Skip to content

Commit aa973ca

Browse files
stevenkuang-tencentstevenkuang-tencent
committed
update hunyuan_moe to hunyuan_v1_moe
Signed-off-by: stevenkuang <[email protected]>
1 parent 5d2c042 commit aa973ca

File tree

8 files changed

+20
-20
lines changed

8 files changed

+20
-20
lines changed

‎convert_hf_to_gguf.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7387,7 +7387,7 @@ def set_gguf_parameters(self):
73877387

73887388
@ModelBase.register("HunYuanMoEV1ForCausalLM")
73897389
class HunYuanMoEModel(TextModel):
7390-
model_arch = gguf.MODEL_ARCH.HUNYUAN_MOE
7390+
model_arch = gguf.MODEL_ARCH.HUNYUAN_V1_MOE
73917391

73927392
def __init__(self, *args, **kwargs):
73937393
super().__init__(*args, **kwargs)

‎convert_hf_to_gguf_update.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -139,7 +139,7 @@ class TOKENIZER_TYPE(IntEnum):
139139
{"name": "chatglm-bpe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/THUDM/glm-4-9b-chat", "chkhsh": "81d72c7348a9f0ebe86f23298d37debe0a5e71149e29bd283904c02262b27516"},
140140
{"name": "glm4", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/THUDM/glm-4-9b-hf", "chkhsh": "a1336059768a55c99a734006ffb02203cd450fed003e9a71886c88acf24fdbc2"},
141141
{"name": "minerva-7b", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", "chkhsh": "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35"},
142-
{"name": "hunyuan", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "chkhsh": "7e57df22b1fe23a7b1e1c7f3dc4e3f96d43a4eb0836d0c6bdc3436d7b2f1c664"},
142+
{"name": "hunyuan-v1-moe", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tencent/Hunyuan-A13B-Instruct", "chkhsh": "7e57df22b1fe23a7b1e1c7f3dc4e3f96d43a4eb0836d0c6bdc3436d7b2f1c664"},
143143
{"name": "hunyuan-v1-dense", "tokt": TOKENIZER_TYPE.BPE, "repo": "", "chkhsh": ""}, # TODO: update hunyuan-v1-dense repo
144144
# falcon-h1 series uses 4 different tokenizers across model sizes (0.5b - 34b), hence we need to define 4 different hashes
145145
{"name": "falcon-h1", "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/tiiuae/Falcon-H1-0.5B-Base", "chkhsh": "a6b57017d60e6edb4d88ecc2845188e0eb333a70357e45dcc9b53964a73bbae6"},

‎gguf-py/gguf/constants.py

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -372,7 +372,7 @@ class MODEL_ARCH(IntEnum):
372372
ARCEE = auto()
373373
ERNIE4_5 = auto()
374374
ERNIE4_5_MOE = auto()
375-
HUNYUAN_MOE = auto()
375+
HUNYUAN_V1_MOE = auto()
376376
HUNYUAN_V1_DENSE = auto()
377377
SMOLLM3 = auto()
378378
LFM2 = auto()
@@ -692,7 +692,7 @@ class MODEL_TENSOR(IntEnum):
692692
MODEL_ARCH.ERNIE4_5: "ernie4_5",
693693
MODEL_ARCH.ERNIE4_5_MOE: "ernie4_5-moe",
694694
MODEL_ARCH.FALCON_H1: "falcon-h1",
695-
MODEL_ARCH.HUNYUAN_MOE: "hunyuan-moe",
695+
MODEL_ARCH.HUNYUAN_V1_MOE: "hunyuan-v1-moe",
696696
MODEL_ARCH.HUNYUAN_V1_DENSE: "hunyuan-v1-dense",
697697
MODEL_ARCH.SMOLLM3: "smollm3",
698698
MODEL_ARCH.LFM2: "lfm2",
@@ -2430,7 +2430,7 @@ class MODEL_TENSOR(IntEnum):
24302430
MODEL_TENSOR.OUTPUT_NORM, # Final layer norm
24312431
MODEL_TENSOR.OUTPUT, # Output projection (lm_head)
24322432
],
2433-
MODEL_ARCH.HUNYUAN_MOE: [
2433+
MODEL_ARCH.HUNYUAN_V1_MOE: [
24342434
MODEL_TENSOR.TOKEN_EMBD,
24352435
MODEL_TENSOR.OUTPUT_NORM,
24362436
MODEL_TENSOR.OUTPUT,

