Skip to content

Commit 095a079

Browse files
gabe-l-hartgabe-l-hart
committed
feat: hparam and arch plumbing for granitemoeshared
Branch: GraniteMoEShared Signed-off-by: Gabe Goodhart <ghart@us.ibm.com>
1 parent 618fac6 commit 095a079

File tree

2 files changed

+90
-68
lines changed

2 files changed

+90
-68
lines changed

‎src/llama-arch.cpp

Lines changed: 89 additions & 68 deletions
Original file line numberDiff line numberDiff line change
@@ -5,74 +5,75 @@
55
#include <map>
66

77
static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
8-
{ LLM_ARCH_LLAMA, "llama" },
9-
{ LLM_ARCH_LLAMA4, "llama4" },
10-
{ LLM_ARCH_DECI, "deci" },
11-
{ LLM_ARCH_FALCON, "falcon" },
12-
{ LLM_ARCH_GROK, "grok" },
13-
{ LLM_ARCH_GPT2, "gpt2" },
14-
{ LLM_ARCH_GPTJ, "gptj" },
15-
{ LLM_ARCH_GPTNEOX, "gptneox" },
16-
{ LLM_ARCH_MPT, "mpt" },
17-
{ LLM_ARCH_BAICHUAN, "baichuan" },
18-
{ LLM_ARCH_STARCODER, "starcoder" },
19-
{ LLM_ARCH_REFACT, "refact" },
20-
{ LLM_ARCH_BERT, "bert" },
21-
{ LLM_ARCH_NOMIC_BERT, "nomic-bert" },
22-
{ LLM_ARCH_NOMIC_BERT_MOE, "nomic-bert-moe" },
23-
{ LLM_ARCH_JINA_BERT_V2, "jina-bert-v2" },
24-
{ LLM_ARCH_BLOOM, "bloom" },
25-
{ LLM_ARCH_STABLELM, "stablelm" },
26-
{ LLM_ARCH_QWEN, "qwen" },
27-
{ LLM_ARCH_QWEN2, "qwen2" },
28-
{ LLM_ARCH_QWEN2MOE, "qwen2moe" },
29-
{ LLM_ARCH_QWEN2VL, "qwen2vl" },
30-
{ LLM_ARCH_QWEN3, "qwen3" },
31-
{ LLM_ARCH_QWEN3MOE, "qwen3moe" },
32-
{ LLM_ARCH_PHI2, "phi2" },
33-
{ LLM_ARCH_PHI3, "phi3" },
34-
{ LLM_ARCH_PHIMOE, "phimoe" },
35-
{ LLM_ARCH_PLAMO, "plamo" },
36-
{ LLM_ARCH_CODESHELL, "codeshell" },
37-
{ LLM_ARCH_ORION, "orion" },
38-
{ LLM_ARCH_INTERNLM2, "internlm2" },
39-
{ LLM_ARCH_MINICPM, "minicpm" },
40-
{ LLM_ARCH_MINICPM3, "minicpm3" },
41-
{ LLM_ARCH_GEMMA, "gemma" },
42-
{ LLM_ARCH_GEMMA2, "gemma2" },
43-
{ LLM_ARCH_GEMMA3, "gemma3" },
44-
{ LLM_ARCH_STARCODER2, "starcoder2" },
45-
{ LLM_ARCH_MAMBA, "mamba" },
46-
{ LLM_ARCH_XVERSE, "xverse" },
47-
{ LLM_ARCH_COMMAND_R, "command-r" },
48-
{ LLM_ARCH_COHERE2, "cohere2" },
49-
{ LLM_ARCH_DBRX, "dbrx" },
50-
{ LLM_ARCH_OLMO, "olmo" },
51-
{ LLM_ARCH_OLMO2, "olmo2" },
52-
{ LLM_ARCH_OLMOE, "olmoe" },
53-
{ LLM_ARCH_OPENELM, "openelm" },
54-
{ LLM_ARCH_ARCTIC, "arctic" },
55-
{ LLM_ARCH_DEEPSEEK, "deepseek" },
56-
{ LLM_ARCH_DEEPSEEK2, "deepseek2" },
57-
{ LLM_ARCH_CHATGLM, "chatglm" },
58-
{ LLM_ARCH_GLM4, "glm4" },
59-
{ LLM_ARCH_BITNET, "bitnet" },
60-
{ LLM_ARCH_T5, "t5" },
61-
{ LLM_ARCH_T5ENCODER, "t5encoder" },
62-
{ LLM_ARCH_JAIS, "jais" },
63-
{ LLM_ARCH_NEMOTRON, "nemotron" },
64-
{ LLM_ARCH_EXAONE, "exaone" },
65-
{ LLM_ARCH_RWKV6, "rwkv6" },
66-
{ LLM_ARCH_RWKV6QWEN2, "rwkv6qwen2" },
67-
{ LLM_ARCH_RWKV7, "rwkv7" },
68-
{ LLM_ARCH_ARWKV7, "arwkv7" },
69-
{ LLM_ARCH_GRANITE, "granite" },
70-
{ LLM_ARCH_GRANITE_MOE, "granitemoe" },
