic,flags=align_corners","support","0","no","BLAS"
-"BLAS","UPSCALE","type=f32,ne=[4,1,3,2],ne_tgt=[1,1,3,2],mode=bicubic,flags=align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bicubic","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bicubic","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bilinear|antialias,transpose=0","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[512,512,3,2],scale_factor=2,mode=bilinear|antialias,transpose=1","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear|antialias","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[5,7,11,13],ne_tgt=[2,5,7,11],mode=bilinear|antialias","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bilinear|align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[1,4,3,2],ne_tgt=[2,8,3,2],mode=bilinear|align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[4,1,3,2],ne_tgt=[1,1,3,2],mode=bilinear|align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[2,5,7,11],ne_tgt=[5,7,11,13],mode=bicubic|align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[1,4,3,2],ne_tgt=[2,8,3,2],mode=bicubic|align_corners","support","0","no","BLAS"
+"BLAS","UPSCALE","type=f32,ne=[4,1,3,2],ne_tgt=[1,1,3,2],mode=bicubic|align_corners","support","0","no","BLAS"
 "BLAS","SUM","type=f32,ne=[10,5,4,3]","support","0","no","BLAS"
 "BLAS","SUM_ROWS","type=f32,ne=[10,5,4,3],permute=0,slice=0","support","0","no","BLAS"
 "BLAS","SUM","type=f32,ne=[11,5,6,3],permute=[0,2,1,3]","support","0","no","BLAS"
@@ -9891,8 +9955,9 @@
 "BLAS","GROUP_NORM","type=f32,ne=[64,64,320,1],num_groups=32,eps=0.000001","support","0","no","BLAS"
 "BLAS","GROUP_NORM","type=f32,ne=[9,9,1280,1],num_groups=32,eps=0.000001","support","0","no","BLAS"
 "BLAS","ACC","type=f32,ne_a=[256,17,1,1],ne_b=[256,16,1,1]","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[512,512,1,1],pad_0=1,pad_1=1","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[512,512,3,1],lp0=1,rp0=1,lp1=1,rp1=1,lp2=1,rp2=1,lp3=1,rp3=1,v=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,1,1],pad_0=1,pad_1=1,circular=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[33,17,2,1],pad_0=4,pad_1=3,circular=1","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,3,1],lp0=1,rp0=1,lp1=1,rp1=1,lp2=1,rp2=1,lp3=1,rp3=1,v=0,circular=0","support","0","no","BLAS"
 "BLAS","PAD_REFLECT_1D","type=f32,ne_a=[512,34,2,1],pad_0=10,pad_1=9","support","0","no","BLAS"
 "BLAS","PAD_REFLECT_1D","type=f32,ne_a=[3000,384,4,1],pad_0=10,pad_1=9","support","0","no","BLAS"
 "BLAS","ROLL","shift0=3,shift1=-2,shift3=1,shift4=-1","support","0","no","BLAS"
@@ -9914,6 +9979,7 @@
 "BLAS","CUMSUM","type=f32,ne=[2048,5,4,3]","support","0","no","BLAS"
 "BLAS","CUMSUM","type=f32,ne=[242004,1,1,1]","support","0","no","BLAS"
 "BLAS","CUMSUM","type=f32,ne=[375960,1,1,1]","support","0","no","BLAS"
+"BLAS","CUMSUM","type=f32,ne=[20481,4,1,1]","support","0","no","BLAS"
 "BLAS","XIELU","type=f32,ne=[10,5,4,3]","support","0","no","BLAS"
 "BLAS","TRI","type=f32,ne=[10,10,4,3],tri_type=3","support","0","no","BLAS"
 "BLAS","TRI","type=f32,ne=[10,10,4,3],tri_type=2","support","0","no","BLAS"
@@ -9923,17 +9989,41 @@
 "BLAS","FILL","type=f32,ne=[303,207,11,3],c=2.000000","support","0","no","BLAS"
 "BLAS","FILL","type=f32,ne=[800,600,4,4],c=-152.000000","support","0","no","BLAS"
 "BLAS","FILL","type=f32,ne=[2048,512,2,2],c=3.500000","support","0","no","BLAS"
+"BLAS","DIAG","type=f32,ne=[10,1,4,3]","support","0","no","BLAS"
+"BLAS","DIAG","type=f32,ne=[79,1,19,13]","support","0","no","BLAS"
