diff --git a/examples/server/public/index.html b/examples/server/public/index.html
index 4b8135e02..60a5ac7ee 100644
--- a/examples/server/public/index.html
+++ b/examples/server/public/index.html
@@ -204,23 +204,38 @@
             <div class="label">System Message</div>
             <textarea class="textarea textarea-bordered h-24" :placeholder="'Default: ' + configDefault.systemMessage" v-model="config.systemMessage"></textarea>
           </label>
-          <template v-for="key in ['temperature', 'top_k', 'top_p', 'min_p', 'max_tokens']">
-            <label class="input input-bordered flex items-center gap-2 mb-2">
-              <b>{{ key }}</b>
-              <input type="text" class="grow" :placeholder="'Default: ' + (configDefault[key] || 'none')" v-model="config[key]" />
-            </label>
+          <template v-for="configKey in ['temperature', 'top_k', 'top_p', 'min_p', 'max_tokens']">
+            <settings-modal-numeric-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]" />
           </template>
           <!-- TODO: add more sampling-related configs, please regroup them into different "collapse" sections -->
-          <div class="collapse collapse-arrow bg-base-200 mb-2">
-            <input type="checkbox" />
-            <div class="collapse-title font-bold">Advanced config</div>
+          <!-- Section: Other sampler settings -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Other sampler settings</summary>
+            <div class="collapse-content">
+              <template v-for="configKey in ['dynatemp_range', 'dynatemp_exponent', 'typical_p', 'xtc_probability', 'xtc_threshold']">
+                <settings-modal-numeric-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]" />
+              </template>
+            </div>
+          </details>
+          <!-- Section: Penalties settings -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Penalties settings</summary>
+            <div class="collapse-content">
+              <template v-for="configKey in ['repeat_last_n', 'repeat_penalty', 'presence_penalty', 'frequency_penalty', 'dry_multiplier', 'dry_base', 'dry_allowed_length', 'dry_penalty_last_n']">
+                <settings-modal-numeric-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]" />
+              </template>
+            </div>
+          </details>
+          <!-- Section: Advanced config -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Advanced config</summary>
             <div class="collapse-content">
               <label class="form-control mb-2">
                 <div class="label inline">Custom JSON config (For more info, refer to <a class="underline" href="https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md" target="_blank" rel="noopener noreferrer">server documentation</a>)</div>
                 <textarea class="textarea textarea-bordered h-24" placeholder="Example: { &quot;mirostat&quot;: 1, &quot;min_p&quot;: 0.1 }" v-model="config.custom"></textarea>
               </label>
             </div>
-          </div>
+          </details>
         </div>
 
         <!-- action buttons -->
@@ -233,6 +248,21 @@
     </dialog>
   </div>
 
+  <!-- Template to be used by settings modal -->
+  <template id="settings-modal-numeric-input">
+    <label class="input input-bordered join-item grow flex items-center gap-2 mb-2">
+      <!-- Show help message on hovering on the input label -->
+      <div class="dropdown dropdown-hover">
+        <div tabindex="0" role="button" class="font-bold">{{ configKey }}</div>
+        <div class="dropdown-content menu bg-base-100 rounded-box z-10 w-64 p-2 shadow mt-4">
+          {{ configInfo[configKey] || '(no help message available)' }}
+        </div>
+      </div>
+      <!-- Here we forward v-model from parent to child component, see: https://stackoverflow.com/questions/47311936/v-model-and-child-components -->
+      <input type="text" class="grow" :placeholder="'Default: ' + (configDefault[configKey] || 'none')" :value="modelValue" @input="$emit('update:modelValue', $event.target.value)" />
+    </label>
+  </template>
+
   <script src="./deps_markdown-it.js"></script>
   <script type="module">
     import { createApp, defineComponent, shallowRef, computed, h } from './deps_vue.esm-browser.js';
@@ -253,12 +283,48 @@
       systemMessage: 'You are a helpful assistant.',
       // make sure these default values are in sync with `common.h`
       temperature: 0.8,
+      dynatemp_range: 0.0,
+      dynatemp_exponent: 1.0,
       top_k: 40,
       top_p: 0.95,
       min_p: 0.05,
+      xtc_probability: 0.0,
+      xtc_threshold: 0.1,
+      typical_p: 1.0,
+      repeat_last_n: 64,
+      repeat_penalty: 1.0,
+      presence_penalty: 0.0,
+      frequency_penalty: 0.0,
+      dry_multiplier: 0.0,
+      dry_base: 1.75,
+      dry_allowed_length: 2,
+      dry_penalty_last_n: -1,
       max_tokens: -1,
       custom: '', // custom json-stringified object
     };
+    const CONFIG_INFO = {
+      apiKey: '',
+      systemMessage: 'The starting message that defines how model should behave.',
+      temperature: 'Controls the randomness of the generated text by affecting the probability distribution of the output tokens. Higher = more random, lower = more focused.',
+      dynatemp_range: 'Addon for the temperature sampler. The added value to the range of dynamic temperature, which adjusts probabilities by entropy of tokens.',
+      dynatemp_exponent: 'Addon for the temperature sampler. Smoothes out the probability redistribution based on the most probable token.',
+      top_k: 'Keeps only k top tokens.',
+      top_p: 'Limits tokens to those that together have a cumulative probability of at least p',
+      min_p: 'Limits tokens based on the minimum probability for a token to be considered, relative to the probability of the most likely token.',
+      xtc_probability: 'XTC sampler cuts out top tokens; this parameter controls the chance of cutting tokens at all. 0 disables XTC.',
+      xtc_threshold: 'XTC sampler cuts out top tokens; this parameter controls the token probability that is required to cut that token.',
+      typical_p: 'Sorts and limits tokens based on the difference between log-probability and entropy.',
+      repeat_last_n: 'Last n tokens to consider for penalizing repetition',
+      repeat_penalty: 'Controls the repetition of token sequences in the generated text',
+      presence_penalty: 'Limits tokens based on whether they appear in the output or not.',
+      frequency_penalty: 'Limits tokens based on how often they appear in the output.',
+      dry_multiplier: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the DRY sampling multiplier.',
+      dry_base: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the DRY sampling base value.',
+      dry_allowed_length: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the allowed length for DRY sampling.',
+      dry_penalty_last_n: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets DRY penalty for the last n tokens.',
+      max_tokens: 'The maximum number of token per output.',
+      custom: '', // custom json-stringified object
+    };
     // config keys having numeric value (i.e. temperature, top_k, top_p, etc)
     const CONFIG_NUMERIC_KEYS = Object.entries(CONFIG_DEFAULT).filter(e => isNumeric(e[1])).map(e => e[0]);
     // list of themes supported by daisyui
@@ -286,6 +352,12 @@
       { props: ["source"] }
     );
 