71-
{ LLM_ARCH_CHAMELEON, "chameleon" },
72-
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
73-
{ LLM_ARCH_PLM, "plm" },
74-
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
75-
{ LLM_ARCH_UNKNOWN, "(unknown)" },
8+
{ LLM_ARCH_LLAMA, "llama" },
9+
{ LLM_ARCH_LLAMA4, "llama4" },
10+
{ LLM_ARCH_DECI, "deci" },
11+
{ LLM_ARCH_FALCON, "falcon" },
12+
{ LLM_ARCH_GROK, "grok" },
13+
{ LLM_ARCH_GPT2, "gpt2" },
14+
{ LLM_ARCH_GPTJ, "gptj" },
15+
{ LLM_ARCH_GPTNEOX, "gptneox" },
16+
{ LLM_ARCH_MPT, "mpt" },
17+
{ LLM_ARCH_BAICHUAN, "baichuan" },
18+
{ LLM_ARCH_STARCODER, "starcoder" },
19+
{ LLM_ARCH_REFACT, "refact" },
20+
{ LLM_ARCH_BERT, "bert" },
21+
{ LLM_ARCH_NOMIC_BERT, "nomic-bert" },
22+
{ LLM_ARCH_NOMIC_BERT_MOE, "nomic-bert-moe" },
23+
{ LLM_ARCH_JINA_BERT_V2, "jina-bert-v2" },
24+
{ LLM_ARCH_BLOOM, "bloom" },
25+
{ LLM_ARCH_STABLELM, "stablelm" },
26+
{ LLM_ARCH_QWEN, "qwen" },
27+
{ LLM_ARCH_QWEN2, "qwen2" },
28+
{ LLM_ARCH_QWEN2MOE, "qwen2moe" },
29+
{ LLM_ARCH_QWEN2VL, "qwen2vl" },
30+
{ LLM_ARCH_QWEN3, "qwen3" },
31+
{ LLM_ARCH_QWEN3MOE, "qwen3moe" },
32+
{ LLM_ARCH_PHI2, "phi2" },
33+
{ LLM_ARCH_PHI3, "phi3" },
34+
{ LLM_ARCH_PHIMOE, "phimoe" },
35+
{ LLM_ARCH_PLAMO, "plamo" },
36+
{ LLM_ARCH_CODESHELL, "codeshell" },
37+
{ LLM_ARCH_ORION, "orion" },
38+
{ LLM_ARCH_INTERNLM2, "internlm2" },
39+
{ LLM_ARCH_MINICPM, "minicpm" },
40+
{ LLM_ARCH_MINICPM3, "minicpm3" },
41+
{ LLM_ARCH_GEMMA, "gemma" },
42+
{ LLM_ARCH_GEMMA2, "gemma2" },
43+
{ LLM_ARCH_GEMMA3, "gemma3" },
44+
{ LLM_ARCH_STARCODER2, "starcoder2" },
45+
{ LLM_ARCH_MAMBA, "mamba" },
46+
{ LLM_ARCH_XVERSE, "xverse" },
47+
{ LLM_ARCH_COMMAND_R, "command-r" },
48+
{ LLM_ARCH_COHERE2, "cohere2" },
49+
{ LLM_ARCH_DBRX, "dbrx" },
50+
{ LLM_ARCH_OLMO, "olmo" },
51+
{ LLM_ARCH_OLMO2, "olmo2" },
52+
{ LLM_ARCH_OLMOE, "olmoe" },
53+
{ LLM_ARCH_OPENELM, "openelm" },
54+
{ LLM_ARCH_ARCTIC, "arctic" },
55+
{ LLM_ARCH_DEEPSEEK, "deepseek" },
56+
{ LLM_ARCH_DEEPSEEK2, "deepseek2" },
57+
{ LLM_ARCH_CHATGLM, "chatglm" },
58+
{ LLM_ARCH_GLM4, "glm4" },
59+
{ LLM_ARCH_BITNET, "bitnet" },
60+
{ LLM_ARCH_T5, "t5" },
61+
{ LLM_ARCH_T5ENCODER, "t5encoder" },
62+
{ LLM_ARCH_JAIS, "jais" },
63+
{ LLM_ARCH_NEMOTRON, "nemotron" },
64+
{ LLM_ARCH_EXAONE, "exaone" },
65+
{ LLM_ARCH_RWKV6, "rwkv6" },
66+
{ LLM_ARCH_RWKV6QWEN2, "rwkv6qwen2" },
67+
{ LLM_ARCH_RWKV7, "rwkv7" },
68+
{ LLM_ARCH_ARWKV7, "arwkv7" },
69+
{ LLM_ARCH_GRANITE, "granite" },
70+
{ LLM_ARCH_GRANITE_MOE, "granitemoe" },
71+
{ LLM_ARCH_GRANITE_MOE_SHARED, "granitemoeshared" },
72+
{ LLM_ARCH_CHAMELEON, "chameleon" },
73+
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
74+
{ LLM_ARCH_PLM, "plm" },
75+
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
76+
{ LLM_ARCH_UNKNOWN, "(unknown)" },
7677
};
7778

