models: Flint (FlintForCausalLM) on ra/main

conversion/__init__.py

@@ -19,6 +19,7 @@
 TEXT_MODEL_MAP: dict[str, str] = {
     "AfmoeForCausalLM": "afmoe",
     "LagunaForCausalLM": "laguna",
+    "FlintForCausalLM": "flint",
     "ApertusForCausalLM": "llama",
     "ArceeForCausalLM": "llama",
     "ArcticForCausalLM": "arctic",

conversion/base.py

@@ -254,7 +254,7 @@
                     if weight_map is None or not isinstance(weight_map, dict):
                         raise ValueError(f"Can't load 'weight_map' from {index_name!r}")
                     tensor_names_from_index.update(weight_map.keys())
                     part_dict: dict[str, None] = dict.fromkeys(weight_map.values(), None) # ty: ignore[invalid-assignment]
                     part_names = sorted(part_dict.keys())
             else:
                 weight_map = {}
@@ -1529,7 +1529,7 @@
             self.gguf_writer.add_expert_group_used_count(n_group_used)
             logger.info(f"gguf: expert groups used count = {n_group_used}")
 
-        if (score_func := self.find_hparam(["score_function", "scoring_func", "score_func", "moe_router_activation", "moe_router_activation_func", "expert_selection_fn", "router_score_func"], optional=True)) is not None:
+        if (score_func := self.find_hparam(["score_function", "scoring_func", "score_func", "moe_router_activation", "moe_router_activation_func", "expert_selection_fn", "router_score_func", "router_score_function"], optional=True)) is not None:
             if score_func == "sigmoid":
                 self.gguf_writer.add_expert_gating_func(gguf.ExpertGatingFuncType.SIGMOID)
             elif score_func == "softmax":
@@ -1586,15 +1586,15 @@
 
         from transformers import AutoTokenizer
         tokenizer = AutoTokenizer.from_pretrained(self.dir_model)
         vocab_size = self.hparams.get("vocab_size", len(tokenizer.vocab))  # ty: ignore[unresolved-attribute]
         assert max(tokenizer.vocab.values()) < vocab_size  # ty: ignore[unresolved-attribute]
 
         tokpre = self.get_vocab_base_pre(tokenizer)
 
         reverse_vocab = {id_: encoded_tok for encoded_tok, id_ in tokenizer.vocab.items()}  # ty: ignore[unresolved-attribute]
         added_vocab = tokenizer.get_added_vocab()  # ty: ignore[unresolved-attribute]
 
         added_tokens_decoder = tokenizer.added_tokens_decoder  # ty: ignore[unresolved-attribute]
 
         for i in range(vocab_size):
             if i not in reverse_vocab:
@@ -1607,7 +1607,7 @@
                     # To avoid unexpected issues - we make sure to normalize non-normalized tokens
                     if not added_tokens_decoder[i].normalized:
                         previous_token = token
                         token = tokenizer.decode(tokenizer.encode(token, add_special_tokens=False))  # ty: ignore[unresolved-attribute, invalid-assignment]
                         if previous_token != token:
                             logger.info(f"{repr(previous_token)} is encoded and decoded back to {repr(token)} using AutoTokenizer")
 
