ikawrakow · ikawrakow · Jun 3, 2025 · Jun 3, 2025 · Jun 3, 2025 · Jun 3, 2025
diff --git a/common/common.cpp b/common/common.cpp
@@ -659,6 +659,11 @@ bool gpt_params_find_arg(int argc, char ** argv, const std::string & arg, gpt_pa
         sparams.xtc_threshold = std::stof(argv[i]);
         return true;
     }
+    if (arg == "--top-n-sigma") {
+        CHECK_ARG
+        sparams.top_n_sigma = std::stof(argv[i]);
+        return true;
+    }
     if (arg == "--cfg-negative-prompt") {
         CHECK_ARG
         sparams.cfg_negative_prompt = argv[i];
@@ -1646,7 +1651,8 @@ void gpt_params_print_usage(int /*argc*/, char ** argv, const gpt_params & param
     options.push_back({ "*",           "       --mirostat-lr N",        "Mirostat learning rate, parameter eta (default: %.1f)", (double)sparams.mirostat_eta });
     options.push_back({ "*",           "       --mirostat-ent N",       "Mirostat target entropy, parameter tau (default: %.1f)", (double)sparams.mirostat_tau });
     options.push_back({ "*",           "       --xtc-probability p",    "xtc probability (default: %.1f, 0.0 = disabled)", (double)sparams.xtc_probability });
-    options.push_back({ "*",           "       --xtc-threshold t",      "xtc threshold (default: %.1f, 0.0 = disabled)", (double)sparams.xtc_threshold});
+    options.push_back({ "*",           "       --xtc-threshold t",      "xtc threshold (default: %.1f, >0.5 = disabled)", (double)sparams.xtc_threshold});
+    options.push_back({ "*",           "       --top-n-sigma t",        "top-n-sigma parmeter (default: %.1f, 0.0 = disabled)", (double)sparams.top_n_sigma});
     options.push_back({ "*",           "       -l TOKEN_ID(+/-)BIAS",   "modifies the likelihood of token appearing in the completion,\n"
                                                                         "i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',\n"
                                                                         "or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'" });
@@ -3410,6 +3416,7 @@ void yaml_dump_non_result_info(FILE * stream, const gpt_params & params, const l
     fprintf(stream, "mirostat_lr: %f # default: 0.1\n", sparams.mirostat_eta);
     fprintf(stream, "xtc_probability: %f # default: 0.0\n", sparams.xtc_probability);
     fprintf(stream, "xtc_threshold: %f # default: 0.0\n", sparams.xtc_threshold);
+    fprintf(stream, "top_n_sigma: %f # default: 0.0\n", sparams.top_n_sigma);
     fprintf(stream, "mlock: %s # default: false\n", params.use_mlock ? "true" : "false");
     fprintf(stream, "model: %s # default: %s\n", params.model.c_str(), DEFAULT_MODEL_PATH);
     fprintf(stream, "model_draft: %s # default:\n", params.model_draft.c_str());

diff --git a/common/sampling.cpp b/common/sampling.cpp
@@ -122,11 +122,11 @@ std::string llama_sampling_print(const llama_sampling_params & params) {
             "\trepeat_last_n = %d, repeat_penalty = %.3f, frequency_penalty = %.3f, presence_penalty = %.3f\n"
             "\ttop_k = %d, tfs_z = %.3f, top_p = %.3f, min_p = %.3f, typical_p = %.3f, temp = %.3f\n"
             "\tmirostat = %d, mirostat_lr = %.3f, mirostat_ent = %.3f\n"
-            "\txtc_probability = %.3f, xtc_threshold = %.3f",
+            "\txtc_probability = %.3f, xtc_threshold = %.3f, top_n_sigma = %.3f",
             params.penalty_last_n, params.penalty_repeat, params.penalty_freq, params.penalty_present,
             params.top_k, params.tfs_z, params.top_p, params.min_p, params.typical_p, params.temp,
             params.mirostat, params.mirostat_eta, params.mirostat_tau,
-            params.xtc_probability, params.xtc_threshold);
+            params.xtc_probability, params.xtc_threshold, params.top_n_sigma);
 
