From 0537a3f6b289c380fa4ba9e8cf29b94896f2ea69 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 01:43:39 -0700
Subject: [PATCH 1/8] merged
 https://github.com/zpin/llama-cpp-python/tree/xtc_dry with latest changes

---
 llama_cpp/_internals.py | 22 +++++++++++++++++++++
 llama_cpp/llama.py      | 44 +++++++++++++++++++++++++++++++++++++++++
 llama_cpp/llama_cpp.py  | 35 ++++++++++++++++++++++++++++++++
 3 files changed, 101 insertions(+)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index b5175a7f2..52a8adba3 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -703,6 +703,28 @@ def add_temp_ext(self, t: float, delta: float, exponent: float):
     def add_xtc(self, p: float, t: float, min_keep: int, seed: int):
         sampler = llama_cpp.llama_sampler_init_xtc(p, t, min_keep, seed)
         llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
+    
+    # def add_dry(self, model: LlamaModel, ctx: LlamaContext, multiplier: float, base: float,
+    #             allowed_length: int, penalty_last_n: int, seq_breakers: list[str] = []):
+
+    #     # Convert Python strings to bytes
+    #     seq_breakers_bytes = [s.encode('utf-8') for s in seq_breakers]
+    #     # Create array of char*
+    #     arr = (ctypes.c_char_p * len(seq_breakers_bytes))(*seq_breakers_bytes)
+    #     sampler = llama_cpp.llama_sampler_init_dry(model.vocab, ctx.n_ctx(), multiplier, base,
+    #                                             allowed_length, penalty_last_n,
+    #                                             arr, len(seq_breakers))
+    #     self._add_sampler(sampler)
+    def add_dry(self, m: float, b: float, l: int, n: int, breakers: List[str]):
+        # Convert breakers to C array
+        seq_breakers_bytes = [s.encode('utf-8') for s in breakers]
+        arr = (ctypes.c_char_p * len(seq_breakers_bytes))(*seq_breakers_bytes)
+        sampler = llama_cpp.llama_sampler_init_dry(
+            None, 0, m, b, l, n,
+            arr,
+            len(breakers)
+        )
+        llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 
     def add_top_n_sigma(self, n: float):
         sampler = llama_cpp.llama_sampler_init_top_n_sigma(n)
diff --git a/llama_cpp/llama.py b/llama_cpp/llama.py
index 71d94ebd8..00eeb8e41 100644
--- a/llama_cpp/llama.py
+++ b/llama_cpp/llama.py
@@ -682,6 +682,13 @@ def _init_sampler(
         mirostat_mode: int = 0,
         mirostat_eta: float = 0.1,
         mirostat_tau: float = 5.0,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
         grammar: Optional[LlamaGrammar] = None,
@@ -749,11 +756,13 @@ def apply_func(token_data_array: llama_cpp.llama_token_data_array_p):
             else:
                 n_probs = 0
                 min_keep = max(1, n_probs)
+                sampler.add_dry(self._model, self._ctx, dry_multiplier, dry_base, dry_allowed_length, dry_penalty_last_n, dry_seq_breakers)
                 sampler.add_top_k(top_k)
                 sampler.add_typical(typical_p, min_keep)
                 sampler.add_top_p(top_p, min_keep)
                 sampler.add_min_p(min_p, min_keep)
                 sampler.add_temp(temp)
+                sampler.add_xtc(xtc_probability, xtc_threshold, min_keep, self._seed)
                 sampler.add_dist(self._seed)
         return sampler
 
