model: add Hy3 (hy_v3) support with MTP speculative decoding (#25395)

* model: add Hy3 (hy_v3) architecture support

Adds Tencent Hunyuan 3 (HF architecture HYV3ForCausalLM, GGUF arch
hy_v3): a MoE decoder stack with per-head Q/K RMSNorm, a sigmoid
router with expert selection bias, an always-active ungated shared
expert, and leading dense block(s) (first_k_dense_replace).

The base implementation is ported from charlie12345's fork
(https://github.com/charlie12345/ROCmFPX, src/models/hyv3.cpp),
adapted to current mainline APIs (hparams.n_layer(), build_qkv,
build_moe_ffn with fused gate_up + scale tensors, output_s).

Note: blk.N.exp_probs_b is stored without a .bias suffix for
compatibility with existing hy_v3 GGUFs produced by that fork.

Co-Authored-By: charlie12345 <charlie12345@users.noreply.github.com>
Co-authored-by: Piotr Wilkin <ilintar@gmail.com>
Assisted-by: Claude Fable 5

common/jinja/value.cpp

@@ -750,11 +750,50 @@ const func_builtins & value_string_t::get_builtins() const {
             res->val_str.mark_input_based_on(args.get_pos(0)->val_str);
             return res;
         }},
+        {"format", [](const func_args & args) -> value {
+            value val_input = args.get_pos(0);
+            if (!is_val<value_string>(val_input)) {
+                throw raised_exception("format() first argument must be a string");
+            }
+            const jinja::string & fmt = val_input->as_string();
+            const bool fmt_is_input = fmt.all_parts_are_input();
+
+            const std::string str = fmt.str();
+            jinja::string result;
+            std::string literal;
+            auto flush_literal = [&]() {
+                if (!literal.empty()) {
+                    result.parts.push_back({fmt_is_input, literal});
+                    literal.clear();
+                }
+            };
+
+            size_t arg_idx = 1; // positional args follow the format string
+            for (size_t i = 0; i < str.size(); ++i) {
+                if (str[i] != '{') {
+                    literal += str[i];
+                    continue;
+                }
+                if (i + 1 >= str.size() || str[i + 1] != '}') {
+                    throw not_implemented_exception("format() only supports simple '{}' placeholders");
+                }
+                ++i;
+                flush_literal();
+                const jinja::string arg_str = args.get_pos(arg_idx++)->as_string();
+                result.parts.insert(result.parts.end(), arg_str.parts.begin(), arg_str.parts.end());
+            }
+            flush_literal();
+            return mk_val<value_string>(result);
+        }},
         {"int", [](const func_args & args) -> value {
             value val_input   = args.get_pos(0);
             value val_default = args.get_kwarg_or_pos("default", 1);
             value val_base    = args.get_kwarg_or_pos("base",    2);
             const int base = val_base->is_undefined() ? 10 : val_base->as_int();
+            if (base != 0 && (base < 2 || base > 36)) {
+                // an out-of-range base makes std::stoi fail fast on the MSVC CRT instead of throwing
+                throw raised_exception("int() base must be 0 or between 2 and 36");
+            }
             if (is_val<value_string>(val_input) == false) {
                 throw raised_exception("int() first argument must be a string");
             }