@@ -1695,6 +1695,9 @@
         if chkhsh == "1444df51289cfa8063b96f0e62b1125440111bc79a52003ea14b6eac7016fd5f":
             # ref: https://huggingface.co/openbmb/MiniCPM-V-4_6
             res = "qwen35"
+        if chkhsh == "9ff8c72e7519833cf13eb20da2d87641873339e004dddddde384920fba6431b5":
+            # ref: https://huggingface.co/fastino/GLiNER2.5-Decide
+            res = "default"
         if chkhsh == "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273":
             # ref: https://huggingface.co/alvarobartt/grok-2-tokenizer
             res = "grok-2"
@@ -1737,11 +1740,6 @@
         if chkhsh == "8e62295832751ca1e8f92f2226f403dea30dc5165e448b5bfa05af5340c64ec7":
             # ref: https://huggingface.co/BAAI/bge-large-zh-v1.5
             res = "bert-bge-large"
-        if chkhsh == "9ff8c72e7519833cf13eb20da2d87641873339e004dddddde384920fba6431b5":
-            # ref: DeBERTa-v3 / GLiNER2.x -- Metaspace pre-tokenizer with
-            # prepend_scheme "always" is llama.cpp's "default" SPM behaviour.
-            # Unigram model, so the writer also needs token scores.
-            res = "default"
         if chkhsh == "b6dc8df998e1cfbdc4eac8243701a65afe638679230920b50d6f17d81c098166":
             # ref: https://huggingface.co/mosaicml/mpt-7b
             res = "mpt"
@@ -1766,6 +1764,9 @@
         if chkhsh == "52df12b4c8d4176e7481aab4b6e8454d1fd0a210a04a574f6d4e067d10e23c3e":
             # ref: https://huggingface.co/CohereLabs/North-Mini-Code-1.0
             res = "cohere2moe"
+        if chkhsh == "d5420e601e979078c670bc503d9ac411c311fa76b76a76267c31e0d6ba816637":
+            # ref: https://huggingface.co/RecursiveSI/flint-preview
+            res = "flint"
         if chkhsh == "e636dc30a262dcc0d8c323492e32ae2b70728f4df7dfe9737d9f920a282b8aea":
             # ref: https://huggingface.co/Qwen/Qwen1.5-7B
             res = "qwen2"
@@ -2000,10 +2001,10 @@
     def _set_vocab_hybriddna(self):
         from transformers import AutoTokenizer
         tokenizer = AutoTokenizer.from_pretrained(self.dir_model, trust_remote_code=True)
         vocab_size = self.hparams.get("vocab_size", len(tokenizer.vocab))  # ty: ignore[unresolved-attribute]
         assert max(tokenizer.vocab.values()) < vocab_size  # ty: ignore[unresolved-attribute]
 
         reverse_vocab = {id_: encoded_tok for encoded_tok, id_ in tokenizer.vocab.items()}  # ty: ignore[unresolved-attribute]
         # k-mers can share text with a base-vocab BPE token (e.g. CCCCCC) and get
         # dropped by get_vocab(); a reserved marker suffix (U+E000) keeps each
         # k-mer's own id (llama.cpp strips it on detokenization)

conversion/flint.py

@@ -0,0 +1,38 @@
+from __future__ import annotations
+
+from .base import ModelBase, gguf
+from .qwen import Qwen2MoeModel
+
+
+@ModelBase.register("FlintForCausalLM")
+@ModelBase.example("RecursiveSI/flint-preview")
+class FlintModel(Qwen2MoeModel):
+    model_arch = gguf.MODEL_ARCH.FLINT
+
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        dense = self.hparams["mlp_only_layers"]
+        if dense != list(range(len(dense))):
+            raise ValueError("Flint dense layers must precede the MoE layers")
+        self.gguf_writer.add_leading_dense_block_count(len(dense))
+        self.gguf_writer.add_expert_shared_count(
+            self.hparams["shared_expert_intermediate_size"] // self.hparams["moe_intermediate_size"]
+        )
+        self.gguf_writer.add_expert_weights_norm(True)
+        self.gguf_writer.add_expert_weights_scale(1.0)
+        self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
+        self.gguf_writer.add_sliding_window_pattern([
+            kind == "sliding_attention" for kind in self.hparams["layer_types"]
+        ])
+        self.gguf_writer.add_rope_pattern([
+            i not in self.hparams["no_rope_layer_indices"] for i in range(self.block_count)
+        ])
+
+    def modify_tensors(self, data_torch, name, bid):
+        if name.endswith(".softmax_offset"):
+            name += ".weight"
+        if name.endswith(".gate.expert_bias"):
+            if not self.hparams["router_expert_bias"]:
+                return
+            name += ".bias"
+        yield from super().modify_tensors(data_torch, name, bid)