@@ -771,6 +780,13 @@ def sample(
         mirostat_mode: int = 0,
         mirostat_eta: float = 0.1,
         mirostat_tau: float = 5.0,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
         grammar: Optional[LlamaGrammar] = None,
@@ -806,6 +822,13 @@ def sample(
                 mirostat_mode=mirostat_mode,
                 mirostat_tau=mirostat_tau,
                 mirostat_eta=mirostat_eta,
+                xtc_probability=xtc_probability,
+                xtc_threshold=xtc_threshold,
+                dry_multiplier=dry_multiplier,
+                dry_allowed_length=dry_allowed_length,
+                dry_base=dry_base,
+                dry_penalty_last_n=dry_penalty_last_n,
+                dry_seq_breakers=dry_seq_breakers,
                 penalize_nl=penalize_nl,
                 logits_processor=logits_processor,
                 grammar=grammar,
@@ -835,6 +858,13 @@ def generate(
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
@@ -874,6 +904,13 @@ def generate(
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
+            xtc_probability=xtc_probability,
+            xtc_threshold=xtc_threshold,
+            dry_multiplier=dry_multiplier,
+            dry_allowed_length=dry_allowed_length,
+            dry_base=dry_base,
+            dry_penalty_last_n=dry_penalty_last_n,
+            dry_seq_breakers=dry_seq_breakers,
             penalize_nl=penalize_nl,
             logits_processor=logits_processor,
             grammar=grammar,
@@ -926,6 +963,13 @@ def generate(
                     mirostat_mode=mirostat_mode,
                     mirostat_tau=mirostat_tau,
                     mirostat_eta=mirostat_eta,
+                    xtc_probability=xtc_probability,
+                    xtc_threshold=xtc_threshold,
+                    dry_multiplier=dry_multiplier,
+                    dry_allowed_length=dry_allowed_length,
+                    dry_base=dry_base,
+                    dry_penalty_last_n=dry_penalty_last_n,
+                    dry_seq_breakers=dry_seq_breakers,
                     logits_processor=logits_processor,
                     grammar=grammar,
                     penalize_nl=penalize_nl,
diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index 711d42a6a..f1e4324b0 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -3884,6 +3884,41 @@ def llama_sampler_init_xtc(
 ) -> llama_sampler_p:
     ...
 
+#    LLAMA_API struct llama_sampler *    llama_sampler_init_dry(
+#            const struct llama_vocab *  vocab,
+#                             int32_t    context_size,
+#                               float    dry_multiplier,
+#                               float    dry_base,
+#                             int32_t    dry_allowed_length,
+#                             int32_t    dry_penalty_last_n,
+#                          const char ** seq_breakers,
+#                              size_t    num_breakers);
+@ctypes_function(
+"llama_sampler_init_dry",
+    [
+        llama_vocab_p_ctypes,
+        ctypes.c_int32,
+        ctypes.c_float,
+        ctypes.c_float,
+        ctypes.c_int32,
+        ctypes.c_int32,
+        ctypes.POINTER(ctypes.c_char_p),
+        ctypes.c_size_t
+    ],
+    llama_sampler_p_ctypes,
+)
+def llama_sampler_init_dry(
+    vocab: llama_vocab_p,
+    context_size: int,
+    dry_multiplier: float,
+    dry_base: float,
+    dry_allowed_length: int,
+    dry_penalty_last_n: int,
+    seq_breakers: list[str],
+    num_breakers: int,
+) -> llama_sampler_p:
+    ...
+
 
 # /// @details Top n sigma sampling as described in academic paper "Top-nσ: Not All Logits Are You Need" https://arxiv.org/pdf/2411.07641
 # LLAMA_API struct llama_sampler * llama_sampler_init_top_n_sigma(float   n);

From 6d0370bb402b8206ff5866c6cd49c72ef189add5 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 02:06:42 -0700
Subject: [PATCH 2/8] added high level support

---
 llama_cpp/llama.py | 63 ++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 63 insertions(+)

