@novastera-oss/llamarn 0.2.7 → 0.3.0

package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py

@@ -13,7 +13,7 @@ class TensorNameMap:
             "transformer.wte",                           # gpt2 gpt-j mpt refact qwen dbrx jais exaone
             "transformer.word_embeddings",               # falcon
             "word_embeddings",                           # bloom
-            "model.embed_tokens",                        # llama-hf nemotron olmoe olmo2 rwkv6qwen2 glm4-0414
+            "model.embed_tokens",                        # llama-hf nemotron olmoe olmo2 rwkv6qwen2 glm4-0414 granite-hybrid
             "tok_embeddings",                            # llama-pth
             "embeddings.word_embeddings",                # bert nomic-bert
             "language_model.embedding.word_embeddings",  # persimmon
@@ -50,6 +50,7 @@ class TensorNameMap:
             "model.pre_ln",               # rwkv7
             "model.layers.0.pre_norm",    # rwkv7
             "backbone.norm",              # wavtokenizer
+            "model.embedding_norm",       # lfm2
         ),
 
         # Position embeddings
@@ -118,7 +119,7 @@ class TensorNameMap:
             "transformer.h.{bid}.input_layernorm",                  # falcon7b
             "h.{bid}.input_layernorm",                              # bloom
             "transformer.h.{bid}.ln_mlp",                           # falcon40b
-            "model.layers.{bid}.input_layernorm",                   # llama-hf nemotron olmoe phimoe
+            "model.layers.{bid}.input_layernorm",                   # llama-hf nemotron olmoe phimoe granite-hybrid
             "layers.{bid}.attention_norm",                          # llama-pth
             "language_model.encoder.layers.{bid}.input_layernorm",  # persimmon
             "model.layers.{bid}.ln1",                               # yi
@@ -136,6 +137,7 @@ class TensorNameMap:
             "model.layers.{bid}.ln1",                               # rwkv7
             "model.layers.{bid}.input_layernorm",                   # llama4
             "transformer_encoder.{bid}.attention_norm",             # neobert
+            "model.layers.{bid}.operator_norm",                     # lfm2
         ),
 
         # Attention norm 2
@@ -220,6 +222,7 @@ class TensorNameMap:
             "transformer.h.{bid}.self_attention.dense",                     # falcon
             "h.{bid}.self_attention.dense",                                 # bloom
             "model.layers.{bid}.self_attn.o_proj",                          # llama-hf nemotron olmoe olmo2 phimoe
+            "model.layers.{bid}.self_attn.out_proj",                        # lfm2
             "model.layers.{bid}.self_attn.linear_attn",                     # deci
             "layers.{bid}.attention.wo",                                    # llama-pth
             "encoder.layer.{bid}.attention.output.dense",                   # bert
@@ -279,6 +282,8 @@ class TensorNameMap:
             "transformer.decoder_layer.{bid}.rms_norm_2",                    # Grok
             "encoder.layers.{bid}.post_attention_layernorm",                 # chatglm
             "transformer.layers.{bid}.ffn_norm",                             # openelm
+            "model.layers.{bid}.pre_ff_layernorm",                           # jamba granite-hybrid
+            "model.layers.{bid}.pre_moe_layernorm",                          # mini-jamba
             "model.layers.{bid}.post_attention_layernorm",                   # llama4
             "transformer_encoder.{bid}.ffn_norm",                            # neobert
         ),
@@ -286,12 +291,14 @@ class TensorNameMap:
         # Post feed-forward norm
         MODEL_TENSOR.FFN_PRE_NORM: (
             "model.layers.{bid}.pre_feedforward_layernorm", # gemma2
+            "model.layers.{bid}.pre_ff_layernorm.weight",
         ),
 
         # Post feed-forward norm
         MODEL_TENSOR.FFN_POST_NORM: (
             "model.layers.{bid}.post_feedforward_layernorm", # gemma2 olmo2
             "model.layers.{bid}.post_mlp_layernorm", # glm-4-0414
+            "model.layers.{bid}.feed_forward.up_proj",
         ),
 