convert_hf_to_gguf_update.py

@@ -101,6 +101,7 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "command-r",        "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereForAI/c4ai-command-r-v01", },
     {"name": "tiny_aya",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereLabs/tiny-aya-base", },
     {"name": "cohere2moe",       "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/CohereLabs/North-Mini-Code-1.0", },
+    {"name": "flint",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/RecursiveSI/flint-preview", },
     {"name": "qwen2",            "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen1.5-7B", },
     {"name": "olmo",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/allenai/OLMo-1.7-7B-hf", },
     {"name": "dbrx",             "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/databricks/dbrx-base", },
@@ -188,6 +189,10 @@ class TOKENIZER_TYPE(IntEnum):
     {"name": "kimi-k2",   "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/moonshotai/Kimi-K2-Base",   "chkhsh": "81212dc7cdb7e0c1074ca62c5aeab0d43c9f52b8a737be7b12a777c953027890"},
     {"name": "qwen2",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/Qwen/Qwen3-Embedding-0.6B", "chkhsh": "d4540891389ea895b53b399da6ac824becc30f2fba0e9ddbb98f92e55ca0e97c"},
     {"name": "qwen35",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/openbmb/MiniCPM-V-4_6", "chkhsh": "1444df51289cfa8063b96f0e62b1125440111bc79a52003ea14b6eac7016fd5f"},
+    # DeBERTa-v3 / GLiNER2.5 is a Unigram tokenizer whose Metaspace pre-tokenizer (prepend_scheme "always")
+    # matches llama.cpp's "default". Listed as BPE only because the generator drops SPM/UGM entries from the
+    # hash table; with a pre-computed chkhsh the type is never used to load anything.
+    {"name": "default", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/fastino/GLiNER2.5-Decide", "chkhsh": "9ff8c72e7519833cf13eb20da2d87641873339e004dddddde384920fba6431b5"},
     {"name": "grok-2",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/alvarobartt/grok-2-tokenizer", "chkhsh": "66b8d4e19ab16c3bfd89bce5d785fb7e0155e8648708a1f42077cb9fe002c273"},
     # jina-v2-de variants
     {"name": "jina-v2-de", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/aari1995/German_Semantic_V3", "chkhsh": "b3d1dd861f1d4c5c0d2569ce36baf3f90fe8a102db3de50dd71ff860d91be3df"},

ggml/src/ggml-backend-meta.cpp

@@ -774,6 +774,13 @@ static struct ggml_backend_meta_split_state ggml_backend_meta_get_split_state(
         return src_ss[0];
     };
 
+    auto handle_pool_1d = [&](const std::vector<ggml_backend_meta_split_state> & src_ss) -> ggml_backend_meta_split_state {
+        if (src_ss[0].axis == GGML_BACKEND_SPLIT_AXIS_0) {
+            return {GGML_BACKEND_SPLIT_AXIS_MIRRORED, {0}, {1}, 1};
+        }
+        return src_ss[0];
+    };
+
     auto handle_flash_attn_ext = [&](const std::vector<ggml_backend_meta_split_state> & src_ss) -> ggml_backend_meta_split_state {
         GGML_ASSERT(tensor->src[3] == nullptr || src_ss[3].axis == GGML_BACKEND_SPLIT_AXIS_MIRRORED);
 
@@ -990,12 +997,14 @@ static struct ggml_backend_meta_split_state ggml_backend_meta_get_split_state(
             case GGML_OP_CONV_3D:
             case GGML_OP_CONV_2D_DW:
             case GGML_OP_CONV_TRANSPOSE_2D:
-            case GGML_OP_POOL_1D:
             case GGML_OP_POOL_2D:
             case GGML_OP_POOL_2D_BACK:
             case GGML_OP_UPSCALE: {
                 split_state = handle_generic(src_ss, /*scalar_only =*/ true);
             } break;
+            case GGML_OP_POOL_1D: {
+                split_state = handle_pool_1d(src_ss);
+            } break;
             case GGML_OP_PAD: {
                 split_state = handle_pad(src_ss);
             } break;