     return std::string(result);
 }
@@ -156,6 +156,7 @@ std::string llama_sampling_type_to_str(llama_sampler_type sampler_type) {
         case llama_sampler_type::MIN_P:       return "min_p";
         case llama_sampler_type::TEMPERATURE: return "temperature";
         case llama_sampler_type::XTC        : return "xtc";
+        case llama_sampler_type::TOP_N_SIGMA: return "top_n_sigma";
         default : return "";
     }
 }
@@ -168,6 +169,7 @@ std::vector<llama_sampler_type> llama_sampling_types_from_names(const std::vecto
         {"min_p",       llama_sampler_type::MIN_P},
         {"tfs_z",       llama_sampler_type::TFS_Z},
         {"xtc",         llama_sampler_type::XTC},
+        {"top_n_sigma", llama_sampler_type::TOP_N_SIGMA},
         {"temperature", llama_sampler_type::TEMPERATURE}
     };
 
@@ -183,6 +185,7 @@ std::vector<llama_sampler_type> llama_sampling_types_from_names(const std::vecto
         {"tfs-z",       llama_sampler_type::TFS_Z},
         {"tfs",         llama_sampler_type::TFS_Z},
         {"xtc",         llama_sampler_type::XTC},
+        {"top-n-sigma", llama_sampler_type::TOP_N_SIGMA},
         {"temp",        llama_sampler_type::TEMPERATURE}
     };
 
@@ -218,6 +221,7 @@ std::vector<llama_sampler_type> llama_sampling_types_from_chars(const std::strin
         {'m', llama_sampler_type::MIN_P},
         {'f', llama_sampler_type::TFS_Z},
         {'x', llama_sampler_type::XTC},
+        {'n', llama_sampler_type::TOP_N_SIGMA},
         {'t', llama_sampler_type::TEMPERATURE}
     };
 
@@ -248,16 +252,18 @@ static void sampler_queue(
     const float         typical_p         = params.typical_p;
     const float         xtc_probability   = params.xtc_probability;
     const float         xtc_threshold     = params.xtc_threshold;
+    const float         top_n_sigma       = params.top_n_sigma;
     const std::vector<llama_sampler_type> & samplers_sequence = params.samplers_sequence;
 
     for (auto sampler_type : samplers_sequence) {
         switch (sampler_type) {
-            case llama_sampler_type::TOP_K    : llama_sample_top_k    (ctx_main, &cur_p, top_k,     min_keep); break;
-            case llama_sampler_type::TFS_Z    : llama_sample_tail_free(ctx_main, &cur_p, tfs_z,     min_keep); break;
-            case llama_sampler_type::TYPICAL_P: llama_sample_typical  (ctx_main, &cur_p, typical_p, min_keep); break;
-            case llama_sampler_type::TOP_P    : llama_sample_top_p    (ctx_main, &cur_p, top_p,     min_keep); break;
-            case llama_sampler_type::MIN_P    : llama_sample_min_p    (ctx_main, &cur_p, min_p,     min_keep); break;
-            case llama_sampler_type::XTC      : llama_sample_xtc      (ctx_main, &cur_p, xtc_probability, xtc_threshold, min_keep); break;
+            case llama_sampler_type::TOP_K      : llama_sample_top_k    (ctx_main, &cur_p, top_k,     min_keep); break;
+            case llama_sampler_type::TFS_Z      : llama_sample_tail_free(ctx_main, &cur_p, tfs_z,     min_keep); break;
+            case llama_sampler_type::TYPICAL_P  : llama_sample_typical  (ctx_main, &cur_p, typical_p, min_keep); break;
+            case llama_sampler_type::TOP_P      : llama_sample_top_p    (ctx_main, &cur_p, top_p,     min_keep); break;
+            case llama_sampler_type::MIN_P      : llama_sample_min_p    (ctx_main, &cur_p, min_p,     min_keep); break;
+            case llama_sampler_type::XTC        : llama_sample_xtc      (ctx_main, &cur_p, xtc_probability, xtc_threshold, min_keep); break;
+            case llama_sampler_type::TOP_N_SIGMA: llama_sample_top_n_sigma(ctx_main, &cur_p, top_n_sigma); break;
             case llama_sampler_type::TEMPERATURE:
                 if (dynatemp_range > 0) {
                     float dynatemp_min = std::max(0.0f, temp - dynatemp_range);