         MODEL_TENSOR.FFN_GATE_INP: (
@@ -301,8 +308,9 @@ class TensorNameMap:
             "transformer.decoder_layer.{bid}.router",           # Grok
             "transformer.blocks.{bid}.ffn.router.layer",        # dbrx
             "model.layers.{bid}.block_sparse_moe.router.layer", # granitemoe
-            "model.layers.{bid}.feed_forward.router",           # llama4
+            "model.layers.{bid}.feed_forward.router",           # llama4 jamba
             "encoder.layers.{bid}.mlp.router.layer",            # nomic-bert-moe
+            "model.layers.{bid}.mlp.gate.wg",                   # hunyuan
         ),
 
         MODEL_TENSOR.FFN_GATE_INP_SHEXP: (
@@ -344,7 +352,7 @@ class TensorNameMap:
             "model.layers.{bid}.residual_mlp.w3",                     # arctic
             "encoder.layers.{bid}.mlp.dense_h_to_4h",                 # chatglm
             "transformer.h.{bid}.mlp.c_fc_1",                         # exaone
-            "model.layers.{bid}.feed_forward.up_proj",                # llama4
+            "model.layers.{bid}.feed_forward.up_proj",                # llama4 jamba granite-hybrid
             "transformer_encoder.{bid}.ffn.w12",                      # neobert
         ),
 
@@ -362,6 +370,8 @@ class TensorNameMap:
             "model.layers.{bid}.mlp.shared_expert.up_proj",          # qwen2moe
             "model.layers.{bid}.mlp.shared_experts.up_proj",         # deepseek deepseek2
             "model.layers.{bid}.feed_forward.shared_expert.up_proj", # llama4
+            "model.layers.{bid}.feed_forward.down_proj",
+            "model.layers.{bid}.mlp.shared_mlp.up_proj",             # hunyuan
         ),
 
         # AWQ-activation gate
@@ -382,7 +392,7 @@ class TensorNameMap:
             "transformer.h.{bid}.mlp.linear_1",           # refact
             "model.layers.{bid}.residual_mlp.w1",         # arctic
             "transformer.h.{bid}.mlp.c_fc_0",             # exaone
-            "model.layers.{bid}.feed_forward.gate_proj",  # llama4
+            "model.layers.{bid}.feed_forward.gate_proj",  # llama4 jamba granite-hybrid
         ),
 
         MODEL_TENSOR.FFN_GATE_EXP: (
@@ -398,6 +408,7 @@ class TensorNameMap:
             "model.layers.{bid}.mlp.shared_expert.gate_proj",          # qwen2moe
             "model.layers.{bid}.mlp.shared_experts.gate_proj",         # deepseek deepseek2
             "model.layers.{bid}.feed_forward.shared_expert.gate_proj", # llama4
+            "model.layers.{bid}.mlp.shared_mlp.gate_proj",             # hunyuan
         ),
 
         # Feed-forward down
@@ -427,7 +438,7 @@ class TensorNameMap:
             "encoder.layer.{bid}.mlp.down_layer",                     # jina-bert-v2
             "encoder.layers.{bid}.mlp.dense_4h_to_h",                 # chatglm
             "model.layers.h.{bid}.mlp.c_proj",                        # exaone
-            "model.layers.{bid}.feed_forward.down_proj",              # llama4
+            "model.layers.{bid}.feed_forward.down_proj",              # llama4 jamba granite-hybrid
             "transformer_encoder.{bid}.ffn.w3",                       # neobert
         ),
 
@@ -447,11 +458,13 @@ class TensorNameMap:
             "model.layers.{bid}.mlp.shared_experts.down_proj",         # deepseek deepseek2
             "model.layers.{bid}.feed_forward.shared_expert.down_proj", # llama4
             "model.layers.{bid}.shared_mlp.output_linear",             # granitemoe
+            "model.layers.{bid}.mlp.shared_mlp.down_proj",             # hunyuan
         ),
 