gguf-py/gguf/constants.py

@@ -543,6 +543,7 @@ class MODEL_ARCH(IntEnum):
     QWEN2VL          = auto()
     QWEN3            = auto()
     QWEN3MOE         = auto()
+    FLINT            = auto()
     QWEN3NEXT        = auto()
     HRM_TEXT         = auto()
     QWEN3VL          = auto()
@@ -1328,6 +1329,7 @@ class MODEL_TENSOR(IntEnum):
     MODEL_ARCH.QWEN2MOE:         "qwen2moe",
     MODEL_ARCH.QWEN2VL:          "qwen2vl",
     MODEL_ARCH.QWEN3:            "qwen3",
+    MODEL_ARCH.FLINT:            "flint",
     MODEL_ARCH.QWEN3MOE:         "qwen3moe",
     MODEL_ARCH.QWEN3NEXT:        "qwen3next",
     MODEL_ARCH.HRM_TEXT:         "hrm_text",
@@ -2853,6 +2855,33 @@ class MODEL_TENSOR(IntEnum):
         MODEL_TENSOR.FFN_DOWN_EXP,
         MODEL_TENSOR.FFN_UP_EXP,
     ],
+
+    MODEL_ARCH.FLINT: [
+        MODEL_TENSOR.TOKEN_EMBD,
+        MODEL_TENSOR.OUTPUT_NORM,
+        MODEL_TENSOR.OUTPUT,
+        MODEL_TENSOR.ATTN_NORM,
+        MODEL_TENSOR.ATTN_QKV,
+        MODEL_TENSOR.ATTN_Q,
+        MODEL_TENSOR.ATTN_Q_NORM,
+        MODEL_TENSOR.ATTN_K,
+        MODEL_TENSOR.ATTN_K_NORM,
+        MODEL_TENSOR.ATTN_V,
+        MODEL_TENSOR.ATTN_OUT,
+        MODEL_TENSOR.FFN_NORM,
+        MODEL_TENSOR.FFN_GATE_INP,
+        MODEL_TENSOR.FFN_GATE_EXP,
+        MODEL_TENSOR.FFN_DOWN_EXP,
+        MODEL_TENSOR.FFN_UP_EXP,
+        MODEL_TENSOR.ATTN_SINKS,
+        MODEL_TENSOR.FFN_GATE,
+        MODEL_TENSOR.FFN_UP,
+        MODEL_TENSOR.FFN_DOWN,
+        MODEL_TENSOR.FFN_GATE_SHEXP,
+        MODEL_TENSOR.FFN_UP_SHEXP,
+        MODEL_TENSOR.FFN_DOWN_SHEXP,
+        MODEL_TENSOR.FFN_EXP_PROBS_B,
+    ],
     MODEL_ARCH.QWEN3NEXT: [
         MODEL_TENSOR.TOKEN_EMBD,
         MODEL_TENSOR.OUTPUT_NORM,

gguf-py/gguf/tensor_mapping.py

@@ -394,6 +394,7 @@ class TensorNameMap:
         ),
 
         MODEL_TENSOR.ATTN_SINKS: (
+            "model.layers.{bid}.self_attn.softmax_offset", # flint
             "model.layers.{bid}.self_attn.sinks", # openai-moe
             "model.layers.{bid}.self_attn.attention_sink_bias", # mimov2
             "model.layers.{bid}.self_attn.learnable_sink_param", # hy-v4

src/llama-arch.cpp

@@ -36,6 +36,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
     { LLM_ARCH_QWEN2MOE,         "qwen2moe"         },
     { LLM_ARCH_QWEN2VL,          "qwen2vl"          },
     { LLM_ARCH_QWEN3,            "qwen3"            },
+    { LLM_ARCH_FLINT,            "flint"            },
     { LLM_ARCH_QWEN3MOE,         "qwen3moe"         },
     { LLM_ARCH_QWEN3NEXT,        "qwen3next"        },
     { LLM_ARCH_QWEN3VL,          "qwen3vl"          },

src/llama-arch.h

@@ -42,6 +42,7 @@ enum llm_arch {
     LLM_ARCH_QWEN2VL,
     LLM_ARCH_QWEN3,
     LLM_ARCH_QWEN3MOE,
+    LLM_ARCH_FLINT,
     LLM_ARCH_QWEN3NEXT,
     LLM_ARCH_QWEN3VL,
     LLM_ARCH_QWEN3VLMOE,

src/llama-graph.cpp

@@ -1585,6 +1585,43 @@ ggml_tensor * llm_graph_context::build_lora_mm(
     return res;
 }
 
