mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 32 Packages Projects Releases Wiki Activity

model: add Solar Open model (#18511)

Browse Source 
* model: add Solar-Open model

* vocab: add solar-open to end eog blacklist

* model: add proper llm type

* chat: basic template for solar open

* typo: fix comment about vocab

* convert: sugested changes

* convert: suggested changes

* chat: change reasoning end tag for solar-open

* llama-chat: add solar-open template
This commit is contained in:
HelloKS
2026-01-02 02:01:43 +09:00
committed by GitHub
parent d5574c919c
commit f4f5019254
11 changed files with 99 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-vocab.cpp
+15 -5
View File
@@ -355,6 +355,7 @@ struct llm_tokenizer_bpe : llm_tokenizer {
case LLAMA_VOCAB_PRE_TYPE_STABLELM2:
case LLAMA_VOCAB_PRE_TYPE_QWEN2:
case LLAMA_VOCAB_PRE_TYPE_HUNYUAN:
case LLAMA_VOCAB_PRE_TYPE_SOLAR_OPEN:
regex_exprs = {
// original regex from tokenizer.json
// "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+"
@@ -2015,6 +2016,10 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
tokenizer_pre == "minimax-m2") {
pre_type = LLAMA_VOCAB_PRE_TYPE_MINIMAX_M2;
clean_spaces = false;
} else if (
tokenizer_pre == "solar-open") {
pre_type = LLAMA_VOCAB_PRE_TYPE_SOLAR_OPEN;
clean_spaces = false;
} else {
throw std::runtime_error(format("unknown pre-tokenizer type: '%s'", tokenizer_pre.c_str()));
}
@@ -2358,6 +2363,8 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
|| t.first == "<|end|>"
|| t.first == "<|return|>" // o200k_harmony
|| t.first == "<|call|>" // o200k_harmony
|| t.first == "<|flush|>" // solar-open
|| t.first == "<|calls|>" // solar-open
|| t.first == "<end_of_turn>"
|| t.first == "<|endoftext|>"
|| t.first == "<|eom_id|>"
@@ -2404,13 +2411,14 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
LLAMA_LOG_WARN("%s: special_eom_id is not in special_eog_ids - the tokenizer config may be incorrect\n", __func__);
}
// TODO: workaround for o200k_harmony tokenizer: the "<|end|>" token should not be EOG
// we don't have a good way to detect this, so for now, if we have "<|return|>" and "<|call|>" tokens,
// TODO: workaround for o200k_harmony and solar-open tokenizer: the "<|end|>" token should not be EOG
// we don't have a good way to detect this, so for now, if we have "<|return|>" and "<|call|>" tokens ("<|calls|>" and "<|flush|>" for solar-open),
// we remove the "<|end|>" token from the EOG list
{
bool has_return = false;
bool has_call = false;
bool has_end = false;
bool has_flush = false;
llama_token end_id = LLAMA_TOKEN_NULL;
@@ -2420,18 +2428,20 @@ void llama_vocab::impl::load(llama_model_loader & ml, const LLM_KV & kv) {
if (id_to_token[tid].text == "<|return|>") {
has_return = true;
} else if (id_to_token[tid].text == "<|call|>") {
} else if (id_to_token[tid].text == "<|call|>" || id_to_token[tid].text == "<|calls|>") {
has_call = true;
} else if (id_to_token[tid].text == "<|flush|>") {
has_flush = true;
} else if (id_to_token[tid].text == "<|end|>") {
has_end = true;
end_id = tid;
}
}
if (has_return && has_call && has_end) {
if ((has_return && has_call && has_end) || (has_call && has_flush && has_end)) {
special_eog_ids.erase(end_id);
id_to_token[end_id].attr = LLAMA_TOKEN_ATTR_USER_DEFINED;
LLAMA_LOG_WARN("%s: special_eog_ids contains both '<|return|>' and '<|call|>' tokens, removing '<|end|>' token from EOG list\n", __func__);
LLAMA_LOG_WARN("%s: special_eog_ids contains both '<|return|>' and '<|call|>', or '<|calls|>' and '<|flush|>' tokens, removing '<|end|>' token from EOG list\n", __func__);
}
}
}