+"BLAS","DIAG","type=f32,ne=[256,1,8,16]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[10,10,4,3],ne_rhs=[3,10,4,3]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[11,11,1,1],ne_rhs=[5,11,1,1]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[17,17,2,4],ne_rhs=[9,17,2,4]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[30,30,7,1],ne_rhs=[8,30,7,1]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[42,42,5,2],ne_rhs=[10,42,5,2]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[64,64,2,2],ne_rhs=[10,64,2,2]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[64,64,2,2],ne_rhs=[64,64,2,2]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[79,79,5,3],ne_rhs=[417,79,5,3]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,2],ne_rhs=[32,128,4,2]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[80,80,2,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[79,80,2,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,2,8],ne_rhs=[81,80,2,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[80,80,8,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[79,80,8,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[80,80,8,8],ne_rhs=[81,80,8,8]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[84,84,4,4],ne_rhs=[32,84,4,4]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[95,95,8,8],ne_rhs=[40,95,8,8]","support","0","no","BLAS"
 "BLAS","SOLVE_TRI","type=f32,ne_lhs=[100,100,4,4],ne_rhs=[41,100,4,4]","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1","support","0","no","BLAS"
-"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,4],ne_rhs=[31,128,4,4]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,4],ne_rhs=[32,128,4,4]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[128,128,3,4],ne_rhs=[32,128,3,4]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[128,128,4,1],ne_rhs=[32,128,4,1]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[64,64,4,4],ne_rhs=[200,64,4,4]","support","0","no","BLAS"
+"BLAS","SOLVE_TRI","type=f32,ne_lhs=[64,64,4,4],ne_rhs=[384,64,4,4]","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0,circular=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0,circular=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=0,circular=1","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=0,circular=1","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1,circular=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1,circular=0","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[512,512,1,1],lp0=0,rp0=1,lp1=0,rp1=1,lp2=0,rp2=0,lp3=0,rp3=0,v=1,circular=1","support","0","no","BLAS"
+"BLAS","PAD","type=f32,ne_a=[11,22,33,44],lp0=1,rp0=2,lp1=3,rp1=4,lp2=5,rp2=6,lp3=7,rp3=8,v=1,circular=1","support","0","no","BLAS"
 "BLAS","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=f32,permute=[0,1,2,3]","support","0","no","BLAS"
 "BLAS","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=f16,permute=[0,1,2,3]","support","0","no","BLAS"
 "BLAS","FLASH_ATTN_EXT","hsk=40,hsv=40,nh=4,nr23=[1,1],kv=113,nb=1,mask=1,sinks=1,max_bias=0.000000,logit_softcap=0.000000,prec=f32,type_KV=bf16,permute=[0,1,2,3]","support","0","no","BLAS"

docs/preset.md

@@ -58,3 +58,40 @@ temp = 0.8
 ctx-size = 1024
 ; (and other configurations)
 ```
+
+### Named presets
+
+If you want to define multiple preset configurations for one or more GGUF models, you can create a blank HF repo containing a single `preset.ini` file that references the actual model(s):
+
+```ini
+[*]
+mmap = 1
+
+[gpt-oss-20b-hf]
