diff --git a/Package.resolved b/Package.resolved
index b689a691..0b86bc46 100644
--- a/Package.resolved
+++ b/Package.resolved
@@ -19,6 +19,15 @@
         "version" : "1.3.1"
       }
     },
+    {
+      "identity" : "llama.swift",
+      "kind" : "remoteSourceControl",
+      "location" : "https://github.com/mattt/llama.swift",
+      "state" : {
+        "revision" : "716419d4d7aa542fce301e809cde7234c68ddbc6",
+        "version" : "2.10549.0"
+      }
+    },
     {
       "identity" : "partialjsondecoder",
       "kind" : "remoteSourceControl",
diff --git a/Sources/AnyLanguageModel/Models/LlamaLanguageModel.swift b/Sources/AnyLanguageModel/Models/LlamaLanguageModel.swift
index 6bcd8045..e20ca8d1 100644
--- a/Sources/AnyLanguageModel/Models/LlamaLanguageModel.swift
+++ b/Sources/AnyLanguageModel/Models/LlamaLanguageModel.swift
@@ -137,6 +137,14 @@ import Foundation
             /// Mirostat sampling mode for adaptive perplexity control.
             public var mirostat: MirostatMode?
 
+            /// Text appended after the assistant header of the rendered prompt.
+            ///
+            /// The model continues generating from this text. Use it to steer the
+            /// start of the response, for example prefilling an empty
+            /// `<think></think>` block to suppress a model's default reasoning
+            /// output when its chat template offers no switch for it.
+            public var assistantPrefill: String?
+
             /// Creates custom generation options for llama.cpp.
             public init(
                 contextSize: UInt32? = nil,
@@ -150,7 +158,8 @@ import Foundation
                 repeatLastN: Int32? = nil,
                 frequencyPenalty: Float? = nil,
                 presencePenalty: Float? = nil,
-                mirostat: MirostatMode? = nil
+                mirostat: MirostatMode? = nil,
+                assistantPrefill: String? = nil
             ) {
                 self.contextSize = contextSize
                 self.batchSize = batchSize
@@ -164,6 +173,7 @@ import Foundation
                 self.frequencyPenalty = frequencyPenalty
                 self.presencePenalty = presencePenalty
                 self.mirostat = mirostat
+                self.assistantPrefill = assistantPrefill
             }
 
             /// Default llama.cpp options used when none are provided at runtime.
@@ -321,6 +331,7 @@ import Foundation
             var frequencyPenalty: Float
             var presencePenalty: Float
             var mirostat: CustomGenerationOptions.MirostatMode?
+            var assistantPrefill: String?
             var sampling: GenerationOptions.SamplingMode?
             var maximumResponseTokens: Int?
 
@@ -337,6 +348,7 @@ import Foundation
                 frequencyPenalty: Float = 0.0,
                 presencePenalty: Float = 0.0,
                 mirostat: CustomGenerationOptions.MirostatMode? = nil,
+                assistantPrefill: String? = nil,
                 sampling: GenerationOptions.SamplingMode? = nil,
                 maximumResponseTokens: Int? = nil
             ) {
@@ -352,6 +364,7 @@ import Foundation
                 self.frequencyPenalty = frequencyPenalty
                 self.presencePenalty = presencePenalty
                 self.mirostat = mirostat
+                self.assistantPrefill = assistantPrefill
                 self.sampling = sampling
                 self.maximumResponseTokens = maximumResponseTokens
             }
@@ -389,6 +402,7 @@ import Foundation
                         frequencyPenalty: base.frequencyPenalty,
                         presencePenalty: base.presencePenalty,
                         mirostat: base.mirostat,
+                        assistantPrefill: base.assistantPrefill,
                         sampling: sampling ?? base.sampling,
                         maximumResponseTokens: maximumResponseTokens ?? base.maximumResponseTokens
                     )
@@ -407,6 +421,7 @@ import Foundation
                 self.frequencyPenalty = options.frequencyPenalty ?? base.frequencyPenalty
                 self.presencePenalty = options.presencePenalty ?? base.presencePenalty
                 self.mirostat = options.mirostat ?? base.mirostat
+                self.assistantPrefill = options.assistantPrefill ?? base.assistantPrefill
                 self.sampling = sampling ?? base.sampling
                 self.maximumResponseTokens = maximumResponseTokens ?? base.maximumResponseTokens
             }
@@ -509,10 +524,11 @@ import Foundation
             if includeSchemaInPrompt, type != String.self {
                 fullPrompt = try formatPrompt(
                     for: session,
-                    extraSystemMessage: schemaPrompt(for: type.generationSchema)
+                    extraSystemMessage: schemaPrompt(for: type.generationSchema),
+                    assistantPrefill: runtimeOptions.assistantPrefill
                 )
             } else {
-                fullPrompt = try formatPrompt(for: session)
+                fullPrompt = try formatPrompt(for: session, assistantPrefill: runtimeOptions.assistantPrefill)
             }
 
