Merge branch 'upstream' into concedo_experimental

# Conflicts:
#	.github/workflows/build.yml
#	CODEOWNERS
#	README.md
#	docs/ops.md
#	docs/ops/SYCL.csv
#	docs/ops/Vulkan.csv
#	ggml/CMakeLists.txt
#	ggml/src/CMakeLists.txt
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-sycl/backend.hpp
#	ggml/src/ggml-sycl/element_wise.cpp
#	ggml/src/ggml-sycl/element_wise.hpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	tests/test-backend-ops.cpp
#	tests/test-thread-safety.cpp

Author: LostRuins

convert_hf_to_gguf.py

@@ -895,8 +895,8 @@ def get_vocab_base_pre(self, tokenizer) -> str:
             # ref: https://huggingface.co/JetBrains/Mellum-4b-base
             res = "mellum"
         if chkhsh == "9b1be57e70d20d9501b2b3186e792d81181ae36ada3903c26f9fea418cf87206":
-            # ref: https://huggingface.co/inclusionAI/LLaDA-MoE-7B-A1B-Base
-            res = "llada-moe"
+            # ref: https://huggingface.co/inclusionAI/Ling-mini-base-2.0
+            res = "bailingmoe2"
         if chkhsh == "53e325976a6e142379c19b09afcae354f2f496f147afa8f9e189a33fe4e3024e":
             # ref: https://huggingface.co/ibm-granite/granite-docling-258M
             res = "granite-docling"
@@ -8060,6 +8060,103 @@ def prepare_tensors(self):
                 raise ValueError(f"Unprocessed experts: {experts}")
 
 
+@ModelBase.register("BailingMoeV2ForCausalLM")
+class BailingMoeV2Model(TextModel):
+    model_arch = gguf.MODEL_ARCH.BAILINGMOE2
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+        if nextn_layers := self.hparams.get("num_nextn_predict_layers", 0):
+            self.block_count = self.hparams["num_hidden_layers"] + nextn_layers
+            self.tensor_map = gguf.get_tensor_name_map(self.model_arch, self.block_count)
+
+    def set_vocab(self):
+        self._set_vocab_gpt2()
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        hparams = self.hparams
+        if (rope_dim := hparams.get("head_dim")) is None:
+            rope_dim = hparams["hidden_size"] // hparams["num_attention_heads"]
+
+        self.gguf_writer.add_rope_dimension_count(int(rope_dim * self.hparams.get("partial_rotary_factor", 0.5)))
+        rope_scaling = self.hparams.get("rope_scaling") or {}
+        if rope_scaling.get("rope_type", rope_scaling.get("type")) == "yarn" and "factor" in rope_scaling:
+            self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
+            self.gguf_writer.add_rope_scaling_factor(rope_scaling["factor"])
+            self.gguf_writer.add_rope_scaling_orig_ctx_len(rope_scaling["original_max_position_embeddings"])
+        else:
+            self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.NONE)
+        self.gguf_writer.add_leading_dense_block_count(hparams["first_k_dense_replace"])
+        self.gguf_writer.add_vocab_size(hparams["vocab_size"])
+        self.gguf_writer.add_expert_feed_forward_length(hparams["moe_intermediate_size"])
+        self.gguf_writer.add_expert_shared_feed_forward_length(hparams.get("moe_shared_expert_intermediate_size", hparams["moe_intermediate_size"] * hparams["num_shared_experts"]))
+        self.gguf_writer.add_expert_weights_scale(hparams["routed_scaling_factor"])
+        self.gguf_writer.add_expert_count(hparams["num_experts"])
+        self.gguf_writer.add_expert_shared_count(hparams["num_shared_experts"])
+        self.gguf_writer.add_expert_group_count(hparams["n_group"])
+        self.gguf_writer.add_expert_group_used_count(hparams["topk_group"])
+        self.gguf_writer.add_expert_weights_norm(hparams["norm_topk_prob"])
+
+        if hparams["score_function"] == "sigmoid":
+            self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SIGMOID)
+        elif hparams["score_function"] == "softmax":
+            self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SOFTMAX)
+        else:
+            raise ValueError(f"Unsupported score_function value: {hparams['score_function']}")
+
+        if (nextn_layers := self.hparams.get("num_nextn_predict_layers")) is not None:
+            self.gguf_writer.add_nextn_predict_layers(nextn_layers)
+
+    _experts: list[dict[str, Tensor]] | None = None
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        if "mlp.experts" in name:
+            n_experts = self.hparams["num_experts"]
+            assert bid is not None
+
+            tensors: list[tuple[str, Tensor]] = []
+
+            if self._experts is None:
+                self._experts = [{} for _ in range(self.block_count)]
+
+            self._experts[bid][name] = data_torch
+
+            if len(self._experts[bid]) >= n_experts * 3:
+                # merge the experts into a single 3d tensor
+                for w_name in ["down_proj", "gate_proj", "up_proj"]:
+                    datas: list[Tensor] = []
+
+                    for xid in range(n_experts):
+                        ename = f"model.layers.{bid}.mlp.experts.{xid}.{w_name}.weight"
+                        datas.append(self._experts[bid][ename])
+                        del self._experts[bid][ename]
+
+                    data_torch = torch.stack(datas, dim=0)
+
+                    merged_name = f"model.layers.{bid}.mlp.experts.{w_name}.weight"
+
+                    new_name = self.map_tensor_name(merged_name)
+
+                    tensors.append((new_name, data_torch))
+
+            return tensors
+
+        if name.endswith(".expert_bias"):
+            name = name.replace(".expert_bias", ".expert_bias.bias")
+
+        return [(self.map_tensor_name(name), data_torch)]
+
+    def prepare_tensors(self):
+        super().prepare_tensors()
+
+        if self._experts is not None:
+            # flatten `list[dict[str, Tensor]]` into `list[str]`
+            experts = [k for d in self._experts for k in d.keys()]
+            if len(experts) > 0:
+                raise ValueError(f"Unprocessed experts: {experts}")
+
+
 @ModelBase.register("GroveMoeForCausalLM", "modeling_grove_moe.GroveMoeForCausalLM")
 class GroveMoeModel(TextModel):
     model_arch = gguf.MODEL_ARCH.GROVEMOE