‎src/llama-arch.cpp

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -84,7 +84,7 @@ static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
8484
{ LLM_ARCH_ARCEE, "arcee" },
8585
{ LLM_ARCH_ERNIE4_5, "ernie4_5" },
8686
{ LLM_ARCH_ERNIE4_5_MOE, "ernie4_5-moe" },
87-
{ LLM_ARCH_HUNYUAN_MOE, "hunyuan-moe" },
87+
{ LLM_ARCH_HUNYUAN_V1_MOE, "hunyuan-v1-moe" },
8888
{ LLM_ARCH_HUNYUAN_V1_DENSE, "hunyuan-v1-dense" },
8989
{ LLM_ARCH_SMOLLM3, "smollm3" },
9090
{ LLM_ARCH_LFM2, "lfm2" },
@@ -1874,7 +1874,7 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
18741874
},
18751875
},
18761876
{
1877-
LLM_ARCH_HUNYUAN_MOE,
1877+
LLM_ARCH_HUNYUAN_V1_MOE,
18781878
{
18791879
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
18801880
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },

‎src/llama-arch.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -88,7 +88,7 @@ enum llm_arch {
8888
LLM_ARCH_ARCEE,
8989
LLM_ARCH_ERNIE4_5,
9090
LLM_ARCH_ERNIE4_5_MOE,
91-
LLM_ARCH_HUNYUAN_MOE,
91+
LLM_ARCH_HUNYUAN_V1_MOE,
9292
LLM_ARCH_HUNYUAN_V1_DENSE,
9393
LLM_ARCH_SMOLLM3,
9494
LLM_ARCH_LFM2,

‎src/llama-chat.cpp

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -65,7 +65,7 @@ static const std::map<std::string, llm_chat_template> LLM_CHAT_TEMPLATES = {
6565
{ "bailing", LLM_CHAT_TEMPLATE_BAILING },
6666
{ "llama4", LLM_CHAT_TEMPLATE_LLAMA4 },
6767
{ "smolvlm", LLM_CHAT_TEMPLATE_SMOLVLM },
68-
{ "hunyuan-moe", LLM_CHAT_TEMPLATE_HUNYUAN_MOE },
68+
{ "hunyuan-v1-moe", LLM_CHAT_TEMPLATE_HUNYUAN_V1_MOE },
6969
{ "hunyuan-v1-dense", LLM_CHAT_TEMPLATE_HUNYUAN_V1_DENSE },
7070
{ "kimi-k2", LLM_CHAT_TEMPLATE_KIMI_K2 },
7171
};
@@ -193,7 +193,7 @@ llm_chat_template llm_chat_detect_template(const std::string & tmpl) {
193193
} else if (tmpl_contains("<|endofuserprompt|>")) {
194194
return LLM_CHAT_TEMPLATE_DOTS1;
195195
} else if (tmpl_contains("<|startoftext|>") && tmpl_contains("<|extra_4|>")) {
196-
return LLM_CHAT_TEMPLATE_HUNYUAN_MOE;
196+
return LLM_CHAT_TEMPLATE_HUNYUAN_V1_MOE;
197197
} else if (tmpl_contains("<｜hy_place▁holder▁no▁2｜>") && tmpl_contains("<｜hy_place▁holder▁no▁3｜>")) {
198198
return LLM_CHAT_TEMPLATE_HUNYUAN_V1_DENSE;
199199
} else if (tmpl_contains("<|im_assistant|>assistant<|im_middle|>")) {
@@ -694,14 +694,14 @@ int32_t llm_chat_apply_template(
694694
if (add_ass) {
695695
ss << "<|response|>";
696696
}
697-
} else if (tmpl == LLM_CHAT_TEMPLATE_HUNYUAN_MOE) {
697+
} else if (tmpl == LLM_CHAT_TEMPLATE_HUNYUAN_V1_MOE) {
698698
// tencent/Hunyuan-A13B-Instruct
699699
for (auto message : chat) {
700700
std::string role(message->role);
701701
if (role == "system") {
702702
ss << "<|startoftext|>" << message->content << "<|extra_4|>";
703703
} else if (role == "assistant") {
704-
ss << "<|startoftext|>" << message->content << "<|eos|>";
704+
ss << message->content << "<|eos|>";
705705
} else {
706706
ss << "<|startoftext|>" << message->content << "<|extra_0|>";
707707
}