             if type == String.self {
@@ -600,7 +616,10 @@ import Foundation
                             llama_set_n_threads(context, runtimeOptions.threads, runtimeOptions.threads)
 
                             var accumulatedText = ""
-                            let fullPrompt = try self.formatPrompt(for: session)
+                            let fullPrompt = try self.formatPrompt(
+                                for: session,
+                                assistantPrefill: runtimeOptions.assistantPrefill
+                            )
 
                             do {
                                 for try await tokenText in generateTextStream(
@@ -673,8 +692,7 @@ import Foundation
             params.n_gpu_layers = 0
 
             // Try to reduce memory usage
-            params.use_mmap = true
-            params.use_mlock = false
+            params.load_mode = LLAMA_LOAD_MODE_MMAP
             return params
         }
 
@@ -829,6 +847,7 @@ import Foundation
                 llama_sampler_chain_add(
                     samplerPtr,
                     llama_sampler_init_penalties(
+                        llama_vocab_n_tokens(vocab),
                         effectiveRepeatLastN,
                         effectiveRepeatPenalty,
                         effectiveFrequencyPenalty,
@@ -958,6 +977,7 @@ import Foundation
                 llama_sampler_chain_add(
                     samplerPointer,
                     llama_sampler_init_penalties(
+                        llama_vocab_n_tokens(vocab),
                         options.repeatLastN,
                         options.repeatPenalty,
                         options.frequencyPenalty,
@@ -1197,6 +1217,7 @@ import Foundation
                     llama_sampler_chain_add(
                         samplerPtr,
                         llama_sampler_init_penalties(
+                            llama_vocab_n_tokens(vocab),
                             effectiveRepeatLastN,
                             effectiveRepeatPenalty,
                             effectiveFrequencyPenalty,
@@ -1368,7 +1389,8 @@ import Foundation
 
         private func formatPrompt(
             for session: LanguageModelSession,
-            extraSystemMessage: String? = nil
+            extraSystemMessage: String? = nil,
+            assistantPrefill: String? = nil
         ) throws -> String {
             guard let model = self.model else {
                 throw LlamaLanguageModelError.modelLoadFailed
@@ -1452,9 +1474,14 @@ import Foundation
                 throw LlamaLanguageModelError.encodingFailed
             }
 
-            return buffer.withUnsafeBytes { rawBuffer in
+            let rendered = buffer.withUnsafeBytes { rawBuffer in
                 String(decoding: rawBuffer.prefix(Int(result)), as: UTF8.self)
             }
+
+            if let assistantPrefill, !assistantPrefill.isEmpty {
+                return rendered + assistantPrefill
+            }
+            return rendered
         }
 
         private func extractText(from segments: [Transcript.Segment]) -> String {
diff --git a/Tests/AnyLanguageModelTests/LlamaLanguageModelTests.swift b/Tests/AnyLanguageModelTests/LlamaLanguageModelTests.swift
index 3d9e32c7..fa2ae7c2 100644
--- a/Tests/AnyLanguageModelTests/LlamaLanguageModelTests.swift
+++ b/Tests/AnyLanguageModelTests/LlamaLanguageModelTests.swift
@@ -45,7 +45,8 @@ import Testing
                 repeatLastN: 48,
                 frequencyPenalty: 0.05,
                 presencePenalty: 0.05,
-                mirostat: .v2(tau: 5.0, eta: 0.2)
+                mirostat: .v2(tau: 5.0, eta: 0.2),
+                assistantPrefill: "<think></think>"
             )
             options[custom: LlamaLanguageModel.self] = custom
 
@@ -62,6 +63,7 @@ import Testing
             #expect(retrieved?.frequencyPenalty == 0.05)
             #expect(retrieved?.presencePenalty == 0.05)
             #expect(retrieved?.mirostat == .v2(tau: 5.0, eta: 0.2))
+            #expect(retrieved?.assistantPrefill == "<think></think>")
         }
 
         @Test func customGenerationOptionsDefaults() {