         MODEL_TENSOR.ATTN_Q_NORM: (
             "language_model.encoder.layers.{bid}.self_attention.q_layernorm",
             "model.layers.{bid}.self_attn.q_layernorm",                       # persimmon
+            "model.layers.{bid}.self_attn.query_layernorm",                   # hunyuan
             "model.layers.{bid}.self_attn.q_norm",                            # cohere olmoe chameleon olmo2
             "transformer.blocks.{bid}.attn.q_ln",                             # sea-lion
             "encoder.layer.{bid}.attention.self.layer_norm_q",                # jina-bert-v2
@@ -461,6 +474,7 @@ class TensorNameMap:
         MODEL_TENSOR.ATTN_K_NORM: (
             "language_model.encoder.layers.{bid}.self_attention.k_layernorm",
             "model.layers.{bid}.self_attn.k_layernorm",                       # persimmon
+            "model.layers.{bid}.self_attn.key_layernorm",                     # hunyuan
             "model.layers.{bid}.self_attn.k_norm",                            # cohere olmoe chameleon olmo2
             "transformer.blocks.{bid}.attn.k_ln",                             # sea-lion
             "encoder.layer.{bid}.attention.self.layer_norm_k",                # jina-bert-v2
@@ -477,42 +491,132 @@ class TensorNameMap:
             "encoder.layers.{bid}.norm2",                   # nomic-bert
             "transformer.decoder_layer.{bid}.rms_norm_3",   # Grok
             "encoder.layer.{bid}.mlp.layernorm",            # jina-bert-v2
-            "encoder.layer.{bid}.layer_norm_2"              # jina-v2-code
+            "encoder.layer.{bid}.layer_norm_2",             # jina-v2-code
+        ),
+
+        MODEL_TENSOR.PER_LAYER_TOKEN_EMBD: (
+            "model.embed_tokens_per_layer",  # gemma3n
+        ),
+
+        MODEL_TENSOR.PER_LAYER_MODEL_PROJ: (
+            "model.per_layer_model_projection",  # gemma3n
+        ),
+
+        MODEL_TENSOR.PER_LAYER_PROJ_NORM: (
+            "model.per_layer_projection_norm",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_PROJ: (
+            "model.altup_projections",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_UNEMBD_PROJ: (
+            "model.altup_unembed_projections",  # gemma3n
+        ),
+
+        MODEL_TENSOR.PER_LAYER_INP_GATE: (
+            "model.layers.{bid}.per_layer_input_gate",  # gemma3n
+        ),
+
+        MODEL_TENSOR.PER_LAYER_PROJ: (
+            "model.layers.{bid}.per_layer_projection",  # gemma3n
+        ),
+
+        MODEL_TENSOR.PER_LAYER_POST_NORM: (
+            "model.layers.{bid}.post_per_layer_input_norm",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_CORRECT_COEF: (
+            "model.layers.{bid}.altup.correction_coefs",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_CORRECT_SCALE: (
+            "model.layers.{bid}.altup.correct_output_scale",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_PREDICT_COEF: (
+            "model.layers.{bid}.altup.prediction_coefs",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_ROUTER: (
+            "model.layers.{bid}.altup.modality_router",  # gemma3n
+        ),
+
+        MODEL_TENSOR.ALTUP_ROUTER_NORM: (
+            "model.layers.{bid}.altup.router_norm",  # gemma3n
+        ),
+
+        MODEL_TENSOR.LAUREL_L: (
+            "model.layers.{bid}.laurel.linear_left",  # gemma3n
+        ),
+
+        MODEL_TENSOR.LAUREL_R: (
+            "model.layers.{bid}.laurel.linear_right",  # gemma3n
+        ),
+
+        MODEL_TENSOR.LAUREL_POST_NORM: (
+            "model.layers.{bid}.laurel.post_laurel_norm",  # gemma3n
         ),
 