+hf          = ggml-org/gpt-oss-20b-GGUF
+batch-size  = 2048
+ubatch-size = 2048
+top-p       = 1.0
+top-k       = 0
+min-p       = 0.01
+temp        = 1.0
+chat-template-kwargs = {"reasoning_effort": "high"}
+
+[gpt-oss-120b-hf]
+hf          = ggml-org/gpt-oss-120b-GGUF
+batch-size  = 2048
+ubatch-size = 2048
+top-p       = 1.0
+top-k       = 0
+min-p       = 0.01
+temp        = 1.0
+chat-template-kwargs = {"reasoning_effort": "high"}
+```
+
+You can then use it via `llama-cli` or `llama-server`, example:
+
+```sh
+llama-server -hf user/repo:gpt-oss-120b-hf
+```
+
+Please make sure to provide the correct `hf-repo` for each child preset. Otherwise, you may get error: `The specified tag is not a valid quantization scheme.`

ggml/src/ggml-blas/CMakeLists.txt

@@ -32,14 +32,12 @@ if (BLAS_FOUND)
                 pkg_check_modules(DepBLAS openblas)
             endif()
         elseif (${GGML_BLAS_VENDOR} MATCHES "FLAME")
-            add_compile_definitions(GGML_BLAS_USE_BLIS)
             pkg_check_modules(DepBLAS blis)
         elseif (${GGML_BLAS_VENDOR} MATCHES "ATLAS")
             pkg_check_modules(DepBLAS blas-atlas)
         elseif (${GGML_BLAS_VENDOR} MATCHES "FlexiBLAS")
             pkg_check_modules(DepBLAS flexiblas_api)
         elseif (${GGML_BLAS_VENDOR} MATCHES "Intel")
-            add_compile_definitions(GGML_BLAS_USE_MKL)
             # all Intel* libraries share the same include path
             pkg_check_modules(DepBLAS mkl-sdl)
         elseif (${GGML_BLAS_VENDOR} MATCHES "NVHPC")
@@ -74,10 +72,26 @@ if (BLAS_FOUND)
 
     target_compile_options(ggml-blas PRIVATE ${BLAS_LINKER_FLAGS})
 
-    if ("${BLAS_INCLUDE_DIRS}" MATCHES "mkl" AND (${GGML_BLAS_VENDOR} MATCHES "Generic" OR ${GGML_BLAS_VENDOR} MATCHES "Intel"))
+    if ("${GGML_BLAS_VENDOR}" STREQUAL "")
+        message(WARNING "GGML_BLAS_VENDOR is not set; some methods may not link properly.")
+    endif()
+
+    if ("${GGML_BLAS_VENDOR}" MATCHES "Intel" OR ("${BLAS_INCLUDE_DIRS}" MATCHES "mkl" AND "${GGML_BLAS_VENDOR}" MATCHES "Generic"))
         add_compile_definitions(GGML_BLAS_USE_MKL)
     endif()
 
+    if ("${GGML_BLAS_VENDOR}" MATCHES "OpenBLAS")
+        add_compile_definitions(GGML_BLAS_USE_OPENBLAS)
+    endif()
+
+    if ("${GGML_BLAS_VENDOR}" MATCHES "FLAME" OR "${GGML_BLAS_VENDOR}" MATCHES "AOCL" OR "${GGML_BLAS_VENDOR}" MATCHES "AOCL_mt")
+        add_compile_definitions(GGML_BLAS_USE_BLIS)
+    endif()
+
+    if ("${GGML_BLAS_VENDOR}" MATCHES "NVPL")
+        add_compile_definitions(GGML_BLAS_USE_NVPL)
+    endif()
+
     target_link_libraries     (ggml-blas PRIVATE ${BLAS_LIBRARIES})
     target_include_directories(ggml-blas PRIVATE ${BLAS_INCLUDE_DIRS})
 else()

ggml/src/ggml-blas/ggml-blas.cpp

@@ -115,15 +115,11 @@ static void ggml_backend_blas_mul_mat(ggml_backend_blas_context * ctx, struct gg
 #endif
     }
 
-#if defined(OPENBLAS_VERSION)
+#if defined(GGML_BLAS_USE_OPENBLAS)
     openblas_set_num_threads(ctx->n_threads);
-#endif
-
-#if defined(GGML_BLAS_USE_BLIS)
+#elif defined(GGML_BLAS_USE_BLIS)
     bli_thread_set_num_threads(ctx->n_threads);
-#endif
-
-#if defined(GGML_BLAS_USE_NVPL)
+#elif defined(GGML_BLAS_USE_NVPL)
     nvpl_blas_set_num_threads(ctx->n_threads);
 #endif
 
@@ -288,7 +284,7 @@ ggml_backend_t ggml_backend_blas_init(void) {
         /* .context = */ ctx,
     };
 
-#if defined(OPENBLAS_VERSION) && defined(GGML_USE_OPENMP)
+#if defined(GGML_BLAS_USE_OPENBLAS) && defined(GGML_USE_OPENMP)
     if (openblas_get_parallel() != OPENBLAS_OPENMP) {