+llm_graph_amax_src1 llm_graph_context::build_amax_src1_f16(
+          ggml_tensor * cur) const {
+    llm_graph_amax_src1 result = {
+        /*.res     =*/ cur,
+        /*.scale_s =*/ nullptr,
+    };
+
+    const int64_t n_embd = cur->ne[0];
+
+    // amax = max|cur| along the rows for each [expert, token] -> [1, n_expert, n_tokens]
+    ggml_tensor * ab = ggml_abs(ctx0, cur);
+    ggml_tensor * amax = ggml_pool_1d(ctx0, ab, GGML_OP_POOL_MAX, n_embd, n_embd, 0);
+
+    // ref: https://github.com/ggml-org/llama.cpp/pull/26223
+    constexpr float k_log2_e = 1.4426950408889634f; // 1 / ln(2)
+    constexpr float k_ln_2   = 0.6931471805599453f;
+
+    ggml_tensor * e;
+    // note: this can technically be:
+    // e = ggml_scale(ctx0, amax, 1.0f/32768.0f);
+    // but we leave x32 headroom for the F16 sums of 32 values in Q8 activation blocks
+    e = ggml_scale(ctx0, amax, 1.0f/1024.0f);
+    e = ggml_log  (ctx0, e);
+    e = ggml_scale(ctx0, e, k_log2_e);
+    e = ggml_ceil (ctx0, e);
+    e = ggml_clamp(ctx0, e, 0.0f, 127.0f);
+    e = ggml_scale(ctx0, e, k_ln_2);
+    e = ggml_exp  (ctx0, e);
+
+    result = {
+        /*.res     =*/ ggml_div(ctx0, cur, e),
+        /*.scale_s =*/ e,
+    };
+
+    return result;
+}
+
 ggml_tensor * llm_graph_context::build_lora_mm_id(
           ggml_tensor * w,   // ggml_tensor * as
           ggml_tensor * cur, // ggml_tensor * b
@@ -2359,12 +2396,25 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
             GGML_ABORT("fatal error");
     }
 
-    experts = build_lora_mm_id(down_exps, cur, selected_experts, down_exps_s, slots); // [n_embd, n_expert_used, n_tokens]
-    if (arch == LLM_ARCH_MISTRAL4) {
-        // src1 can exceed F16 range
-        ggml_prec_set_src(experts, GGML_PREC_F32, 1);
+    {
+        // TODO: this graph-level rescaling should only be applied when the corresponding backend does not support
+        //       ggml_prec_set_src(experts, GGML_PREC_BF16, 1);
+        //       ref: https://github.com/ggml-org/llama.cpp/pull/30656#issuecomment-6269597922
+        llm_graph_amax_src1 amax_src1 = { nullptr, nullptr };
+
+        if (arch == LLM_ARCH_MISTRAL4 || arch == LLM_ARCH_MELLUM || arch == LLM_ARCH_FLINT) {
+            amax_src1 = build_amax_src1_f16(cur);
+            cur = amax_src1.res;
+        }
+
+        experts = build_lora_mm_id(down_exps, cur, selected_experts, down_exps_s, slots); // [n_embd, n_expert_used, n_tokens]
+        //ggml_prec_set_src(experts, GGML_PREC_BF16, 1); // (see TODO above)
+        cb(experts, "ffn_moe_down", il);
+
+        if (amax_src1.scale_s) {
+            experts = ggml_mul(ctx0, experts, amax_src1.scale_s);
+        }
     }
-    cb(experts, "ffn_moe_down", il);
 
     if (down_exps_s) {
         cb(experts, "ffn_moe_down_scaled", il);

src/llama-graph.h

@@ -1013,6 +1013,11 @@ struct llm_graph_qkv {
     ggml_tensor * v; // [n_embd_head, n_head_kv, n_tokens]
 };
 
+struct llm_graph_amax_src1 {
+    ggml_tensor * res;     // cur rescaled
+    ggml_tensor * scale_s; // scale to undo on the matmul result, or nullptr
+};
+
 struct llm_graph_context {
     const llm_arch arch;
 
@@ -1109,6 +1114,11 @@ struct llm_graph_context {
               ggml_tensor * w_s   = nullptr,
               ggml_tensor * slots = nullptr) const;
 
+    // rescale `cur` (the src1 of a matmul) by a power of two, computed from its maximum absolute
+    // keep both individual values and Q8 activation block sums within F16 range
+    llm_graph_amax_src1 build_amax_src1_f16(
+              ggml_tensor * cur) const;
+
     ggml_tensor * build_norm(
              ggml_tensor * cur,
              ggml_tensor * mw,

src/llama-kv-cache.cpp

@@ -880,7 +880,10 @@ bool llama_kv_cache::update(llama_context * lctx, bool do_shift, const stream_co
         LLAMA_LOG_DEBUG("%s: applying K-shift\n", __func__);
 