         MODEL_TENSOR.SSM_IN: (
-            "model.layers.{bid}.in_proj",
-            "backbone.layers.{bid}.mixer.in_proj",
+            "model.layers.{bid}.in_proj",           # mamba-hf
+            "backbone.layers.{bid}.mixer.in_proj",  # mamba
+            "model.layers.{bid}.mamba.in_proj",     # jamba falcon-h1 granite-hybrid
         ),
 
         MODEL_TENSOR.SSM_CONV1D: (
-            "model.layers.{bid}.conv1d",
-            "backbone.layers.{bid}.mixer.conv1d",
+            "model.layers.{bid}.conv1d",           # mamba-hf
+            "backbone.layers.{bid}.mixer.conv1d",  # mamba
+            "model.layers.{bid}.mamba.conv1d",     # jamba falcon-h1 granite-hybrid
         ),
 
         MODEL_TENSOR.SSM_X: (
-            "model.layers.{bid}.x_proj",
-            "backbone.layers.{bid}.mixer.x_proj",
+            "model.layers.{bid}.x_proj",           # mamba-hf
+            "backbone.layers.{bid}.mixer.x_proj",  # mamba
+            "model.layers.{bid}.mamba.x_proj",     # jamba
         ),
 
         MODEL_TENSOR.SSM_DT: (
-            "model.layers.{bid}.dt_proj",
-            "backbone.layers.{bid}.mixer.dt_proj",
+            "model.layers.{bid}.dt_proj",           # mamba-hf
+            "backbone.layers.{bid}.mixer.dt_proj",  # mamba
+            "model.layers.{bid}.mamba.dt_proj",     # jamba falcon-h1 granite-hybrid
+        ),
+
+        MODEL_TENSOR.SSM_DT_NORM: (
+            "model.layers.{bid}.mamba.dt_layernorm",  # jamba
         ),
 
         MODEL_TENSOR.SSM_A: (
-            "model.layers.{bid}.A_log",
-            "backbone.layers.{bid}.mixer.A_log",
+            "model.layers.{bid}.A_log",           # mamba-hf
+            "backbone.layers.{bid}.mixer.A_log",  # mamba
+            "model.layers.{bid}.mamba.A_log",     # jamba falcon-h1 granite-hybrid
+        ),
+
+        MODEL_TENSOR.SSM_B_NORM: (
+            "model.layers.{bid}.mamba.b_layernorm",  # jamba
+            "model.layers.{bid}.mamba.B_layernorm",  # mini-jamba
+        ),
+
+        MODEL_TENSOR.SSM_C_NORM: (
+            "model.layers.{bid}.mamba.c_layernorm",  # jamba
+            "model.layers.{bid}.mamba.C_layernorm",  # mini-jamba
         ),
 
         MODEL_TENSOR.SSM_D: (
-            "model.layers.{bid}.D",
-            "backbone.layers.{bid}.mixer.D",
+            "model.layers.{bid}.D",           # mamba-hf
+            "backbone.layers.{bid}.mixer.D",  # mamba
+            "model.layers.{bid}.mamba.D",     # jamba falcon-h1 granite-hybrid
+        ),
+
+        MODEL_TENSOR.SSM_NORM: (
+            "model.layers.{bid}.mamba.norm", # falcon-h1 granite-hybrid
+            "backbone.layers.{bid}.mixer.norm",  # mamba2
         ),
 
         MODEL_TENSOR.SSM_OUT: (
-            "model.layers.{bid}.out_proj",
-            "backbone.layers.{bid}.mixer.out_proj",
+            "model.layers.{bid}.out_proj",           # mamba-hf
+            "backbone.layers.{bid}.mixer.out_proj",  # mamba
+            "model.layers.{bid}.mamba.out_proj",     # jamba falcon-h1 granite-hybrid
         ),
 
         MODEL_TENSOR.TIME_MIX_W0: (
@@ -914,6 +1018,18 @@ class TensorNameMap:
             "backbone.posnet.{bid}.proj_out", # wavtokenizer
         ),
 
