nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 33 Packages Projects Releases Wiki Activity

model : support granite multilingual embeddings R2 (ibm-granite/granite-embedding-{97,311}m-multilingual-r2) (#22716)

Browse Source 
* Add support for the ibm-granite/granite-embedding-{97m,311m}-multilingual-r2 embedding models:

* Added a version of the gpt4o tokenizer that has a fixed regex (better handling of marks), and different token merging setting for the 97m model
* Reused gemma4 tokenizer for the 311m model

* granite-embedding-*-multilingual-r2 : add support SwiGLU FFN for Granite Embedding Multilingual R2

* added new GGUF key <arch>.hidden_activation (LLM_KV_HIDDEN_ACT) + writer
* added a forward declaration of llm_ffn_op_type to llama-hparams.h
* added llm_ffn_op in hparams
* added LLM_FFN_NONE = 0 sentinel to llm_ffn_op_type (value-initialization), modern-bert: explicitly assigns LLM_FFN_GEGLU before reading GGUF (unchanged).
* centralized hidden_act mapping in llama-model.cpp, added llm_ffn_op_type_from_string() helper, mirroring rope_scaling_type/llama_rope_scaling_type_from_string()
* modern-bert reads the GGUF key (when present) and uses the resulting op in its FFN graph

* Added granite-embedding-{97m,311m}-multilingual-r2 to the converter code

* Added the hashes for the granite embedding multilingual R2 models
* Set the hidden_activation in the GGUF if the field is present in config.json (such as for the granite embedding models)
This commit is contained in:
Hans Florian
2026-06-02 17:55:11 +02:00
committed by GitHub
parent 2187e00337
commit bfb4308b05
14 changed files with 140 additions and 57 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-vocab.cpp
+16 -1
View File
@@ -432,6 +432,15 @@ struct llm_tokenizer_bpe : llm_tokenizer {
"[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))*((?=[\\p{L}])([^A-Z]))+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}])([^a-z]))+((?=[\\p{L}])([^A-Z]))*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
};
break;
case LLAMA_VOCAB_PRE_TYPE_GRANITE_EMB_MULTI:
// Same lookaheads as GPT4O but with \p{M} added so combining marks
// (diacritics) attach to their base letters. Avoids excessive
// backtracking on scripts that use them heavily (Bengali, Hindi,
// Telugu, Thai, ...). See PR #22716 for benchmarks.
regex_exprs = {
"[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}\\p{M}])([^a-z]))*((?=[\\p{L}\\p{M}])([^A-Z]))+(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|[^\\r\\n\\p{L}\\p{N}]?((?=[\\p{L}\\p{M}])([^a-z]))+((?=[\\p{L}\\p{M}])([^A-Z]))*(?:'[sS]|'[tT]|'[rR][eE]|'[vV][eE]|'[mM]|'[lL][lL]|'[dD])?|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n/]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+",
};
break;
case LLAMA_VOCAB_PRE_TYPE_TINY_AYA:
regex_exprs = {
// original regex from tokenizer.json: "\\d{1,3}(?=(?:\\d{3})*\\b)"
@@ -2142,7 +2151,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
tokenizer_pre == "jais-2") {
pre_type = LLAMA_VOCAB_PRE_TYPE_JAIS2;
} else if (
tokenizer_pre == "gemma4") {
tokenizer_pre == "gemma4" ||
tokenizer_pre == "granite-embed-multi-311m") {
pre_type = LLAMA_VOCAB_PRE_TYPE_GEMMA4;
escape_whitespaces = true;
} else if (
@@ -2252,6 +2262,11 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
tokenizer_pre == "talkie") {
pre_type = LLAMA_VOCAB_PRE_TYPE_GPT4O;
clean_spaces = false;
} else if (
tokenizer_pre == "granite-embed-multi-97m") {
pre_type = LLAMA_VOCAB_PRE_TYPE_GRANITE_EMB_MULTI;
clean_spaces = false;
ignore_merges = true;
} else if (
tokenizer_pre == "tiny_aya") {
pre_type = LLAMA_VOCAB_PRE_TYPE_TINY_AYA;