diff --git a/llama_cpp/llama.py b/llama_cpp/llama.py
index 00eeb8e41..64b5ffb30 100644
--- a/llama_cpp/llama.py
+++ b/llama_cpp/llama.py
@@ -1186,6 +1186,13 @@ def _create_completion(
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -1374,6 +1381,13 @@ def logit_bias_processor(
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
+            xtc_probability=xtc_probability,
+            xtc_threshold=xtc_threshold,
+            dry_multiplier=dry_multiplier,
+            dry_allowed_length=dry_allowed_length,
+            dry_base=dry_base,
+            dry_penalty_last_n=dry_penalty_last_n,
+            dry_seq_breakers=dry_seq_breakers,
             frequency_penalty=frequency_penalty,
             presence_penalty=presence_penalty,
             repeat_penalty=repeat_penalty,
@@ -1806,6 +1820,13 @@ def create_completion(
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -1835,6 +1856,13 @@ def create_completion(
             mirostat_mode: The mirostat sampling mode.
             mirostat_tau: The target cross-entropy (or surprise) value you want to achieve for the generated text. A higher value corresponds to more surprising or less predictable text, while a lower value corresponds to less surprising or more predictable text.
             mirostat_eta: The learning rate used to update `mu` based on the error between the target and observed surprisal of the sampled word. A larger learning rate will cause `mu` to be updated more quickly, while a smaller learning rate will result in slower updates.
+            xtc_probability: ???
+            xtc_threshold: ???
+            dry_multiplier: ???
+            dry_allowed_length: ???
+            dry_base: ???
+            dry_penalty_last_n: ???
+            dry_seq_breakers: ???
             model: The name to use for the model in the completion object.
             stopping_criteria: A list of stopping criteria to use.
             logits_processor: A list of logits processors to use.
@@ -1869,6 +1897,13 @@ def create_completion(
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
+            xtc_probability=xtc_probability,
+            xtc_threshold=xtc_threshold,
+            dry_multiplier=dry_multiplier,
+            dry_allowed_length=dry_allowed_length,
+            dry_base=dry_base,
+            dry_penalty_last_n=dry_penalty_last_n,
+            dry_seq_breakers=dry_seq_breakers,
             model=model,
             stopping_criteria=stopping_criteria,
             logits_processor=logits_processor,
@@ -1903,6 +1938,13 @@ def __call__(
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -1966,6 +2008,13 @@ def __call__(
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
+            xtc_probability=xtc_probability,
+            xtc_threshold=xtc_threshold,
+            dry_multiplier=dry_multiplier,
+            dry_allowed_length=dry_allowed_length,
+            dry_base=dry_base,
+            dry_penalty_last_n=dry_penalty_last_n,
+            dry_seq_breakers=dry_seq_breakers,
             model=model,
             stopping_criteria=stopping_criteria,
             logits_processor=logits_processor,
@@ -1997,6 +2046,13 @@ def create_chat_completion(
         mirostat_mode: int = 0,
         mirostat_tau: float = 5.0,
         mirostat_eta: float = 0.1,
+        xtc_probability: float = 0.0,
+        xtc_threshold: float = 0.1,
+        dry_multiplier: float = 0.0,
+        dry_allowed_length: int = 2,
+        dry_base: float = 1.75,
+        dry_penalty_last_n: int = -1,
+        dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         logits_processor: Optional[LogitsProcessorList] = None,
         grammar: Optional[LlamaGrammar] = None,
@@ -2070,6 +2126,13 @@ def create_chat_completion(
             mirostat_mode=mirostat_mode,
             mirostat_tau=mirostat_tau,
             mirostat_eta=mirostat_eta,
+            xtc_probability=xtc_probability,
+            xtc_threshold=xtc_threshold,
+            dry_multiplier=dry_multiplier,
+            dry_allowed_length=dry_allowed_length,
+            dry_base=dry_base,
+            dry_penalty_last_n=dry_penalty_last_n,
+            dry_seq_breakers=dry_seq_breakers,
             model=model,
             logits_processor=logits_processor,
             grammar=grammar,

From 06a2cddc2b4ee2a3291411a616a9547dbe7c72ac Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 02:55:54 -0700
Subject: [PATCH 3/8] maybe fixed it?

---
 llama_cpp/_internals.py | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index 52a8adba3..0659788fd 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -715,16 +715,16 @@ def add_xtc(self, p: float, t: float, min_keep: int, seed: int):
     #                                             allowed_length, penalty_last_n,
     #                                             arr, len(seq_breakers))
     #     self._add_sampler(sampler)
-    def add_dry(self, m: float, b: float, l: int, n: int, breakers: List[str]):
-        # Convert breakers to C array
-        seq_breakers_bytes = [s.encode('utf-8') for s in breakers]
-        arr = (ctypes.c_char_p * len(seq_breakers_bytes))(*seq_breakers_bytes)
-        sampler = llama_cpp.llama_sampler_init_dry(
-            None, 0, m, b, l, n,
-            arr,
-            len(breakers)
-        )
-        llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
+    # def add_dry(self, m: float, b: float, l: int, n: int, breakers: List[str]):
+    #     # Convert breakers to C array
+    #     seq_breakers_bytes = [s.encode('utf-8') for s in breakers]
+    #     arr = (ctypes.c_char_p * len(seq_breakers_bytes))(*seq_breakers_bytes)
+    #     sampler = llama_cpp.llama_sampler_init_dry(
+    #         None, 0, m, b, l, n,
+    #         arr,
+    #         len(breakers)
+    #     )
+    #     llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 
     def add_top_n_sigma(self, n: float):
         sampler = llama_cpp.llama_sampler_init_top_n_sigma(n)
@@ -790,8 +790,8 @@ def add_dry(
         model: LlamaModel,
         n_ctx_train: int,
         dry_multiplier: float,
-        dry_base: float,
         dry_allowed_length: int,
+        dry_base: float,
         dry_penalty_last_n: int,
         seq_breakers: List[str]
     ):
@@ -804,8 +804,8 @@ def add_dry(
             model.vocab,
             n_ctx_train,
             dry_multiplier,
-            dry_base,
             dry_allowed_length,
+            dry_base,
             dry_penalty_last_n,
             breaker_ptrs,
             len(seq_breakers)