+        MODEL_TENSOR.SHORTCONV_CONV: (
+            "model.layers.{bid}.conv.conv",
+        ),
+
+        MODEL_TENSOR.SHORTCONV_INPROJ: (
+            "model.layers.{bid}.conv.in_proj",
+        ),
+
+        MODEL_TENSOR.SHORTCONV_OUTPROJ: (
+            "model.layers.{bid}.conv.out_proj",
+        ),
+
         #############################################################################
         ## Vision encoder
 

package/cpp/llama.cpp/gguf-py/gguf/vocab.py

@@ -7,7 +7,10 @@ import os
 from pathlib import Path
 from typing import Any, Callable, Sequence, Mapping, Iterable, Protocol, ClassVar, runtime_checkable
 
-from sentencepiece import SentencePieceProcessor
+try:
+    from sentencepiece import SentencePieceProcessor
+except ImportError:
+    SentencePieceProcessor = None
 
 import gguf
 
@@ -116,6 +119,7 @@ class SpecialVocab:
         logger.warning(f'Special token type {typ}, id {tid} out of range, must be under {self.n_vocab} - skipping')
 
     def _try_load_from_tokenizer_json(self, path: Path) -> bool:
+        tokenizer = None
         tokenizer_file = path / 'tokenizer.json'
         if tokenizer_file.is_file():
             with open(tokenizer_file, encoding = 'utf-8') as f:
@@ -149,15 +153,110 @@ class SpecialVocab:
             added_tokens = tokenizer.get('added_tokens', {})
         else:
             added_tokens = {}
+        tokenizer_config = None
         tokenizer_config_file = path / 'tokenizer_config.json'
-        if not tokenizer_config_file.is_file():
+        if tokenizer_config_file.is_file():
+            with open(tokenizer_config_file, encoding = 'utf-8') as f:
+                tokenizer_config = json.load(f)
+        if tokenizer:
+            special_bos = (tokenizer_config or {}).get('bos_token')
+            special_cls = (tokenizer_config or {}).get('cls_token')
+            special_eos = (tokenizer_config or {}).get('eos_token')
+            special_sep = (tokenizer_config or {}).get('sep_token')
+            if not special_bos and special_cls and tokenizer_config:
+                tokenizer_config['bos_token'] = special_bos = special_cls
+            if not special_eos and special_sep and tokenizer_config:
+                tokenizer_config['eos_token'] = special_eos = special_sep
+            if post_processor := tokenizer.get('post_processor'):
+                for processor in post_processor.get('processors', [post_processor]):
+                    if processor.get('type') == 'RobertaProcessing':
+                        self.add_special_token['bos'] = True
+                        self.add_special_token['eos'] = True
+                        self.add_special_token['sep'] = True
+                        if not special_cls and tokenizer_config:
+                            special_cls = processor.get('cls', [special_bos])[0]
+                            tokenizer_config['cls_token'] = special_cls
+                        if not special_sep and tokenizer_config:
+                            special_sep = processor.get('sep', [special_eos])[0]
+                            tokenizer_config['sep_token'] = special_sep
+                        continue
+                    # Crude parsing of TemplateProcessing to determine if BOS/SEP/EOS should be added
+                    # Only works with simple templates, **will** get it wrong on unusual sequences
+                    if processor.get('type') == 'TemplateProcessing':
+                        tmpl_single = processor.get('single', [])
+                        tmpl_pair = processor.get('pair', [])
+                        special_first = None
+                        special_last = None
+                        if len(tmpl_single) > 1:
+                            if special_first := tmpl_single[0].get('SpecialToken', {}).get('id'):
+                                if not tokenizer_config:
+                                    special_bos = special_first
+                                self.add_special_token['bos'] = True if special_first in (special_bos, special_cls) else False
+                                if special_first not in (special_bos, special_cls):
+                                    logger.warning(f'Unknown leading special token {special_first!r} in TemplateProcessing<single>')
+                            if special_last := tmpl_single[-1].get('SpecialToken', {}).get('id'):
+                                if not tokenizer_config:
+                                    special_eos = special_last
+                                elif special_last != special_eos:
+                                    if 'eot' not in self.special_token_types:
+                                        self.special_token_types = tuple(self.special_token_types) + ('eot', )
+                                        tokenizer_config['eot_token'] = special_eos
+                                    elif 'eom' not in self.special_token_types:
+                                        self.special_token_types = tuple(self.special_token_types) + ('eom', )