‎src/llama-chat.h

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -45,7 +45,7 @@ enum llm_chat_template {
4545
LLM_CHAT_TEMPLATE_LLAMA4,
4646
LLM_CHAT_TEMPLATE_SMOLVLM,
4747
LLM_CHAT_TEMPLATE_DOTS1,
48-
LLM_CHAT_TEMPLATE_HUNYUAN_MOE,
48+
LLM_CHAT_TEMPLATE_HUNYUAN_V1_MOE,
4949
LLM_CHAT_TEMPLATE_HUNYUAN_V1_DENSE,
5050
LLM_CHAT_TEMPLATE_KIMI_K2,
5151
LLM_CHAT_TEMPLATE_UNKNOWN,

‎src/llama-model.cpp

Lines changed: 7 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -1733,7 +1733,7 @@ void llama_model::load_hparams(llama_model_loader & ml) {
17331733
type = LLM_TYPE_UNKNOWN;
17341734
}
17351735
} break;
1736-
case LLM_ARCH_HUNYUAN_MOE:
1736+
case LLM_ARCH_HUNYUAN_V1_MOE:
17371737
{
17381738
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);
17391739
ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH, hparams.n_ff_exp);
@@ -5078,7 +5078,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
50785078
layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {ffn_intermediate_size}, TENSOR_NOT_REQUIRED);
50795079
}
50805080
} break;
5081-
case LLM_ARCH_HUNYUAN_MOE:
5081+
case LLM_ARCH_HUNYUAN_V1_MOE:
50825082
{
50835083
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
50845084

@@ -16580,8 +16580,8 @@ struct llm_build_arcee : public llm_graph_context {
1658016580
}
1658116581
};
1658216582

16583-
struct llm_build_hunyuan_moe : public llm_graph_context {
16584-
llm_build_hunyuan_moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
16583+
struct llm_build_hunyuan_v1_moe : public llm_graph_context {
16584+
llm_build_hunyuan_v1_moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
1658516585
const int64_t n_embd_head = hparams.n_embd_head_v;
1658616586

1658716587
GGML_ASSERT(n_embd_head == hparams.n_embd_head_k);
@@ -17615,9 +17615,9 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
1761517615
{
1761617616
llm = std::make_unique<llm_build_ernie4_5_moe>(*this, params);
1761717617
} break;
17618-
case LLM_ARCH_HUNYUAN_MOE:
17618+
case LLM_ARCH_HUNYUAN_V1_MOE:
1761917619
{
17620-
llm = std::make_unique<llm_build_hunyuan_moe>(*this, params);
17620+
llm = std::make_unique<llm_build_hunyuan_v1_moe>(*this, params);
1762117621
} break;
1762217622
case LLM_ARCH_HUNYUAN_V1_DENSE:
1762317623
{
@@ -17831,7 +17831,7 @@ llama_rope_type llama_model_rope_type(const llama_model * model) {
1783117831
case LLM_ARCH_EXAONE4:
1783217832
case LLM_ARCH_MINICPM3:
1783317833
case LLM_ARCH_DOTS1:
17834-
case LLM_ARCH_HUNYUAN_MOE:
17834+
case LLM_ARCH_HUNYUAN_V1_MOE:
1783517835
case LLM_ARCH_HUNYUAN_V1_DENSE:
1783617836
case LLM_ARCH_LFM2:
1783717837
return LLAMA_ROPE_TYPE_NEOX;

0 commit comments

Comments
 (0)