Add seqrep API docs since "himom" probably falls a bit short

Author: KerfuffleV2

llama.h

@@ -407,7 +407,11 @@ extern "C" {
     /// @details Frequency and presence penalties described in OpenAI API https://platform.openai.com/docs/api-reference/parameter-details.
     LLAMA_API void llama_sample_frequency_and_presence_penalties(struct llama_context * ctx, llama_token_data_array * candidates, const llama_token * last_tokens, size_t last_tokens_size, float alpha_frequency, float alpha_presence);
 
-    /// @details himom
+    /// @details Sequence repetition penalty with semi-fuzzy matching.
+    /// @params min_length The minimum length of a matching sequence of tokens.
+    /// @params tolerance Tolerance for non-matching tokens in a sequence.
+    /// @params flat_penalty Flat penalty applied to the token that can continue a repeated sequence.
+    /// @params stacking_penalty Scaling penalty applied to the token that can continue a repeated sequence. The penalty is multiplied by the total length of sequences that are continued by this token.
     LLAMA_API void llama_sample_seqrep_penalty(struct llama_context * ctx, llama_token_data_array * candidates, const llama_token * last_tokens_p, size_t last_tokens_size, size_t min_length, size_t tolerance, float flat_penalty, float length_penalty);
 
     /// @details Apply classifier-free guidance to the logits as described in academic paper "Stay on topic with Classifier-Free Guidance" https://arxiv.org/abs/2306.17806