+                                        tokenizer_config['eom_token'] = special_eos
+                                    else:
+                                        logger.warning(f'Overriding EOS token {special_eos!r} with {special_last!r} without EOT/EOM fallback!')
+                                    tokenizer_config['eos_token'] = special_eos = special_last
+                                self.add_special_token['eos'] = True if special_last == special_eos else False
+                                if special_last != special_eos:
+                                    logger.warning(f'Unknown trailing special token {special_last!r} in TemplateProcessing<single>')
+                        if tmpl_pair:
+                            seq_start = 1 if special_first and tmpl_pair[0].get('SpecialToken', {}).get('id') == special_first else 0
+                            seq_stop = -1 if special_last and tmpl_pair[-1].get('SpecialToken', {}).get('id') == special_last else None
+                            if (special_first and seq_start == 0) or (special_last and seq_stop is None):
+                                logger.warning('TemplateProcessing<single> leading/trailing special tokens do not match TemplateProcessing<pair>')
+                            if tmpl_pair := tmpl_pair[slice(seq_start, seq_stop)]:
+                                tmpl_a = tmpl_pair[0].get('Sequence', {}).get('id')
+                                tmpl_b = tmpl_pair[-1].get('Sequence', {}).get('id')
+                                if tmpl_a != 'A' or tmpl_b != 'B':
+                                    logger.warning(f'Unknown sequence {tmpl_a}...{tmpl_b} in TemplateProcessing<pair>')
+                                # A [sep] [eos] B
+                                if tmpl_a == 'A' and tmpl_b == 'B' and (tmpl_pair := tmpl_pair[1:-1]):
+                                    add_sep = False
+                                    if special_entry := tmpl_pair[0].get('SpecialToken', {}).get('id'):
+                                        if special_entry in (special_sep, special_eos) and not special_last:
+                                            add_sep = True
+                                        if special_entry not in (special_sep, special_eos):
+                                            logger.warning(f'Unknown separator token {special_entry!r} in TemplateProcessing<pair>')
+                                    else:
+                                        logger.warning(f'Unknown middle sequence {tmpl_pair[0]!r} in TemplateProcessing<pair>')
+                                    if len(tmpl_pair) == 2:
+                                        if special_entry := tmpl_pair[1].get('SpecialToken', {}).get('id'):
+                                            if special_entry in (special_sep, special_eos):
+                                                add_sep = True
+                                            if special_entry not in (special_sep, special_eos):
+                                                logger.warning(f'Unknown second separator token {special_entry!r} in TemplateProcessing<pair>')
+                                        else:
+                                            logger.warning(f'Unknown second middle sequence {tmpl_pair[1]!r} in TemplateProcessing<pair>')
+                                    self.add_special_token['sep'] = add_sep
+                                    if add_sep and not special_sep and tokenizer_config:
+                                        tokenizer_config['sep_token'] = special_eos
+                        continue
+        if not tokenizer_config:
             return True
-        with open(tokenizer_config_file, encoding = 'utf-8') as f:
-            tokenizer_config = json.load(f)
         chat_template_alt = None
-        chat_template_file = path / 'chat_template.json'
-        if chat_template_file.is_file():
-            with open(chat_template_file, encoding = 'utf-8') as f:
+        chat_template_json = path / 'chat_template.json'
+        chat_template_jinja = path / 'chat_template.jinja'
+        if chat_template_jinja.is_file():
+            with open(chat_template_jinja, encoding = 'utf-8') as f:
+                chat_template_alt = f.read()
+            if additional_templates := list((path / 'additional_chat_templates').glob('*.jinja')):
+                chat_template_alt = [{'name': 'default', 'template': chat_template_alt}]
+                for template_path in additional_templates:
+                    with open(template_path, encoding = 'utf-8') as fp:
+                        chat_template_alt.append({'name': template_path.stem, 'template': fp.read()})
+        elif chat_template_json.is_file():
+            with open(chat_template_json, encoding = 'utf-8') as f:
                 chat_template_alt = json.load(f).get('chat_template')
         chat_template = tokenizer_config.get('chat_template', chat_template_alt)
         if chat_template is None or isinstance(chat_template, (str, list)):
@@ -302,6 +401,9 @@ class SentencePieceVocab(Vocab):
     name = "spm"
 
