Added support for min_p (#921)

* Added support for min_p

My small contribution to this great project.

Ref: https://github.com/ggerganov/llama.cpp/pull/3841

Closes: https://github.com/abetlen/llama-cpp-python/issues/911

* Fix for negative temp (sample_softmax)

llama_cpp/llama.py

@@ -1046,6 +1046,8 @@ class Llama:
         self,
         top_k: int = 40,
         top_p: float = 0.95,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         temp: float = 0.80,
         repeat_penalty: float = 1.1,
         frequency_penalty: float = 0.0,
@@ -1108,7 +1110,10 @@ class Llama:
                 grammar=grammar,
             )
 
-        if temp == 0.0:
+        if temp < 0.0:
+            self._ctx.sample_softmax(candidates=self._candidates)
+            id = self._candidates.candidates.data[0].id
+        elif temp == 0.0:
             id = self._ctx.sample_token_greedy(candidates=self._candidates)
         elif mirostat_mode == 1:
             self._ctx.sample_temp(candidates=self._candidates, temp=temp)
@@ -1130,8 +1135,9 @@ class Llama:
         else:
             self._ctx.sample_top_k(candidates=self._candidates, k=top_k, min_keep=1)
             self._ctx.sample_tail_free(candidates=self._candidates, z=tfs_z, min_keep=1)
-            self._ctx.sample_typical(candidates=self._candidates, p=1.0, min_keep=1)
+            self._ctx.sample_typical(candidates=self._candidates, p=typical_p, min_keep=1)
             self._ctx.sample_top_p(candidates=self._candidates, p=top_p, min_keep=1)
+            self._ctx.sample_min_p(candidates=self._candidates, p=min_p, min_keep=1)
             self._ctx.sample_temp(candidates=self._candidates, temp=temp)
             id = self._ctx.sample_token(candidates=self._candidates)
         if grammar is not None:
@@ -1143,6 +1149,8 @@ class Llama:
         tokens: Sequence[int],
         top_k: int = 40,
         top_p: float = 0.95,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         temp: float = 0.80,
         repeat_penalty: float = 1.1,
         reset: bool = True,
@@ -1200,6 +1208,8 @@ class Llama:
             token = self.sample(
                 top_k=top_k,
                 top_p=top_p,
+                min_p=min_p,
+                typical_p=typical_p,
                 temp=temp,
                 repeat_penalty=repeat_penalty,
                 frequency_penalty=frequency_penalty,
@@ -1298,6 +1308,8 @@ class Llama:
         max_tokens: Optional[int] = 16,
         temperature: float = 0.8,
         top_p: float = 0.95,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         logprobs: Optional[int] = None,
         echo: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
@@ -1398,6 +1410,8 @@ class Llama:
             prompt_tokens,
             top_k=top_k,
             top_p=top_p,
+            min_p=min_p,
+            typical_p=typical_p,
             temp=temperature,
             tfs_z=tfs_z,
             mirostat_mode=mirostat_mode,
@@ -1772,6 +1786,8 @@ class Llama:
         max_tokens: Optional[int] = 16,
         temperature: float = 0.8,
         top_p: float = 0.95,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         logprobs: Optional[int] = None,
         echo: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
@@ -1818,6 +1834,8 @@ class Llama:
             max_tokens=max_tokens,
             temperature=temperature,
             top_p=top_p,
+            min_p=min_p,
+            typical_p=typical_p,
             logprobs=logprobs,
             echo=echo,
             stop=stop,
@@ -1849,6 +1867,8 @@ class Llama:
         max_tokens: int = 128,
         temperature: float = 0.8,
         top_p: float = 0.95,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         logprobs: Optional[int] = None,
         echo: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
@@ -1895,6 +1915,8 @@ class Llama:
             max_tokens=max_tokens,
             temperature=temperature,
             top_p=top_p,
+            min_p=min_p,
+            typical_p=typical_p,
             logprobs=logprobs,
             echo=echo,
             stop=stop,
@@ -1924,6 +1946,8 @@ class Llama:
         temperature: float = 0.2,
         top_p: float = 0.95,
         top_k: int = 40,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         stream: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
         seed: Optional[int] = None,
@@ -1970,6 +1994,8 @@ class Llama:
             temperature=temperature,
             top_p=top_p,
             top_k=top_k,
+            min_p=min_p,
+            typical_p=typical_p,
             stream=stream,
             stop=stop,
             seed=seed,