diff --git a/common/sampling.h b/common/sampling.h
@@ -16,6 +16,7 @@ enum class llama_sampler_type : char {
     MIN_P       = 'm',
     TFS_Z       = 'f',
     XTC         = 'x',
+    TOP_N_SIGMA = 'n',
     TYPICAL_P   = 'y',
     TEMPERATURE = 't'
 };
@@ -41,7 +42,8 @@ typedef struct llama_sampling_params {
     float       mirostat_tau          = 5.00f;              // target entropy
     float       mirostat_eta          = 0.10f;              // learning rate
     float       xtc_probability       = 0.0f;               // xtc probability
-    float       xtc_threshold         = 1.0f;               // xtc threashold, disabled if > 0.5
+    float       xtc_threshold         = 1.0f;               // xtc threshold, disabled if > 0.5
+    float       top_n_sigma           = 0.0f;               // top-n-sigma
     bool        penalize_nl           = false;              // consider newlines as a repeatable token
     uint32_t    seed                  = LLAMA_DEFAULT_SEED; // the seed used to initialize llama_sampling_context
 

diff --git a/examples/main/README.md b/examples/main/README.md
@@ -239,6 +239,17 @@ The `--mirostat-ent` option sets the Mirostat target entropy (tau), which repres
 
 Example usage: `--mirostat 2 --mirostat-lr 0.05 --mirostat-ent 3.0`
 
+### XTC Sampling
+
+-   --xtc-probability p: xtc probability (default: 0.0 => disabled)
+-   --xtc-threshold t  : xtc threshold   (default: 1.0 => disabled)
+
+### Top-n-sigma Sampling
+
+Sets all logits $L_i$ to $-\infty$ where $L_i < L_{\rm max} - n \sigma$. Here $L_{\rm max}$ is the maximum logit, $\sigma$ is the logit standard deviation, and $n$ is the top-n-sigma parameter.
+
+-   --top-n-sigma t          top-n-sigma parmeter (default: 0.0 => disabled)
+
 ### Logit Bias
 
 -   `-l TOKEN_ID(+/-)BIAS, --logit-bias TOKEN_ID(+/-)BIAS`: Modify the likelihood of a token appearing in the generated text completion.

diff --git a/examples/server/README.md b/examples/server/README.md
@@ -98,6 +98,9 @@ sampling:
                                   (default: 0, 0 = disabled, 1 = Mirostat, 2 = Mirostat 2.0)
          --mirostat-lr N          Mirostat learning rate, parameter eta (default: 0.1)
          --mirostat-ent N         Mirostat target entropy, parameter tau (default: 5.0)
+         --xtc-probability p      xtc probability (default: 0.0 => disabled)
+         --xtc-threshold t        xtc threshold (default: 1.0 => disabled)
+         --top-n-sigma t          top-n-sigma parmeter (default: 0.0 => disabled)
          -l TOKEN_ID(+/-)BIAS     modifies the likelihood of token appearing in the completion,
                                   i.e. `--logit-bias 15043+1` to increase likelihood of token ' Hello',
                                   or `--logit-bias 15043-1` to decrease likelihood of token ' Hello'

diff --git a/include/llama.h b/include/llama.h
@@ -1216,6 +1216,13 @@ extern "C" {
                            float   threshold,
                            size_t  min_keep);
 
+    /// @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 void llama_sample_top_n_sigma(
+            struct llama_context * ctx,
+          llama_token_data_array * candidates_p,
+                           float   top_n_sigma);
+
+
     /// @details Mirostat 1.0 algorithm described in the paper https://arxiv.org/abs/2007.14966. Uses tokens instead of words.
     /// @param candidates A vector of `llama_token_data` containing the candidate tokens, their probabilities (p), and log-odds (logit) for the current position in the generated text.
     /// @param 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.

diff --git a/src/llama-sampling.cpp b/src/llama-sampling.cpp
@@ -435,7 +435,7 @@ void llama_sample_temp_impl(struct llama_sampling * smpl, llama_token_data_array
 }
 
 void llama_sample_xtc_impl(struct llama_sampling * smpl, llama_token_data_array * candidates, float probability, float threshold, size_t min_keep) {
-    if (probability < 0 || threshold > 0.5f || candidates->size < 2) {
+    if (probability <= 0 || threshold > 0.5f || candidates->size < 2) {
         return;
     }
     GGML_ASSERT(smpl);
@@ -468,6 +468,64 @@ void llama_sample_xtc_impl(struct llama_sampling * smpl, llama_token_data_array
 