convert_hf_to_gguf_update.py

@@ -139,7 +139,7 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "lfm2",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LiquidAI/LFM2-Tokenizer"},
     {"name": "exaone4",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-4.0-32B", },
     {"name": "mellum",           "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/JetBrains/Mellum-4b-base", },
-    {"name": "llada-moe",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/inclusionAI/LLaDA-MoE-7B-A1B-Base", },
+    {"name": "bailingmoe2",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/inclusionAI/Ling-mini-base-2.0", },
     {"name": "granite-docling",  "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/ibm-granite/granite-docling-258M", },
 ]
 

docs/backend/zDNN.md

@@ -27427,7 +27427,8 @@ Current version indicated by LITEVER below.
 			</div>
 			<div id="oaicustom" class="menutext hidden">
 				<span id="oaidesc">
-				Entering your OpenAI API key will allow you to use KoboldAI Lite with their API.<br><br>
+				Entering your <a class="color_blueurl" href="https://platform.openai.com/api-keys">OpenAI API key</a> will allow you to use KoboldAI Lite with their API.<br>
+				<b>Or, set a custom API URL to use ANY 3rd-party OAI Compatible API</b><br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the OpenAI API, <span class="color_red">unless CORS proxy is used</span>. Reconnecting resets all custom OpenAI URLs.<br>Only Temperature, Top-P and Repetition Penalty samplers are used.<br><br>
 				<span class="color_green" style="font-weight: bold;">Please input OpenAI API URL and Key.</span><br>
 				<div style="display:inline-flex">
@@ -27438,17 +27439,17 @@ Current version indicated by LITEVER below.
 				</div>
 				</span>
 				<span id="openrouterdesc" class="hidden">
-				Entering your OpenRouter API key will allow you to use KoboldAI Lite with their API.<br><br>
+				Entering your <a class="color_blueurl" href="https://openrouter.ai/settings/keys">OpenRouter API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the OpenRouter API and is not transmitted to us.<br>Only Temperature, Top-P and Repetition Penalty samplers are used.<br><br>
 				<span class="color_green" style="font-weight: bold;">Please input OpenRouter Key.</span><br><br>
 				</span>
 				<span id="mistralaidesc" class="hidden">
-				Entering your MistralAI API key will allow you to use KoboldAI Lite with their API.<br><br>
+				Entering your <a class="color_blueurl" href="https://docs.mistral.ai/getting-started/quickstart">MistralAI API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the MistralAI API and is not transmitted to us.<br>Only Temperature and Top-P samplers are used.<br><br>
 				<span class="color_green" style="font-weight: bold;">Please input MistralAI Key.</span><br><br>
 				</span>
 				<span id="featherlessdesc" class="hidden">
-				Entering your Featherless API key will allow you to use KoboldAI Lite with their API.<br><br>
+				Entering your <a class="color_blueurl" href="https://featherless.ai/docs/getting-started">Featherless API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the Featherless API and is not transmitted to us.<br>Only Temperature, Top-P, Top-K, Min-P and Repetition Penalty samplers are used.<br><br>
