Renamed dry_range to dry_penalty_last_n to match llama.cpp, changed default DRY range from 0 to -1

Author: zpin

llama_cpp/llama.py

@@ -685,7 +685,7 @@ def _init_sampler(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -754,7 +754,7 @@ 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_range, dry_seq_breakers)
+                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)
@@ -783,7 +783,7 @@ def sample(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -825,7 +825,7 @@ def sample(
                 dry_multiplier=dry_multiplier,
                 dry_allowed_length=dry_allowed_length,
                 dry_base=dry_base,
-                dry_range=dry_range,
+                dry_penalty_last_n=dry_penalty_last_n,
                 dry_seq_breakers=dry_seq_breakers,
                 penalize_nl=penalize_nl,
                 logits_processor=logits_processor,
@@ -861,7 +861,7 @@ def generate(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         penalize_nl: bool = True,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -907,7 +907,7 @@ def generate(
             dry_multiplier=dry_multiplier,
             dry_allowed_length=dry_allowed_length,
             dry_base=dry_base,
-            dry_range=dry_range,
+            dry_penalty_last_n=dry_penalty_last_n,
             dry_seq_breakers=dry_seq_breakers,
             penalize_nl=penalize_nl,
             logits_processor=logits_processor,
@@ -966,7 +966,7 @@ def generate(
                     dry_multiplier=dry_multiplier,
                     dry_allowed_length=dry_allowed_length,
                     dry_base=dry_base,
-                    dry_range=dry_range,
+                    dry_penalty_last_n=dry_penalty_last_n,
                     dry_seq_breakers=dry_seq_breakers,
                     logits_processor=logits_processor,
                     grammar=grammar,
@@ -1189,7 +1189,7 @@ def _create_completion(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
@@ -1384,7 +1384,7 @@ def logit_bias_processor(
             dry_multiplier=dry_multiplier,
             dry_allowed_length=dry_allowed_length,
             dry_base=dry_base,
-            dry_range=dry_range,
+            dry_penalty_last_n=dry_penalty_last_n,
             dry_seq_breakers=dry_seq_breakers,
             frequency_penalty=frequency_penalty,
             presence_penalty=presence_penalty,
@@ -1823,7 +1823,7 @@ def create_completion(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
@@ -1893,7 +1893,7 @@ def create_completion(
             dry_multiplier=dry_multiplier,
             dry_allowed_length=dry_allowed_length,
             dry_base=dry_base,
-            dry_range=dry_range,
+            dry_penalty_last_n=dry_penalty_last_n,
             dry_seq_breakers=dry_seq_breakers,
             model=model,
             stopping_criteria=stopping_criteria,
@@ -1934,7 +1934,7 @@ def __call__(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         stopping_criteria: Optional[StoppingCriteriaList] = None,
@@ -2004,7 +2004,7 @@ def __call__(
             dry_multiplier=dry_multiplier,
             dry_allowed_length=dry_allowed_length,
             dry_base=dry_base,
-            dry_range=dry_range,
+            dry_penalty_last_n=dry_penalty_last_n,
             dry_seq_breakers=dry_seq_breakers,
             model=model,
             stopping_criteria=stopping_criteria,
@@ -2042,7 +2042,7 @@ def create_chat_completion(
         dry_multiplier: float = 0.0,
         dry_allowed_length: int = 2,
         dry_base: float = 1.75,
-        dry_range: int = 0,
+        dry_penalty_last_n: int = -1,
         dry_seq_breakers: list[str] = [],
         model: Optional[str] = None,
         logits_processor: Optional[LogitsProcessorList] = None,
@@ -2122,7 +2122,7 @@ def create_chat_completion(
             dry_multiplier=dry_multiplier,
             dry_allowed_length=dry_allowed_length,
             dry_base=dry_base,
-            dry_range=dry_range,
+            dry_penalty_last_n=dry_penalty_last_n,
             dry_seq_breakers=dry_seq_breakers,
             model=model,
             logits_processor=logits_processor,