     def __init__(self, base_path: Path):
+        if SentencePieceProcessor is None:
+            raise RuntimeError("sentencepiece is not installed")
+
         added_tokens: dict[str, int] = {}
         if (fname_tokenizer := base_path / 'tokenizer.model').exists():
             # normal location

package/cpp/llama.cpp/gguf-py/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "gguf"
-version = "0.17.0"
+version = "0.17.1"
 description = "Read and write ML models in GGUF for GGML"
 authors = ["GGML <ggml@ggml.ai>"]
 packages = [
@@ -22,7 +22,7 @@ python = ">=3.8"
 numpy = ">=1.17"
 tqdm = ">=4.27"
 pyyaml = ">=5.1"
-sentencepiece = ">=0.1.98,<=0.2.0"
+sentencepiece = { version = ">=0.1.98,<=0.2.0", optional = true }
 PySide6 = { version = "^6.9", python = ">=3.9,<3.14", optional = true }
 
 [tool.poetry.dev-dependencies]

package/cpp/llama.cpp/include/llama.h

@@ -79,46 +79,6 @@ extern "C" {
         LLAMA_VOCAB_TYPE_RWKV = 5, // RWKV tokenizer based on greedy tokenization
     };
 
-    // pre-tokenization types
-    enum llama_vocab_pre_type {
-        LLAMA_VOCAB_PRE_TYPE_DEFAULT        = 0,
-        LLAMA_VOCAB_PRE_TYPE_LLAMA3         = 1,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM   = 2,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER = 3,
-        LLAMA_VOCAB_PRE_TYPE_FALCON         = 4,
-        LLAMA_VOCAB_PRE_TYPE_MPT            = 5,
-        LLAMA_VOCAB_PRE_TYPE_STARCODER      = 6,
-        LLAMA_VOCAB_PRE_TYPE_GPT2           = 7,
-        LLAMA_VOCAB_PRE_TYPE_REFACT         = 8,
-        LLAMA_VOCAB_PRE_TYPE_COMMAND_R      = 9,
-        LLAMA_VOCAB_PRE_TYPE_STABLELM2      = 10,
-        LLAMA_VOCAB_PRE_TYPE_QWEN2          = 11,
-        LLAMA_VOCAB_PRE_TYPE_OLMO           = 12,
-        LLAMA_VOCAB_PRE_TYPE_DBRX           = 13,
-        LLAMA_VOCAB_PRE_TYPE_SMAUG          = 14,
-        LLAMA_VOCAB_PRE_TYPE_PORO           = 15,
-        LLAMA_VOCAB_PRE_TYPE_CHATGLM3       = 16,
-        LLAMA_VOCAB_PRE_TYPE_CHATGLM4       = 17,
-        LLAMA_VOCAB_PRE_TYPE_VIKING         = 18,
-        LLAMA_VOCAB_PRE_TYPE_JAIS           = 19,
-        LLAMA_VOCAB_PRE_TYPE_TEKKEN         = 20,
-        LLAMA_VOCAB_PRE_TYPE_SMOLLM         = 21,
-        LLAMA_VOCAB_PRE_TYPE_CODESHELL      = 22,
-        LLAMA_VOCAB_PRE_TYPE_BLOOM          = 23,
-        LLAMA_VOCAB_PRE_TYPE_GPT3_FINNISH   = 24,
-        LLAMA_VOCAB_PRE_TYPE_EXAONE         = 25,
-        LLAMA_VOCAB_PRE_TYPE_CHAMELEON      = 26,
-        LLAMA_VOCAB_PRE_TYPE_MINERVA        = 27,
-        LLAMA_VOCAB_PRE_TYPE_DEEPSEEK3_LLM  = 28,
-        LLAMA_VOCAB_PRE_TYPE_GPT4O          = 29,
-        LLAMA_VOCAB_PRE_TYPE_SUPERBPE       = 30,
-        LLAMA_VOCAB_PRE_TYPE_TRILLION       = 31,
-        LLAMA_VOCAB_PRE_TYPE_BAILINGMOE     = 32,
-        LLAMA_VOCAB_PRE_TYPE_LLAMA4         = 33,
-        LLAMA_VOCAB_PRE_TYPE_PIXTRAL        = 34,
-        LLAMA_VOCAB_PRE_TYPE_SEED_CODER     = 35,
-    };
-
     enum llama_rope_type {
         LLAMA_ROPE_TYPE_NONE   = -1,
         LLAMA_ROPE_TYPE_NORM   = 0,
@@ -390,6 +350,7 @@ extern "C" {
         void * imatrix;                       // pointer to importance matrix data
         void * kv_overrides;                  // pointer to vector containing overrides
         void * tensor_types;                  // pointer to vector containing tensor types
+        void * prune_layers;                  // pointer to vector containing layer indices to prune
     } llama_model_quantize_params;
 
