]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commitdiff
sampling : do not set min_keep to n_probs (#5564)
authorGeorgi Gerganov <redacted>
Sun, 18 Feb 2024 17:38:06 +0000 (19:38 +0200)
committerGeorgi Gerganov <redacted>
Sun, 18 Feb 2024 17:38:06 +0000 (19:38 +0200)
common/sampling.cpp

index 53013138a9eb48e9c86156c653d2f45d34f0128c..611c327bb744745b8cc12ac3fd684d877e558969 100644 (file)
@@ -121,7 +121,7 @@ static void sampler_queue(
                    struct llama_context * ctx_main,
             const llama_sampling_params & params,
                  llama_token_data_array & cur_p,
-                                 size_t & min_keep) {
+                                 size_t   min_keep) {
     const float         temp              = params.temp;
     const float         dynatemp_range    = params.dynatemp_range;
     const float         dynatemp_exponent = params.dynatemp_exponent;
@@ -248,10 +248,7 @@ static llama_token llama_sampling_sample_impl(
             llama_sample_temp(ctx_main, &cur_p, temp);
             id = llama_sample_token_mirostat_v2(ctx_main, &cur_p, mirostat_tau, mirostat_eta, &ctx_sampling->mirostat_mu);
         } else {
-            // temperature sampling
-            size_t min_keep = std::max(1, params.n_probs);
-
-            sampler_queue(ctx_main, params, cur_p, min_keep);
+            sampler_queue(ctx_main, params, cur_p, 1);
 
             id = llama_sample_token(ctx_main, &cur_p);