sampling : add support for backend sampling

common/arg.cpp

@@ -1513,6 +1513,13 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.sampling.grammar = json_schema_to_grammar(json::parse(schema));
         }
     ).set_sparam());
+    add_opt(common_arg(
+        {"--backend-sampling"},
+        "enable backend sampling (default: disabled)",
+        [](common_params & params) {
+            params.sampling.backend_sampling = true;
+        }
+    ).set_sparam());
     add_opt(common_arg(
         {"--pooling"}, "{none,mean,cls,last,rank}",
         "pooling type for embeddings, use model default if unspecified",

common/common.cpp

@@ -1019,6 +1019,21 @@ struct common_init_result common_init_from_params(common_params & params) {
 
     auto cparams = common_context_params_to_llama(params);
 
+    // backend sampling initialization
+    if (params.sampling.backend_sampling) {
+        llama_sampler * backend_chain = common_sampler_backend_init(model, params.sampling);
+        if (backend_chain != nullptr) {
+            iparams.samplers_seq_config.resize(cparams.n_seq_max);
+            for (int i = 0; i < (int) cparams.n_seq_max; ++i) {
+                iparams.samplers_seq_config[i] = { i, llama_sampler_clone(backend_chain) };
+            }
+            cparams.samplers   = iparams.samplers_seq_config.data();
+            cparams.n_samplers = cparams.n_seq_max;
+
+            llama_sampler_free(backend_chain);
+        }
+    }
+
     llama_context * lctx = llama_init_from_model(model, cparams);
     if (lctx == NULL) {
         LOG_ERR("%s: failed to create context with model '%s', try reducing --n-gpu-layers if you're running out of VRAM\n",

common/common.h

@@ -213,6 +213,8 @@ struct common_params_sampling {
     std::vector<llama_logit_bias> logit_bias;     // logit biases to apply
     std::vector<llama_logit_bias> logit_bias_eog; // pre-calculated logit biases for EOG tokens
 
+    bool backend_sampling = false; // enable backend sampling
+
     // print the parameters into a string
     std::string print() const;
 };
@@ -654,6 +656,9 @@ struct common_init_result {
     llama_context_ptr context;
 
     std::vector<llama_adapter_lora_ptr> lora;
+
+    std::vector<llama_sampler_ptr> samplers;
+    std::vector<llama_sampler_seq_config> samplers_seq_config;
 };
 
 struct common_init_result     common_init_from_params(common_params & params);

common/llguidance.cpp

@@ -106,12 +106,16 @@ static void llama_sampler_llg_free(llama_sampler * smpl) {
 }
 
 static llama_sampler_i llama_sampler_llg_i = {
-    /* .name   = */ llama_sampler_llg_name,
-    /* .accept = */ llama_sampler_llg_accept_impl,
-    /* .apply  = */ llama_sampler_llg_apply,
-    /* .reset  = */ llama_sampler_llg_reset,
-    /* .clone  = */ llama_sampler_llg_clone,
-    /* .free   = */ llama_sampler_llg_free,
+    /* .name                = */ llama_sampler_llg_name,
+    /* .accept              = */ llama_sampler_llg_accept_impl,
+    /* .apply               = */ llama_sampler_llg_apply,
+    /* .reset               = */ llama_sampler_llg_reset,
+    /* .clone               = */ llama_sampler_llg_clone,
+    /* .free                = */ llama_sampler_llg_free,
+    /* .apply_ggml          = */ NULL,
+    /* .accept_ggml         = */ NULL,
+    /* .set_input_ggml      = */ NULL,
+    /* .set_backend_context = */ NULL,
 };
 
 static size_t llama_sampler_llg_tokenize_fn(const void * user_data, const uint8_t * bytes, size_t bytes_len,