     typedef struct llama_logit_bias {
@@ -943,12 +904,14 @@ extern "C" {
     // Requires the context to have a memory.
     // For encode-decoder contexts, processes the batch using the decoder.
     // Positive return values does not mean a fatal error, but rather a warning.
-    // Upon non-zero return values, the memory state is restored to the state before this call
+    // Upon fatal-error or abort, the ubatches that managed to be been processed will remain in the memory state of the context
+    //   To handle this correctly, query the memory state using llama_memory_seq_pos_min() and llama_memory_seq_pos_max()
+    // Upon other return values, the memory state is restored to the state before this call
     //    0 - success
     //    1 - could not find a KV slot for the batch (try reducing the size of the batch or increase the context)
-    //    2 - aborted
+    //    2 - aborted     (processed ubatches will remain in the context's memory)
     //   -1 - invalid input batch
-    // < -1 - error
+    // < -1 - fatal error (processed ubatches will remain in the context's memory)
     LLAMA_API int32_t llama_decode(
             struct llama_context * ctx,
               struct llama_batch   batch);
@@ -1044,6 +1007,7 @@ extern "C" {
 
     LLAMA_API bool llama_vocab_get_add_bos(const struct llama_vocab * vocab);
     LLAMA_API bool llama_vocab_get_add_eos(const struct llama_vocab * vocab);
+    LLAMA_API bool llama_vocab_get_add_sep(const struct llama_vocab * vocab);
 
     LLAMA_API llama_token llama_vocab_fim_pre(const struct llama_vocab * vocab);
     LLAMA_API llama_token llama_vocab_fim_suf(const struct llama_vocab * vocab);
@@ -1087,6 +1051,7 @@ extern "C" {
     /// @param tokens The tokens pointer must be large enough to hold the resulting tokens.
     /// @return Returns the number of tokens on success, no more than n_tokens_max
     /// @return Returns a negative number on failure - the number of tokens that would have been returned
+    /// @return Returns INT32_MIN on overflow (e.g., tokenization result size exceeds int32_t limit)
     /// @param add_special Allow to add BOS and EOS tokens if model is configured to do so.
     /// @param parse_special Allow tokenizing special and/or control tokens which otherwise are not exposed and treated
     ///                      as plaintext. Does not insert a leading space.