         GGML_LOG_DEBUG("%s: warning: ggml is using OpenMP, but OpenBLAS was compiled without OpenMP support\n", __func__);
     }
@@ -329,7 +325,7 @@ static const char * ggml_backend_blas_device_get_description(ggml_backend_dev_t
         return "BLIS";
     #elif defined(GGML_BLAS_USE_NVPL)
         return "NVPL";
-    #elif defined(OPENBLAS_VERSION)
+    #elif defined(GGML_BLAS_USE_OPENBLAS)
         return "OpenBLAS";
     #else
         return "BLAS";

ggml/src/ggml-cuda/mmq.cu

@@ -190,7 +190,7 @@ void ggml_cuda_mul_mat_q(
     {
         const int64_t s11 = src1->nb[1] / ts_src1;
         const int64_t s12 = src1->nb[2] / ts_src1;
-        const int64_t s13 = src1->nb[2] / ts_src1;
+        const int64_t s13 = src1->nb[3] / ts_src1;
 
         if (use_native_mxfp4) {
             quantize_mmq_mxfp4_cuda(src1_d, ids_src1.get(), src1_q8_1.get(), src0->type, ne10, s11, s12, s13,
@@ -333,28 +333,31 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11, int64_t
     }
 
     if (amd_wmma_available(cc)) {
-        // RDNA 4 is consistently worse on rocblas
-        // https://github.com/ggml-org/llama.cpp/pull/18537#issuecomment-3706422301
         if (GGML_CUDA_CC_IS_RDNA3(cc)) {
-            // High expert counts almost always better on MMQ
-            // due to a large amount of graph splits
+            // High expert counts are almost always better on MMQ due to
+            //     the synchronization overhead in the cuBLAS/hipBLAS path:
             // https://github.com/ggml-org/llama.cpp/pull/18202
             if (n_experts >= 64) {
                 return true;
             }
 
+            // For some quantization types MMQ can have lower peak TOPS than hipBLAS
+            //     so it's only faster for sufficiently small batch sizes:
             switch (type) {
-                // These quants are really bad on MMQ
                 case GGML_TYPE_Q2_K:
+                    return ne11 <= 128;
                 case GGML_TYPE_Q6_K:
-                // These quants are usually worse but not always
+                    return ne11 <= (GGML_CUDA_CC_IS_RDNA3_0(cc) ? 128 : 256);
                 case GGML_TYPE_IQ2_XS:
                 case GGML_TYPE_IQ2_S:
-                    return ne11 <= 128;
+                    return GGML_CUDA_CC_IS_RDNA3_5(cc) || ne11 <= 128;
                 default:
                     return true;
             }
         }
+
+        // For RDNA4 MMQ is consistently faster than dequantization + hipBLAS:
+        // https://github.com/ggml-org/llama.cpp/pull/18537#issuecomment-3706422301
         return true;
     }
 

ggml/src/ggml-opencl/CMakeLists.txt

@@ -122,6 +122,7 @@ set(GGML_OPENCL_KERNELS
     upscale
     tanh
     expm1
+    softplus
     pad
     repeat
     mul_mat_f16_f32

ggml/src/ggml-opencl/ggml-opencl.cpp

@@ -540,6 +540,8 @@ struct ggml_backend_opencl_context {
     cl_kernel kernel_tanh_f16_nd;
     cl_kernel kernel_expm1_f32_nd;
     cl_kernel kernel_expm1_f16_nd;
+    cl_kernel kernel_softplus_f32_nd;
+    cl_kernel kernel_softplus_f16_nd;
     cl_kernel kernel_upscale;
     cl_kernel kernel_upscale_bilinear;
     cl_kernel kernel_concat_f32_contiguous;
@@ -1826,6 +1828,31 @@ static void load_cl_kernels(ggml_backend_opencl_context *backend_ctx, ggml_cl_ve
         CL_CHECK(clReleaseProgram(prog));
     }
 
+    // softplus
+    {
+#ifdef GGML_OPENCL_EMBED_KERNELS
+        const std::string kernel_src {
+            #include "softplus.cl.h"
+        };
+#else