From 1b1aa180cac8ce0686e16052ea4e8fdc680dc420 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 03:09:00 -0700
Subject: [PATCH 4/8] ugh

---
 llama_cpp/_internals.py | 18 +++++++++---------
 1 file changed, 9 insertions(+), 9 deletions(-)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index 0659788fd..ba26039f9 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -790,8 +790,8 @@ def add_dry(
         model: LlamaModel,
         n_ctx_train: int,
         dry_multiplier: float,
-        dry_allowed_length: int,
         dry_base: float,
+        dry_allowed_length: int,
         dry_penalty_last_n: int,
         seq_breakers: List[str]
     ):
@@ -801,14 +801,14 @@ def add_dry(
             breaker_ptrs[i] = breaker.encode("utf-8")
         
         sampler = llama_cpp.llama_sampler_init_dry(
-            model.vocab,
-            n_ctx_train,
-            dry_multiplier,
-            dry_allowed_length,
-            dry_base,
-            dry_penalty_last_n,
-            breaker_ptrs,
-            len(seq_breakers)
+            vocab=model.vocab,
+            n_ctx_train=n_ctx_train,
+            dry_multiplier=dry_multiplier,
+            dry_base=dry_base,
+            dry_allowed_length=dry_allowed_length,
+            dry_penalty_last_n=dry_penalty_last_n,
+            breaker_ptrs=breaker_ptrs,
+            num_breakers=len(seq_breakers)
         )
         llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 

From b37e24336475d21fe26a418b74daea31da1e4570 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 03:23:29 -0700
Subject: [PATCH 5/8] I guess it doesn't like kwargs?

---
 llama_cpp/_internals.py | 24 ++++++++++++++++--------
 1 file changed, 16 insertions(+), 8 deletions(-)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index ba26039f9..aa5f7da1f 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -801,14 +801,22 @@ def add_dry(
             breaker_ptrs[i] = breaker.encode("utf-8")
         
         sampler = llama_cpp.llama_sampler_init_dry(
-            vocab=model.vocab,
-            n_ctx_train=n_ctx_train,
-            dry_multiplier=dry_multiplier,
-            dry_base=dry_base,
-            dry_allowed_length=dry_allowed_length,
-            dry_penalty_last_n=dry_penalty_last_n,
-            breaker_ptrs=breaker_ptrs,
-            num_breakers=len(seq_breakers)
+            # vocab=model.vocab,
+            # n_ctx_train=n_ctx_train,
+            # dry_multiplier=dry_multiplier,
+            # dry_base=dry_base,
+            # dry_allowed_length=dry_allowed_length,
+            # dry_penalty_last_n=dry_penalty_last_n,
+            # breaker_ptrs=breaker_ptrs,
+            # num_breakers=len(seq_breakers)
+            model.vocab,
+            n_ctx_train,
+            dry_multiplier,
+            dry_base,
+            dry_allowed_length,
+            dry_penalty_last_n,
+            breaker_ptrs,
+            len(seq_breakers)
         )
         llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 

From c29697d74036ca44d527192e60456028db805627 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 03:41:07 -0700
Subject: [PATCH 6/8] I wish I knew C++

---
 llama_cpp/_internals.py | 6 +++---
 llama_cpp/llama_cpp.py  | 8 ++++----
 2 files changed, 7 insertions(+), 7 deletions(-)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index aa5f7da1f..59e335db2 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -809,14 +809,14 @@ def add_dry(
             # dry_penalty_last_n=dry_penalty_last_n,
             # breaker_ptrs=breaker_ptrs,
             # num_breakers=len(seq_breakers)
-            model.vocab,
+            # model.vocab,
             n_ctx_train,
             dry_multiplier,
             dry_base,
             dry_allowed_length,
             dry_penalty_last_n,
-            breaker_ptrs,
-            len(seq_breakers)
+            # breaker_ptrs,
+            # len(seq_breakers)
         )
         llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 
diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index f1e4324b0..22225012b 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -3908,14 +3908,14 @@ def llama_sampler_init_xtc(
     llama_sampler_p_ctypes,
 )
 def llama_sampler_init_dry(
-    vocab: llama_vocab_p,
-    context_size: int,
+    # vocab: llama_vocab_p,
+    # context_size: int,
     dry_multiplier: float,
     dry_base: float,
     dry_allowed_length: int,
     dry_penalty_last_n: int,
-    seq_breakers: list[str],
-    num_breakers: int,
+    # seq_breakers: list[str],
+    # num_breakers: int,
 ) -> llama_sampler_p:
     ...
 