+    // inout field to be used by settings modal
+    const SettingsModalNumericInput = defineComponent({
+      template: document.getElementById('settings-modal-numeric-input').innerHTML,
+      props: ['configKey', 'configDefault', 'configInfo', 'modelValue'],
+    });
+
     // coversations is stored in localStorage
     // format: { [convId]: { id: string, lastModified: number, messages: [...] } }
     // convId is a string prefixed with 'conv-'
@@ -380,6 +452,7 @@
     const mainApp = createApp({
       components: {
         VueMarkdown,
+        SettingsModalNumericInput,
       },
       data() {
         return {
@@ -397,6 +470,7 @@
           // const
           themes: THEMES,
           configDefault: {...CONFIG_DEFAULT},
+          configInfo: {...CONFIG_INFO},
         }
       },
       computed: {},
@@ -473,8 +547,22 @@
               stream: true,
               cache_prompt: true,
               temperature: this.config.temperature,
+              dynatemp_range: this.config.dynatemp_range,
+              dynatemp_exponent: this.config.dynatemp_exponent,
               top_k: this.config.top_k,
               top_p: this.config.top_p,
+              min_p: this.config.min_p,
+              typical_p: this.config.typical_p,
+              xtc_probability: this.config.xtc_probability,
+              xtc_threshold: this.config.xtc_threshold,
+              repeat_last_n: this.config.repeat_last_n,
+              repeat_penalty: this.config.repeat_penalty,
+              presence_penalty: this.config.presence_penalty,
+              frequency_penalty: this.config.frequency_penalty,
+              dry_multiplier: this.config.dry_multiplier,
+              dry_base: this.config.dry_base,
+              dry_allowed_length: this.config.dry_allowed_length,
+              dry_penalty_last_n: this.config.dry_penalty_last_n,
               max_tokens: this.config.max_tokens,
               ...(this.config.custom.length ? JSON.parse(this.config.custom) : {}),
               ...(this.config.apiKey ? { api_key: this.config.apiKey } : {}),
diff --git a/ggml/src/ggml-vulkan.cpp b/ggml/src/ggml-vulkan.cpp
index a8e78c4db..6c4c92262 100644
--- a/ggml/src/ggml-vulkan.cpp
+++ b/ggml/src/ggml-vulkan.cpp
@@ -3147,7 +3147,7 @@ static void ggml_vk_mul_mat_q_f16(ggml_backend_vk_context * ctx, vk_context& sub
     const bool qx_needs_dequant = mmp == nullptr || x_non_contig;
     const bool qy_needs_dequant = (src1->type != GGML_TYPE_F16 && !y_f32_kernel) || y_non_contig;
 