+        const std::string kernel_src = read_file("softplus.cl");
+#endif
+        cl_program prog;
+        if (!kernel_src.empty()) {
+            prog =
+                build_program_from_source(backend_ctx->context, backend_ctx->device, kernel_src.c_str(), compile_opts);
+            CL_CHECK((backend_ctx->kernel_softplus_f32_nd = clCreateKernel(prog, "kernel_softplus_f32_nd", &err), err));
+            CL_CHECK((backend_ctx->kernel_softplus_f16_nd = clCreateKernel(prog, "kernel_softplus_f16_nd", &err), err));
+            GGML_LOG_CONT(".");
+        } else {
+            GGML_LOG_WARN("ggml_opencl: softplus kernel source not found or empty. Softplus operation will not be available.\n");
+            prog = nullptr;
+            backend_ctx->kernel_softplus_f32_nd = nullptr;
+            backend_ctx->kernel_softplus_f16_nd = nullptr;
+        }
+        CL_CHECK(clReleaseProgram(prog));
+    }
+
     // upscale
     {
 #ifdef GGML_OPENCL_EMBED_KERNELS
@@ -3138,6 +3165,9 @@ static bool ggml_opencl_supports_op(ggml_backend_dev_t dev, const struct ggml_te
                 case GGML_UNARY_OP_EXPM1:
                    return (op->src[0]->type == GGML_TYPE_F32 && op->type == GGML_TYPE_F32) ||
                           (op->src[0]->type == GGML_TYPE_F16 && op->type == GGML_TYPE_F16);
+                case GGML_UNARY_OP_SOFTPLUS:
+                   return (op->src[0]->type == GGML_TYPE_F32 && op->type == GGML_TYPE_F32) ||
+                          (op->src[0]->type == GGML_TYPE_F16 && op->type == GGML_TYPE_F16);
                 default:
                     return false;
             }
@@ -6596,6 +6626,108 @@ static void ggml_cl_expm1(ggml_backend_t backend, const ggml_tensor * src0, cons
     backend_ctx->enqueue_ndrange_kernel(kernel, 3, global_work_size, local_work_size_ptr, dst);
 }
 
+static void ggml_cl_softplus(ggml_backend_t backend, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst) {
+    GGML_ASSERT(src0);
+    GGML_ASSERT(src0->extra);
+    GGML_ASSERT(dst);
+    GGML_ASSERT(dst->extra);
+
+    UNUSED(src1);
+
+    ggml_backend_opencl_context *backend_ctx = (ggml_backend_opencl_context *)backend->context;
+
+    ggml_tensor_extra_cl * extra0 = (ggml_tensor_extra_cl *)src0->extra;
+    ggml_tensor_extra_cl * extrad = (ggml_tensor_extra_cl *)dst->extra;
+
+    cl_ulong offset0_abs = extra0->offset + src0->view_offs;
+    cl_ulong offsetd_abs = extrad->offset + dst->view_offs;
+
+    cl_kernel kernel;
+    if (dst->type == GGML_TYPE_F32) {
+        kernel = backend_ctx->kernel_softplus_f32_nd;
+    } else if (dst->type == GGML_TYPE_F16) {
+        kernel = backend_ctx->kernel_softplus_f16_nd;
+    } else {
+        GGML_ASSERT(false && "Unsupported type for ggml_cl_softplus");
+    }
+    GGML_ASSERT(kernel != nullptr);
+
+    const int ne00 = src0->ne[0];
+    const int ne01 = src0->ne[1];
+    const int ne02 = src0->ne[2];
+    const int ne03 = src0->ne[3];
+
+    const cl_ulong nb00 = src0->nb[0];
+    const cl_ulong nb01 = src0->nb[1];
+    const cl_ulong nb02 = src0->nb[2];
+    const cl_ulong nb03 = src0->nb[3];
+
+    const int ne10 = dst->ne[0];
+    const int ne11 = dst->ne[1];
+    const int ne12 = dst->ne[2];
+    const int ne13 = dst->ne[3];
+
+    const cl_ulong nb10 = dst->nb[0];
+    const cl_ulong nb11 = dst->nb[1];
+    const cl_ulong nb12 = dst->nb[2];
+    const cl_ulong nb13 = dst->nb[3];
+
+    CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem),   &extra0->data_device));