7879
static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
@@ -1483,6 +1484,26 @@ static const std::map<llm_arch, std::map<llm_tensor, const char *>> LLM_TENSOR_N
14831484
{ LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" },
14841485
},
14851486
},
1487+
{
1488+
LLM_ARCH_GRANITE_MOE_SHARED,
1489+
{
1490+
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
1491+
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },
1492+
{ LLM_TENSOR_OUTPUT, "output" },
1493+
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
1494+
{ LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" },
1495+
{ LLM_TENSOR_ATTN_K, "blk.%d.attn_k" },
1496+
{ LLM_TENSOR_ATTN_V, "blk.%d.attn_v" },
1497+
{ LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
1498+
{ LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" },
1499+
{ LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" },
1500+
{ LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" },
1501+
{ LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" },
1502+
{ LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" },
1503+
{ LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" },
1504+
{ LLM_TENSOR_FFN_UP_SHEXP, "blk.%d.ffn_up_shexp" },
1505+
},
1506+
},
14861507
{
14871508
LLM_ARCH_CHAMELEON,
14881509
{

‎src/llama-arch.h

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -72,6 +72,7 @@ enum llm_arch {
7272
LLM_ARCH_ARWKV7,
7373
LLM_ARCH_GRANITE,
7474
LLM_ARCH_GRANITE_MOE,
75+
LLM_ARCH_GRANITE_MOE_SHARED,
7576
LLM_ARCH_CHAMELEON,
7677
LLM_ARCH_WAVTOKENIZER_DEC,
7778
LLM_ARCH_PLM,

0 commit comments

Comments
 (0)