 }
 
+void llama_sample_top_n_sigma_impl(struct llama_sampling * smpl, llama_token_data_array * candidates, float top_n_sigma) {
+
+    if (top_n_sigma <= 0.0f || candidates->size < 4) {
+        // top_n_sigma <= 0: disabled
+        // candidates->size < 4: no point in applying the transformation for fewer than 4 logits.
+        return;
+    }
+
+    const int64_t t_start_sample_us = ggml_time_us();
+
+    float max = candidates->data[0].logit;
+    float mean = 0;
+    size_t count = 0;
+    for (int i = 0; i < (int)candidates->size; ++i) {
+        // Only count non-negative infinity values
+        if (candidates->data[i].logit != -INFINITY) {
+            max = std::max(max, candidates->data[i].logit);
+            mean += candidates->data[i].logit;
+            ++count;
+        }
+    }
+    if (count < 4) {
+        return; // again, tandard deviation is not well defined for so few logits (4 is actually pushing it)
+    }
+    mean /= count;
+
+    float sigma2 = 0;
+    for (int i = 0; i < (int)candidates->size; ++i) {
+        if (candidates->data[i].logit != -INFINITY) {
+            float delta = candidates->data[i].logit - mean;
+            sigma2 += delta*delta;
+        }
+    }
+    float sigma = sqrtf(sigma2/count);
+    float thresh = max - top_n_sigma*sigma;
+
+    int n_masked = 0;
+    for (int i = 0; i < (int)candidates->size; ++i) {
+        if (candidates->data[i].logit != -INFINITY && candidates->data[i].logit < thresh) {
+            candidates->data[i].logit = -INFINITY;
+            ++n_masked;
+        }
+    }
+
+    // do we really want to compute softmax unconditionally?
+    // The following coresponds to mainline implementation with the minor optimization
+    // that we only call the relativly expensive softmax if we masked away some tokens.
+    if (n_masked > 0 || !candidates->sorted) {
+        llama_sample_softmax_impl(nullptr, candidates);
+    }
+
+    if (smpl) {
+        smpl->t_sample_us += ggml_time_us() - t_start_sample_us;
+        smpl->n_sample++;
+    }
+}
+
+
 void llama_sample_repetition_penalties_impl(
         struct llama_sampling * smpl,
        llama_token_data_array * candidates,

diff --git a/src/llama-sampling.h b/src/llama-sampling.h
@@ -33,6 +33,7 @@ void llama_sample_typical_impl  (struct llama_sampling * smpl, llama_token_data_
 void llama_sample_entropy_impl  (struct llama_sampling * smpl, llama_token_data_array * candidates, float min_temp, float max_temp, float exponent_val);
 void llama_sample_temp_impl     (struct llama_sampling * smpl, llama_token_data_array * candidates, float temp);
 void llama_sample_xtc_impl      (struct llama_sampling * smpl, llama_token_data_array * candidates, float probability, float threshold, size_t min_keep);
+void llama_sample_top_n_sigma_impl(struct llama_sampling * smpl, llama_token_data_array * candidates, float top_n_sigma);
 
 void llama_sample_repetition_penalties_impl(
         struct llama_sampling * smpl,

diff --git a/src/llama.cpp b/src/llama.cpp
@@ -23270,6 +23270,10 @@ void llama_sample_xtc(struct llama_context * ctx, llama_token_data_array * candi
     llama_sample_xtc_impl(ctx ? &ctx->sampling : nullptr, candidates_p, probability, threshold, min_keep);
 }
 
+void llama_sample_top_n_sigma(struct llama_context * ctx, llama_token_data_array * candidates_p, float top_n_sigma) {
+    llama_sample_top_n_sigma_impl(ctx ? &ctx->sampling : nullptr, candidates_p, top_n_sigma);
+}
+
 void llama_sample_repetition_penalties(
             struct llama_context * ctx,
           llama_token_data_array * candidates,