mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 28 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-chat.cpp
+11
View File
@@ -74,6 +74,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
{ "seed_oss", LLM_CHAT_TEMPLATE_SEED_OSS },
{ "grok-2", LLM_CHAT_TEMPLATE_GROK_2 },
{ "pangu-embedded", LLM_CHAT_TEMPLATE_PANGU_EMBED },
{ "solar-open", LLM_CHAT_TEMPLATE_SOLAR_OPEN },
};
llm_chat_template llm_chat_template_from_str(const std::string & name) {
@@ -216,6 +217,8 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
return LLM_CHAT_TEMPLATE_GROK_2;
} else if (tmpl_contains(LU8("[unused9]系统：[unused10]"))) {
return LLM_CHAT_TEMPLATE_PANGU_EMBED;
} else if (tmpl_contains("<|begin|>") && tmpl_contains("<|end|>") && tmpl_contains("<|content|>")) {
return LLM_CHAT_TEMPLATE_SOLAR_OPEN;
}
return LLM_CHAT_TEMPLATE_UNKNOWN;
}
@@ -845,6 +848,14 @@ int32_t llm_chat_apply_template(
if (add_ass) {
ss << "[unused9]助手：";
}
} else if (tmpl == LLM_CHAT_TEMPLATE_SOLAR_OPEN) {
for (auto message : chat) {
std::string role(message->role);
ss << "<|begin|>" << role << "<|content|>" << message->content << "<|end|>";
}
if (add_ass) {
ss << "<|begin|>assistant";
}
} else {
// template not supported
return -1;
src/llama-chat.h
+1
View File
@@ -54,6 +54,7 @@ enum llm_chat_template {
LLM_CHAT_TEMPLATE_SEED_OSS,
LLM_CHAT_TEMPLATE_GROK_2,
LLM_CHAT_TEMPLATE_PANGU_EMBED,
LLM_CHAT_TEMPLATE_SOLAR_OPEN,
LLM_CHAT_TEMPLATE_UNKNOWN,
};
src/llama-model.cpp
+5 -3
View File
@@ -126,6 +126,7 @@ const char * llm_type_name(llm_type type) {
case LLM_TYPE_31B_A3_5B: return "31B.A3.5B";
case LLM_TYPE_80B_A3B: return "80B.A3B";
case LLM_TYPE_100B_A6B: return "100B.A6B";
case LLM_TYPE_102B_A12B: return "102B.A12B";
case LLM_TYPE_106B_A12B: return "106B.A12B";
case LLM_TYPE_230B_A10B: return "230B.A10B";
case LLM_TYPE_235B_A22B: return "235B.A22B";
@@ -1778,6 +1779,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
switch (hparams.n_layer) {
case 47: type = LLM_TYPE_106B_A12B; break; // GLM-4.5-Air (46 layers + 1 NextN layer)
case 48: type = LLM_TYPE_102B_A12B; break; // Solar Open
case 93: type = LLM_TYPE_355B_A32B; break; // GLM-4.5 (92 layers + 1 NextN layer)
default: type = LLM_TYPE_UNKNOWN;
}
@@ -5206,9 +5208,9 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), { n_embd, n_embd_head_k * n_head }, flags);
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), { n_embd, n_embd_k_gqa }, flags);
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), { n_embd, n_embd_v_gqa }, flags);
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), { n_embd_head_k * n_head }, flags);
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), { n_embd_k_gqa }, flags);
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), { n_embd_v_gqa }, flags);
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), { n_embd_head_k * n_head }, TENSOR_NOT_REQUIRED | flags);
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), { n_embd_k_gqa }, TENSOR_NOT_REQUIRED | flags);
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), { n_embd_v_gqa }, TENSOR_NOT_REQUIRED | flags);
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd_head_k * n_head, n_embd }, flags);
src/llama-model.h
+1
View File
@@ -119,6 +119,7 @@ enum llm_type {
LLM_TYPE_31B_A3_5B,
LLM_TYPE_80B_A3B, // Qwen3 Next
LLM_TYPE_100B_A6B,
LLM_TYPE_102B_A12B, // Solar-Open
LLM_TYPE_106B_A12B, // GLM-4.5-Air
LLM_TYPE_230B_A10B, // Minimax M2
LLM_TYPE_235B_A22B,
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__);
}
}
}
src/llama-vocab.h
+1
View File
@@ -51,6 +51,7 @@ enum llama_vocab_pre_type {
LLAMA_VOCAB_PRE_TYPE_GRANITE_DOCLING = 40,
LLAMA_VOCAB_PRE_TYPE_MINIMAX_M2 = 41,
LLAMA_VOCAB_PRE_TYPE_AFMOE = 42,
LLAMA_VOCAB_PRE_TYPE_SOLAR_OPEN = 43,
};
struct LLM_KV;