+    CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_ulong), &offset0_abs));
+    CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem),   &extrad->data_device));
+    CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_ulong), &offsetd_abs));
+
+    CL_CHECK(clSetKernelArg(kernel, 4, sizeof(int),      &ne00));
+    CL_CHECK(clSetKernelArg(kernel, 5, sizeof(int),      &ne01));
+    CL_CHECK(clSetKernelArg(kernel, 6, sizeof(int),      &ne02));
+    CL_CHECK(clSetKernelArg(kernel, 7, sizeof(int),      &ne03));
+    CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_ulong), &nb00));
+    CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_ulong), &nb01));
+    CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_ulong),&nb02));
+    CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_ulong),&nb03));
+
+    CL_CHECK(clSetKernelArg(kernel, 12, sizeof(int),     &ne10));
+    CL_CHECK(clSetKernelArg(kernel, 13, sizeof(int),     &ne11));
+    CL_CHECK(clSetKernelArg(kernel, 14, sizeof(int),     &ne12));
+    CL_CHECK(clSetKernelArg(kernel, 15, sizeof(int),     &ne13));
+    CL_CHECK(clSetKernelArg(kernel, 16, sizeof(cl_ulong),&nb10));
+    CL_CHECK(clSetKernelArg(kernel, 17, sizeof(cl_ulong),&nb11));
+    CL_CHECK(clSetKernelArg(kernel, 18, sizeof(cl_ulong),&nb12));
+    CL_CHECK(clSetKernelArg(kernel, 19, sizeof(cl_ulong),&nb13));
+
+    size_t global_work_size[3];
+    if (ne10 == 0 || ne11 == 0 || ne12 == 0 || ne13 == 0) { // Handle case of 0 elements
+        return;
+    }
+    global_work_size[0] = (size_t)ne10;
+    global_work_size[1] = (size_t)ne11;
+    global_work_size[2] = (size_t)ne12;
+
+    size_t lws0 = 16, lws1 = 4, lws2 = 1;
+    if (ne10 < 16) lws0 = ne10;
+    if (ne11 < 4) lws1 = ne11;
+    if (ne12 < 1) lws2 = ne12 > 0 ? ne12 : 1;
+
+    while (lws0 * lws1 * lws2 > 256 && lws0 > 1) lws0 /= 2;
+    while (lws0 * lws1 * lws2 > 256 && lws1 > 1) lws1 /= 2;
+    while (lws0 * lws1 * lws2 > 256 && lws2 > 1) lws2 /= 2;
+
+
+    size_t local_work_size[] = {lws0, lws1, lws2};
+
+    size_t* local_work_size_ptr = local_work_size;
+    if (!backend_ctx->non_uniform_workgroups) {
+        if (global_work_size[0] % local_work_size[0] != 0 ||
+            global_work_size[1] % local_work_size[1] != 0 ||
+            global_work_size[2] % local_work_size[2] != 0) {
+            local_work_size_ptr = NULL;
+        }
+    }
+    if (global_work_size[0] == 0 || global_work_size[1] == 0 || global_work_size[2] == 0) return;
+
+    backend_ctx->enqueue_ndrange_kernel(kernel, 3, global_work_size, local_work_size_ptr, dst);
+}
+
 static void ggml_cl_repeat(ggml_backend_t backend, const ggml_tensor * src0, const ggml_tensor * src1_shape_def, ggml_tensor * dst) {
     GGML_ASSERT(src0);
     GGML_ASSERT(src0->extra);
@@ -9775,6 +9907,12 @@ bool ggml_cl_compute_forward(ggml_backend_t backend, struct ggml_tensor * tensor
                     }
                     func = ggml_cl_expm1;
                     break;
+                case GGML_UNARY_OP_SOFTPLUS:
+                    if (!any_on_device) {
+                        return false;
+                    }
+                    func = ggml_cl_softplus;
+                    break;
                 default:
                     return false;
             } break;

ggml/src/ggml-opencl/kernels/softplus.cl

@@ -0,0 +1,88 @@
+#pragma OPENCL EXTENSION cl_khr_fp16 : enable
+
+//------------------------------------------------------------------------------
+// softplus
+//------------------------------------------------------------------------------