llama_cpp/llama_chat_format.py

@@ -26,6 +26,8 @@ class LlamaChatCompletionHandler(Protocol):
         temperature: float = 0.2,
         top_p: float = 0.95,
         top_k: int = 40,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         stream: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
         seed: Optional[int] = None,
@@ -287,6 +289,8 @@ def register_chat_format(name: str):
             temperature: float = 0.2,
             top_p: float = 0.95,
             top_k: int = 40,
+            min_p: float = 0.05,
+            typical_p: float = 1.0,
             stream: bool = False,
             stop: Optional[Union[str, List[str]]] = [],
             seed: Optional[int] = None,
@@ -330,6 +334,8 @@ def register_chat_format(name: str):
                 temperature=temperature,
                 top_p=top_p,
                 top_k=top_k,
+                min_p=min_p,
+                typical_p=typical_p,
                 stream=stream,
                 stop=stop,
                 seed=seed,
@@ -579,6 +585,8 @@ def functionary_chat_handler(
     temperature: float = 0.2,
     top_p: float = 0.95,
     top_k: int = 40,
+    min_p: float = 0.05,
+    typical_p: float = 1.0,
     stream: bool = False,
     stop: Optional[Union[str, List[str]]] = [],
     response_format: Optional[llama_types.ChatCompletionRequestResponseFormat] = None,
@@ -761,6 +769,8 @@ def functionary_chat_handler(
             temperature=temperature,
             top_p=top_p,
             top_k=top_k,
+            min_p=min_p,
+            typical_p=typical_p,
             stream=stream,
             stop=["user:", "</s>"],
             max_tokens=max_tokens,
@@ -831,6 +841,8 @@ def functionary_chat_handler(
         temperature=temperature,
         top_p=top_p,
         top_k=top_k,
+        min_p=min_p,
+        typical_p=typical_p,
         presence_penalty=presence_penalty,
         frequency_penalty=frequency_penalty,
         repeat_penalty=repeat_penalty,
@@ -929,6 +941,8 @@ class Llava15ChatHandler:
         temperature: float = 0.2,
         top_p: float = 0.95,
         top_k: int = 40,
+        min_p: float = 0.05,
+        typical_p: float = 1.0,
         stream: bool = False,
         stop: Optional[Union[str, List[str]]] = [],
         response_format: Optional[
@@ -1045,6 +1059,8 @@ class Llava15ChatHandler:
                 temperature=temperature,
                 top_p=top_p,
                 top_k=top_k,
+                min_p=min_p,
+                typical_p=typical_p,
                 stream=stream,
                 stop=stop,
                 max_tokens=max_tokens,

llama_cpp/server/app.py

@@ -521,6 +521,14 @@ top_p_field = Field(
     + "Top-p sampling, also known as nucleus sampling, is another text generation method that selects the next token from a subset of tokens that together have a cumulative probability of at least p. This method provides a balance between diversity and quality by considering both the probabilities of tokens and the number of tokens to sample from. A higher value for top_p (e.g., 0.95) will lead to more diverse text, while a lower value (e.g., 0.5) will generate more focused and conservative text.",
 )
 
+min_p_field = Field(
+    default=0.05,
+    ge=0.0,
+    le=1.0,
+    description="Sets a minimum base probability threshold for token selection.\n\n"
+    + "The Min-P sampling method was designed as an alternative to Top-P, and aims to ensure a balance of quality and variety. The parameter min_p represents the minimum probability for a token to be considered, relative to the probability of the most likely token. For example, with min_p=0.05 and the most likely token having a probability of 0.9, logits with a value less than 0.045 are filtered out.",
+)
+
 stop_field = Field(
     default=None,
     description="A list of tokens at which to stop generation. If None, no stop tokens are used.",
@@ -593,6 +601,7 @@ class CreateCompletionRequest(BaseModel):
     max_tokens: int = max_tokens_field
     temperature: float = temperature_field
     top_p: float = top_p_field
+    min_p: float = min_p_field
     echo: bool = Field(
         default=False,
         description="Whether to echo the prompt in the generated text. Useful for chatbots.",
@@ -788,6 +797,7 @@ class CreateChatCompletionRequest(BaseModel):
     )
     temperature: float = temperature_field
     top_p: float = top_p_field
+    min_p: float = min_p_field
     stop: Optional[List[str]] = stop_field
     stream: bool = stream_field
     presence_penalty: Optional[float] = presence_penalty_field