         // apply K-shift if needed
-        if (hparams.rope_type != LLAMA_ROPE_TYPE_NONE) {
+        if (hparams.rope_type != LLAMA_ROPE_TYPE_NONE &&
+                std::any_of(layers.begin(), layers.end(), [&](const auto & layer) {
+                    return hparams.has_rope(layer.il);
+                })) {
             ggml_backend_sched_reset(sched);
 
             auto * res = lctx->get_gf_res_reserve();

src/llama-model.cpp

@@ -113,6 +113,8 @@ static llama_model * llama_model_mapping(llm_arch arch, const llama_model_params
             return new llama_model_qwen2moe(params);
         case LLM_ARCH_QWEN3:
             return new llama_model_qwen3(params);
+        case LLM_ARCH_FLINT:
+            return new llama_model_flint(params);
         case LLM_ARCH_QWEN3MOE:
             return new llama_model_qwen3moe(params);
         case LLM_ARCH_QWEN3VL:
@@ -3170,6 +3172,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
         case LLM_ARCH_DREAM:
         case LLM_ARCH_QWEN2MOE:
         case LLM_ARCH_QWEN3:
+        case LLM_ARCH_FLINT:
         case LLM_ARCH_QWEN3MOE:
         case LLM_ARCH_LLADA_MOE:
         case LLM_ARCH_RND1:

src/llama-vocab.cpp

@@ -281,6 +281,13 @@ struct llm_tokenizer_bpe : llm_tokenizer {
     llm_tokenizer_bpe(const llama_vocab & vocab) {
         GGML_ASSERT(vocab.get_type() == LLAMA_VOCAB_TYPE_BPE);
         switch (vocab.get_pre_type()) {
+            case LLAMA_VOCAB_PRE_TYPE_FLINT:
+                regex_exprs = {
+                    "\\p{N}{1,3}",
+                    "[一-龥぀-ゟ゠-ヿ]+",
+                    "[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]*[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?[\\p{Lu}\\p{Lt}\\p{Lm}\\p{Lo}\\p{M}]+[\\p{Ll}\\p{Lm}\\p{Lo}\\p{M}]*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?| ?[\\p{P}\\p{S}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
+                };
+                break;
             case LLAMA_VOCAB_PRE_TYPE_LLAMA3:
                 regex_exprs = {
                     // original regex from tokenizer.json
@@ -2278,6 +2285,9 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                 LLAMA_LOG_WARN("%s: ************************************        \n", __func__);
                 LLAMA_LOG_WARN("%s:                                             \n", __func__);
                 pre_type = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
+            } else if (tokenizer_pre == "flint") {
+                pre_type = LLAMA_VOCAB_PRE_TYPE_FLINT;
+                clean_spaces = false;
             } else if (tokenizer_pre == "default") {
                 pre_type = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
             } else if (tokenizer_pre == "minicpm5") {
@@ -2997,7 +3007,6 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
         // this is currently determined based on the token text, which is obviously not ideal
         // ref: https://github.com/ggml-org/llama.cpp/issues/9606
         special_eog_ids.clear();
-
         if (special_fim_pad_id != LLAMA_TOKEN_NULL && special_eog_ids.count(special_fim_pad_id) == 0) {
             special_eog_ids.insert(special_fim_pad_id);
         }
@@ -3030,6 +3039,7 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
                     || t.first == "[EOT]" // Kimi-K2
                     || t.first == "[EOS]" // Kimi-K2
                     || t.first == "<|end_of_text|>"
+                    || (pre_type == LLAMA_VOCAB_PRE_TYPE_FLINT && (t.first == "<|user|>" || t.first == "<|observation|>"))
                     || t.first == "<end_of_utterance>" // smoldocling
                     || t.first == "<eos>"            // gemma4
                     || t.first == "<turn|>"          // gemma4

src/llama-vocab.h

@@ -70,6 +70,7 @@ enum llama_vocab_pre_type {
     LLAMA_VOCAB_PRE_TYPE_UFAKZEKA          = 59,
     LLAMA_VOCAB_PRE_TYPE_MMBERT            = 60,
     LLAMA_VOCAB_PRE_TYPE_K2_HORIZON        = 61,
+    LLAMA_VOCAB_PRE_TYPE_FLINT             = 62,
 };
 
 struct LLM_KV;