+inline float softplus_f32(float x){
+    float ax = fabs(x);
+    float m = fmax(x, 0.0f);
+    return log1p(exp(-ax)) + m;
+}
+
+kernel void kernel_softplus_f32_nd(
+        global void * p_src0_base,
+        ulong off_src0_abs,
+        global void * p_dst_base,
+        ulong off_dst_abs,
+        int ne00,
+        int ne01,
+        int ne02,
+        int ne03,
+        ulong nb00,
+        ulong nb01,
+        ulong nb02,
+        ulong nb03,
+        int ne10,
+        int ne11,
+        int ne12,
+        int ne13,
+        ulong nb10,
+        ulong nb11,
+        ulong nb12,
+        ulong nb13
+) {
+    int i0 = get_global_id(0);
+    int i1 = get_global_id(1);
+    int i2 = get_global_id(2);
+
+    if (i0 < ne10 && i1 < ne11 && i2 < ne12) {
+        for (int i3 = 0; i3 < ne13; ++i3) {
+            ulong src_offset_in_tensor = (ulong)i0*nb00 + (ulong)i1*nb01 + (ulong)i2*nb02 + (ulong)i3*nb03;
+            global const float *src_val_ptr = (global const float *)((global char *)p_src0_base + off_src0_abs + src_offset_in_tensor);
+
+            ulong dst_offset_in_tensor = (ulong)i0*nb10 + (ulong)i1*nb11 + (ulong)i2*nb12 + (ulong)i3*nb13;
+            global float *dst_val_ptr = (global float *)((global char *)p_dst_base + off_dst_abs + dst_offset_in_tensor);
+
+            *dst_val_ptr = softplus_f32(*src_val_ptr);
+        }
+    }
+}
+
+kernel void kernel_softplus_f16_nd(
+        global void * p_src0_base,
+        ulong off_src0_abs,
+        global void * p_dst_base,
+        ulong off_dst_abs,
+        int ne00,
+        int ne01,
+        int ne02,
+        int ne03,
+        ulong nb00,
+        ulong nb01,
+        ulong nb02,
+        ulong nb03,
+        int ne10,
+        int ne11,
+        int ne12,
+        int ne13,
+        ulong nb10,
+        ulong nb11,
+        ulong nb12,
+        ulong nb13
+) {
+    int i0 = get_global_id(0);
+    int i1 = get_global_id(1);
+    int i2 = get_global_id(2);
+
+    if (i0 < ne10 && i1 < ne11 && i2 < ne12) {
+        for (int i3 = 0; i3 < ne13; ++i3) {
+            ulong src_offset_in_tensor = (ulong)i0*nb00 + (ulong)i1*nb01 + (ulong)i2*nb02 + (ulong)i3*nb03;
+            global const half *src_val_ptr = (global const half *)((global char *)p_src0_base + off_src0_abs + src_offset_in_tensor);
+
+            ulong dst_offset_in_tensor = (ulong)i0*nb10 + (ulong)i1*nb11 + (ulong)i2*nb12 + (ulong)i3*nb13;
+            global half *dst_val_ptr = (global half *)((global char *)p_dst_base + off_dst_abs + dst_offset_in_tensor);
+
+            *dst_val_ptr = (half)(softplus_f32((float)(*src_val_ptr)));
+        }
+    }
+}

scripts/pr2wt.sh

@@ -40,7 +40,7 @@ org_repo=${org_repo%.git}
 
 echo "org/repo: $org_repo"
 
-meta=$(curl -sSf -H "Accept: application/vnd.github+json" "https://api.github.com/repos/$org_repo/pulls/$PR")
+meta=$(curl -sSLf -H "Accept: application/vnd.github+json" "https://api.github.com/repos/$org_repo/pulls/$PR")
 
 url_remote=$(echo "$meta" | jq -r '.head.repo.clone_url')
 head_ref=$(echo "$meta" | jq -r '.head.ref')

tests/test-backend-ops.cpp

@@ -454,6 +454,28 @@ static bool ggml_is_view_op(enum ggml_op op) {
     return op == GGML_OP_VIEW || op == GGML_OP_RESHAPE || op == GGML_OP_PERMUTE || op == GGML_OP_TRANSPOSE;
 }
 
+static bool backend_has_feature(ggml_backend_t backend, const char * feature_name) {
+    ggml_backend_dev_t dev = ggml_backend_get_device(backend);
+    ggml_backend_reg_t reg = ggml_backend_dev_backend_reg(dev);