 				<span class="color_green" style="font-weight: bold;">Please input Featherless Key.</span><br><br>
 				</span>
@@ -27463,8 +27464,8 @@ Current version indicated by LITEVER below.
 				<span class="color_green" style="font-weight: bold;">No Key Required.</span><br><br>
 				</span>
 				<span id="nvidianimdesc" class="hidden">
-				Entering your Nvidia NIM API key will allow you to use KoboldAI Lite with their API.<br><br>
-				KoboldAI Lite takes no responsibility for your usage of this feature. Due to CORS restrictions, your connection WILL be proxied.<br>Only Temperature, Top-P and Repetition Penalty samplers are used.<br><br>
+				Entering your <a class="color_blueurl" href="https://build.nvidia.com/explore/discover">Nvidia NIM API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
+				KoboldAI Lite takes no responsibility for your usage of this feature. <span class="color_red">Due to CORS restrictions, your connection WILL be proxied.</span><br>Only Temperature, Top-P and Repetition Penalty samplers are used.<br><br>
 				<span class="color_green" style="font-weight: bold;">Please input Nvidia NIM Key.</span><br><br>
 				</span>
 
@@ -27685,7 +27686,7 @@ Current version indicated by LITEVER below.
 
 			</div>
 			<div id="geminicustom" class="menutext hidden">
-				Uses Gemini by Google.<br><br>
+				Entering your <a class="color_blueurl" href="https://docs.mistral.ai/getting-started/quickstart">Google Gemini API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the Gemini API and is not transmitted to us.<br><br>
 				<div>
 				<select title="Gemini AI Model Selection" style="padding:4px; width:calc(100% - 110px); display:inline-block" class="form-control" id="custom_gemini_model" onchange="togglegeminimodel()">
@@ -27739,13 +27740,10 @@ Current version indicated by LITEVER below.
 				</div>
 			</div>
 			<div id="coherecustom" class="menutext hidden">
-				Uses Cohere's models through their own API.<br><br>
+				Entering your <a class="color_blueurl" href="https://dashboard.cohere.com/api-keys">Cohere API key</a> will allow you to use KoboldAI Lite with their API.<br><br>
 				KoboldAI Lite takes no responsibility for your usage of this feature. Your API key is used directly with the Cohere API and is not transmitted to us.<br><br>
 				<select title="Cohere AI Model Selection" style="padding:4px;" class="form-control" id="custom_cohere_model">
-					<option value="command" selected="selected">command</option>
-					<option value="command-r">command-r</option>
-					<option value="command-r-plus">command-r-plus</option>
-					<option value="command-r-08-2024">command-r-08-2024</option>
+					<option value="command-r-08-2024" selected="selected">command-r-08-2024</option>
 					<option value="command-r-plus-08-2024">command-r-plus-08-2024</option>
 					<option value="command-r7b-12-2024">command-r7b-12-2024</option>
 					<option value="command-a-03-2025">command-a-03-2025</option>

docs/build-riscv64-spacemit.md

@@ -0,0 +1,19 @@
+#pragma once
+
+#include "ggml.h"
+#include "ggml-backend.h"
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+// backend API
+GGML_BACKEND_API ggml_backend_t ggml_backend_hexagon_init(void);
+
+GGML_BACKEND_API bool ggml_backend_is_hexagon(ggml_backend_t backend);
+
+GGML_BACKEND_API ggml_backend_reg_t ggml_backend_hexagon_reg(void);
+
+#ifdef  __cplusplus
+}
+#endif