From d0da4f10db9c187bc19a524d7a20e6dddade7a99 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 03:55:04 -0700
Subject: [PATCH 7/8] I really hate that I don't know C++

---
 llama_cpp/llama_cpp.py | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index 22225012b..58f93b06e 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -3896,20 +3896,20 @@ def llama_sampler_init_xtc(
 @ctypes_function(
 "llama_sampler_init_dry",
     [
-        llama_vocab_p_ctypes,
+        # llama_vocab_p_ctypes,
         ctypes.c_int32,
         ctypes.c_float,
         ctypes.c_float,
         ctypes.c_int32,
         ctypes.c_int32,
-        ctypes.POINTER(ctypes.c_char_p),
-        ctypes.c_size_t
+        # ctypes.POINTER(ctypes.c_char_p),
+        # ctypes.c_size_t
     ],
     llama_sampler_p_ctypes,
 )
 def llama_sampler_init_dry(
     # vocab: llama_vocab_p,
-    # context_size: int,
+    context_size: int,
     dry_multiplier: float,
     dry_base: float,
     dry_allowed_length: int,

From 08b3c9f1967ee3daa55f03cfd81efa5b5f0b77b7 Mon Sep 17 00:00:00 2001
From: Pathos <29135741+Pathos14489@users.noreply.github.com>
Date: Wed, 8 Oct 2025 06:59:41 -0700
Subject: [PATCH 8/8] I think I might be an idiot... Excuse: I'm tired

---
 llama_cpp/_internals.py | 10 +++++-----
 llama_cpp/llama_cpp.py  | 12 ++++++------
 2 files changed, 11 insertions(+), 11 deletions(-)

diff --git a/llama_cpp/_internals.py b/llama_cpp/_internals.py
index 59e335db2..6f396e49a 100644
--- a/llama_cpp/_internals.py
+++ b/llama_cpp/_internals.py
@@ -788,7 +788,7 @@ def add_penalties(
     def add_dry(
         self,
         model: LlamaModel,
-        n_ctx_train: int,
+        ctx: LlamaContext,
         dry_multiplier: float,
         dry_base: float,
         dry_allowed_length: int,
@@ -809,14 +809,14 @@ def add_dry(
             # dry_penalty_last_n=dry_penalty_last_n,
             # breaker_ptrs=breaker_ptrs,
             # num_breakers=len(seq_breakers)
-            # model.vocab,
-            n_ctx_train,
+            model.vocab,
+            ctx.n_ctx(),
             dry_multiplier,
             dry_base,
             dry_allowed_length,
             dry_penalty_last_n,
-            # breaker_ptrs,
-            # len(seq_breakers)
+            breaker_ptrs,
+            len(seq_breakers)
         )
         llama_cpp.llama_sampler_chain_add(self.sampler, sampler)
 
diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py
index 58f93b06e..f1e4324b0 100644
--- a/llama_cpp/llama_cpp.py
+++ b/llama_cpp/llama_cpp.py
@@ -3896,26 +3896,26 @@ def llama_sampler_init_xtc(
 @ctypes_function(
 "llama_sampler_init_dry",
     [
-        # llama_vocab_p_ctypes,
+        llama_vocab_p_ctypes,
         ctypes.c_int32,
         ctypes.c_float,
         ctypes.c_float,
         ctypes.c_int32,
         ctypes.c_int32,
-        # ctypes.POINTER(ctypes.c_char_p),
-        # ctypes.c_size_t
+        ctypes.POINTER(ctypes.c_char_p),
+        ctypes.c_size_t
     ],
     llama_sampler_p_ctypes,
 )
 def llama_sampler_init_dry(
-    # vocab: llama_vocab_p,
+    vocab: llama_vocab_p,
     context_size: int,
     dry_multiplier: float,
     dry_base: float,
     dry_allowed_length: int,
     dry_penalty_last_n: int,
-    # seq_breakers: list[str],
-    # num_breakers: int,
+    seq_breakers: list[str],
+    num_breakers: int,
 ) -> llama_sampler_p:
     ...