+
+    auto get_features = (ggml_backend_get_features_t) ggml_backend_reg_get_proc_address(reg, "ggml_backend_get_features");
+    if (!get_features) {
+        return false;
+    }
+
+    const ggml_backend_feature * features = get_features(reg);
+    if (!features) {
+        return false;
+    }
+
+    for (const ggml_backend_feature * f = features; f->name; ++f) {
+        if (strcmp(f->name, feature_name) == 0 && strcmp(f->value, "1") == 0) {
+            return true;
+        }
+    }
+    return false;
+}
+
 enum test_mode {
     MODE_TEST,
     MODE_PERF,
@@ -1101,6 +1123,11 @@ struct test_case {
         return 1e-7;
     }
 
+    virtual double max_nmse_err(ggml_backend_t backend) {
+        GGML_UNUSED(backend);
+        return max_nmse_err();
+    }
+
     virtual double max_maa_err() {
         return 1e-4;
     }
@@ -1109,6 +1136,10 @@ struct test_case {
         return max_nmse_err();
     }
 
+    virtual double max_err(ggml_backend_t backend) {
+        return max_nmse_err(backend);
+    }
+
     virtual double err(const float * a, const float * b, size_t n) {
         return nmse(a, b, n);
     }
@@ -1378,8 +1409,8 @@ struct test_case {
             }
 
             double err = ud->tc->err(f1.data(), f2.data(), f1.size());
-            if (err > ud->tc->max_err()) {
-                printf("[%s] ERR = %.9f > %.9f ", ggml_op_desc(t1), err, ud->tc->max_err());
+            if (err > ud->tc->max_err(ud->backend1)) {
+                printf("[%s] ERR = %.9f > %.9f ", ggml_op_desc(t1), err, ud->tc->max_err(ud->backend1));
                 //for (int i = 0; i < (int) f1.size(); i++) {
                 //    printf("%5d %9.6f %9.6f, diff = %9.6f\n", i, f1[i], f2[i], f1[i] - f2[i]);
                 //}
@@ -3686,6 +3717,14 @@ struct test_mul_mat : public test_case {
         return 5e-4;
     }
 
+    double max_nmse_err(ggml_backend_t backend) override {
+        // for blackwell we quantize activations to mxfp4 instead of q8_1 so we add higher tolerance
+        if (type_a == GGML_TYPE_MXFP4 && backend_has_feature(backend, "BLACKWELL_NATIVE_FP4")) {
+            return 2e-2;
+        }
+        return max_nmse_err();
+    }
+
     int64_t grad_nmax() override {
         return 20000;
     }
@@ -3814,6 +3853,14 @@ struct test_mul_mat_id : public test_case {
         return 5e-4;
     }
 
+    double max_nmse_err(ggml_backend_t backend) override {
+        // for blackwell we quantize activations to mxfp4 instead of q8_1 so we add higher tolerance
+        if (type_a == GGML_TYPE_MXFP4 && backend_has_feature(backend, "BLACKWELL_NATIVE_FP4")) {
+            return 2e-2;
+        }
+        return max_nmse_err();
+    }
+
     uint64_t op_flops(ggml_tensor * t) override {
         GGML_UNUSED(t);
         return 2 * m * k * n * n_used;

tools/server/tests/unit/test_completion.py

@@ -393,12 +393,12 @@ def test_completion_unified(n_ctx, n_slots, n_predict_vals, expected_success):
     for res, n_predict, expect_ok in zip(results, n_predict_vals, expected_success):
         if expect_ok:
             assert res.status_code == 200
+
+        # note: https://github.com/ggml-org/llama.cpp/pull/18700#issuecomment-3728695581
+        if res.status_code == 200:
             assert "content" in res.body
             if "timings" in res.body:
                 assert res.body["timings"]["predicted_n"] == n_predict
-        else:
-            assert res.status_code == 500
-            assert "content" not in res.body
 
 
 @pytest.mark.parametrize(