-    if (mmp == nullptr) {
+    if (qx_needs_dequant) {
         // Fall back to dequant + f16 mulmat
         mmp = ggml_vk_get_mul_mat_mat_pipeline(ctx, GGML_TYPE_F16, y_f32_kernel ? GGML_TYPE_F32 : GGML_TYPE_F16);
     }
@@ -3630,9 +3630,19 @@ static void ggml_vk_mul_mat_vec_nc_f16_f32(ggml_backend_vk_context * ctx, vk_con
 
 static void ggml_vk_mul_mat(ggml_backend_vk_context * ctx, vk_context& subctx, const ggml_tensor * src0, const ggml_tensor * src1, ggml_tensor * dst, bool dryrun = false) {
     VK_LOG_DEBUG("ggml_vk_mul_mat(" << src0 << ", " << src1 << ", " << dst << ")");
-    if (src0->type == GGML_TYPE_F16 && ggml_is_permuted(src0) && ggml_is_permuted(src1) && dst->ne[1] == 1) {
+    if (src0->type == GGML_TYPE_F16 && ggml_is_permuted(src0) && ggml_is_permuted(src1) && dst->ne[1] == 1 &&
+        // detect 0213 permutation, and batch size of 1
+        src0->nb[0] <= src0->nb[2] &&
+        src0->nb[2] <= src0->nb[1] &&
+        src0->nb[1] <= src0->nb[3] &&
+        src1->nb[0] <= src1->nb[2] &&
+        src1->nb[2] <= src1->nb[1] &&
+        src1->nb[1] <= src1->nb[3] &&
+        src0->ne[3] == 1 &&
+        src1->ne[3] == 1) {
         ggml_vk_mul_mat_vec_p021_f16_f32(ctx, subctx, src0, src1, dst, dryrun);
-    } else if (src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && dst->ne[1] == 1) {
+    } else if (src0->type == GGML_TYPE_F16 && !ggml_is_contiguous(src0) && !ggml_is_transposed(src1) && dst->ne[1] == 1 &&
+               !ggml_is_permuted(src0) && !ggml_is_permuted(src1)) {
         ggml_vk_mul_mat_vec_nc_f16_f32(ctx, subctx, src0, src1, dst, dryrun);
     } else if (dst->ne[1] == 1 && (src0->type == GGML_TYPE_F32 || src0->type == GGML_TYPE_F16 || ggml_is_quantized(src0->type))) {
         ggml_vk_mul_mat_vec_q_f16(ctx, subctx, src0, src1, dst, dryrun);
@@ -3708,7 +3718,7 @@ static void ggml_vk_mul_mat_id_q_f16(ggml_backend_vk_context * ctx, vk_context&
     const bool qx_needs_dequant = mmp == nullptr || x_non_contig;
     const bool qy_needs_dequant = (src1->type != GGML_TYPE_F16 && !y_f32_kernel) || y_non_contig;
 
-    if (mmp == nullptr) {
+    if (qx_needs_dequant) {
         GGML_ABORT("fatal error");
     }
 
@@ -4470,7 +4480,7 @@ static void ggml_vk_op_f32(ggml_backend_vk_context * ctx, vk_context& subctx, co
             const uint32_t OH = is_2D ? dst->ne[2] : 1;
             const uint32_t OW =         dst->ne[1];
 
-            const uint32_t batch = src1->ne[3];
+            const uint32_t batch = src1->ne[is_2D ? 3 : 2];
 
             elements = { OW * KW * KH, OH, batch * IC };
         } break;
@@ -4915,7 +4925,7 @@ static void ggml_vk_im2col(ggml_backend_vk_context * ctx, vk_context& subctx, co
     const uint32_t OW =         dst->ne[1];
 
     const uint32_t offset_delta = src1->nb[is_2D ? 2 : 1] / 4; // nb is byte offset, src is type float32
-    const uint32_t batch_offset = src1->nb[3] / 4; // nb is byte offset, src is type float32
+    const uint32_t batch_offset = src1->nb[is_2D ? 3 : 2] / 4; // nb is byte offset, src is type float32
 
     const uint32_t pelements = OW * KW * KH;
 
@@ -6804,6 +6814,11 @@ static bool ggml_backend_vk_device_supports_op(ggml_backend_dev_t dev, const ggm
                 if (a->ne[3] != b->ne[3]) {
                     return false;
                 }
+                if (!(ggml_vk_dim01_contiguous(op->src[0]) || op->src[0]->type == GGML_TYPE_F32 || op->src[0]->type == GGML_TYPE_F16) ||
+                    !(ggml_vk_dim01_contiguous(op->src[1]) || op->src[1]->type == GGML_TYPE_F32 || op->src[1]->type == GGML_TYPE_F16)) {
+                    return false;
+                }
+
                 return true;
             } break;
         case GGML_OP_GET_ROWS: