Skip to content

GitLab

Unverified Commit f6f37130 authored by Bruce MacDonald's avatar Bruce MacDonald Committed by GitHub
Browse files

convert: qwen2 from safetensors (#8408) 

Add native support for converting Qwen2 family models (including Qwen2.5)
from safetensors to gguf format so we can run it.
parent a30f3472
Hide whitespace changes
Inline Side-by-side
Showing with 397 additions and 0 deletions
convert/convert.go
View file @ f6f37130
......@@ -187,6 +187,8 @@ func ConvertModel(fsys fs.FS, ws io.WriteSeeker) error {
conv = &gemma2Model{}
case "Phi3ForCausalLM":
conv = &phi3Model{}
case "Qwen2ForCausalLM":
conv = &qwen2Model{}
case "BertModel":
conv = &bertModel{}
default:
......
convert/convert_qwen2.go 0 → 100644
View file @ f6f37130
package convert
import "github.com/ollama/ollama/llm"
type qwen2Model struct {
ModelParameters
MaxPositionEmbeddings uint32 `json:"max_position_embeddings"`
HiddenSize uint32 `json:"hidden_size"`
HiddenLayers uint32 `json:"num_hidden_layers"`
IntermediateSize uint32 `json:"intermediate_size"`
NumAttentionHeads uint32 `json:"num_attention_heads"`
NumKeyValueHeads uint32 `json:"num_key_value_heads"`
RopeTheta float32 `json:"rope_theta"`
RopeScaling struct {
Type string `json:"type"`
Factor ropeFactor `json:"factor"`
OriginalMaxPositionEmbeddings uint32 `json:"original_max_position_embeddings"`
} `json:"rope_scaling"`
RMSNormEPS float32 `json:"rms_norm_eps"`
}
var _ ModelConverter = (*qwen2Model)(nil)
func (q *qwen2Model) KV(t *Tokenizer) llm.KV {
kv := q.ModelParameters.KV(t)
kv["general.architecture"] = "qwen2"
kv["qwen2.block_count"] = q.HiddenLayers
kv["qwen2.context_length"] = q.MaxPositionEmbeddings
kv["qwen2.embedding_length"] = q.HiddenSize
kv["qwen2.feed_forward_length"] = q.IntermediateSize
kv["qwen2.attention.head_count"] = q.NumAttentionHeads
kv["qwen2.attention.head_count_kv"] = q.NumKeyValueHeads
kv["qwen2.rope.freq_base"] = q.RopeTheta
kv["qwen2.attention.layer_norm_rms_epsilon"] = q.RMSNormEPS
switch q.RopeScaling.Type {
case "":
// no scaling
case "yarn":
kv["qwen2.rope.scaling.type"] = q.RopeScaling.Type
kv["qwen2.rope.scaling.factor"] = q.RopeScaling.Factor
default:
panic("unknown rope scaling type")
}
return kv
}
func (q *qwen2Model) Tensors(ts []Tensor) []llm.Tensor {
var out []llm.Tensor
for _, t := range ts {
out = append(out, llm.Tensor{
Name: t.Name(),
Kind: t.Kind(),
Shape: t.Shape(),
WriterTo: t,
})
}
return out
}
func (p *qwen2Model) Replacements() []string {
return []string{
"lm_head", "output",
"model.embed_tokens", "token_embd",
"model.layers", "blk",
"input_layernorm", "attn_norm",
"self_attn.k_proj", "attn_k",
"self_attn.v_proj", "attn_v",
"self_attn.q_proj", "attn_q",
"self_attn.o_proj", "attn_output",
"mlp.down_proj", "ffn_down",
"mlp.gate_proj", "ffn_gate",
"mlp.up_proj", "ffn_up",
"post_attention_layernorm", "ffn_norm",
"model.norm", "output_norm",
}
}
convert/convert_test.go
View file @ f6f37130
......@@ -108,6 +108,7 @@ func TestConvertModel(t *testing.T) {
"Phi-3-mini-128k-instruct",
"all-MiniLM-L6-v2",
"gemma-2-9b-it",
"Qwen2.5-0.5B-Instruct",
}
for i := range cases {
......
convert/testdata/Qwen2.5-0.5B-Instruct.json 0 → 100644
View file @ f6f37130
{
"general.architecture": "qwen2",
"general.file_type": "1",
"general.parameter_count": "494032768",
"general.quantization_version": "2",
"output_norm.weight": "93a01a6db3419e85320a244bbf8ae81c43033b1d10c342bea3797ff2ce348390",
"qwen2.attention.head_count": "14",
"qwen2.attention.head_count_kv": "2",
"qwen2.attention.layer_norm_rms_epsilon": "1e-06",
"qwen2.block_count": "24",
"qwen2.context_length": "32768",
"qwen2.embedding_length": "896",
"qwen2.feed_forward_length": "4864",
"qwen2.rope.freq_base": "1e+06",
"token_embd.weight": "d74257dc547b48be5ae7b93f1c9af072c0c42dbbb85503078e25c59cd09e68d0",
"tokenizer.ggml.add_eos_token": "false",
"tokenizer.ggml.add_padding_token": "false",
"tokenizer.ggml.eos_token_id": "151645",
"tokenizer.ggml.merges": "6b1b1c58f1223d74f9095929d3e6416cdd74784440221a5507b87b8197f2bfd2",
"tokenizer.ggml.model": "gpt2",
"tokenizer.ggml.padding_token_id": "151643",
"tokenizer.ggml.pre": "qwen2",
"tokenizer.ggml.scores": "94e247e531e8b0fa3d248f3de09c9beae0c87da8106208a8edfaac0b8ec4b53d",
"tokenizer.ggml.token_type": "b178dbc9d1b2e08f84d02918e00fc2de2619a250e6c188c91a6605f701860055",
"tokenizer.ggml.tokens": "1d93f6679b23a1152b725f7f473792d54d53c1040c5250d3e46b42f81e0a1a34",
"blk.0.attn_k.bias": "5ce6617845f66c34515978d23d52e729c298d8bffa28c356a0428bef17142cf1",
"blk.0.attn_k.weight": "a960832a9e0e83e4d95402e5d1a01cc74300fcca0c381237162126330e1a7af8",
"blk.0.attn_norm.weight": "32c7d51cd0958f1f1771174192db341f9770516d7595a2f0fd18a4d78bd5aba3",
"blk.0.attn_output.weight": "c67e6e7e868354a11bf9121c70ee56c140b20eec611a8955e7dfe54a21d40a98",
"blk.0.attn_q.bias": "3e9e994eb1f03bccfc82f8bb3c324c920d42d547e07de5be83be12c428645063",
"blk.0.attn_q.weight": "dc12132f789b97cfa1e3f5775ceb835247fa67aa47400fd09c8f9f3769208583",
"blk.0.attn_v.bias": "a3fd0757b31fdc78af5ec320332d239c1a79d34e8804df06c5454e86955e8cc9",
"blk.0.attn_v.weight": "f43094a2134c7ee2dcc52aac3c8b7d9d64fb0295a8adb94cabfd49213f017b84",
"blk.0.ffn_down.weight": "18c2aec92db14f21976838a8c35d5575f80d0e4b1e05ccc0d8388d5877e80147",
"blk.0.ffn_gate.weight": "a3a1c4ef38f8f750eabadfe3d83bbb0f77941eec1cc1a388e51852e99c8691f6",
"blk.0.ffn_norm.weight": "b59b779c42d44b5c4cec41e39b4eb61e0491a07c1b3e946ccb5b8d5c657eda3f",
"blk.0.ffn_up.weight": "db64f09987ea59449e90abae5a2ffcc20efd9203f0eebec77a6aacb5809d6cff",
"blk.1.attn_k.bias": "a5c8c5671703ec0aa0143ff70a20ffdd67b5d5790ca1dfa5bba4e87e4071ed9f",
"blk.1.attn_k.weight": "835c7c7cc95b3cb2e55bd9cac585aa0760a033896621d3e06421f3378c540f7d",
"blk.1.attn_norm.weight": "f4c36fb6c14fce721fab0de78cc118d6f66e3a3d3ea0017bb14aade24c3c5434",
"blk.1.attn_output.weight": "cc1e80310c97cef068e48e40b7096f32fa2138519d6209c6a1a9994985999016",
"blk.1.attn_q.bias": "bc332780e66b0aac80ec5e63ac32344919a840db2fcc8f87bcef16a43a54138e",
"blk.1.attn_q.weight": "d766f06c925cce38d4b31b2165b3448e1fb49a7d561985f95d9cd2fcba52367a",
"blk.1.attn_v.bias": "9f486626fb6ed9ac84970a71e9b9818dd2758501fd3f61bb1c08540dcc7a8631",
"blk.1.attn_v.weight": "e873d1e5bd4f4d6abfd47c0f55119c2c111105838753ee273a03c5ccea25ce5c",
"blk.1.ffn_down.weight": "b3ce82b093f187344de04284b1783a452de1b72640914609b8f830dc81580521",
"blk.1.ffn_gate.weight": "5cd44ad237edaca525a28a3ac13975d1b565f576d6a8003237a341ae0d156f2e",
"blk.1.ffn_norm.weight": "4ac774ee8afaee119610c46aa1ff89fc6c9084a29d226075bc4aa4d2f15f746c",
"blk.1.ffn_up.weight": "042d81ab5f1983d85c81213232f3bfc05a9302d9dfaa98d931ebba326b6058b8",
"blk.10.attn_k.bias": "767ecfeacd60a2c2221ac4d76c357190849dd9cdf64ced418d9d0c7949101401",
"blk.10.attn_k.weight": "a9f3df343227537636be8202303453086375091944e498bad11e0b91e45e8c71",
"blk.10.attn_norm.weight": "01acd0e7b3e363f873dbfde6f0995ffcce83f5aaa10ff91c31dbf775035f6d5a",
"blk.10.attn_output.weight": "a531fe660769604ab869f01b203eb115e025cad4c0baeacdd1bcca99cf6d0264",
"blk.10.attn_q.bias": "356a02c9163dd660c1340fbe1e049b335ac6178891e00996131bba9ab4cb3e59",
"blk.10.attn_q.weight": "81be0cfb227339d83f954cd8dcf35828441211c6e1d184060e3eb76085041e2f",
"blk.10.attn_v.bias": "ed0450653284b62f8bf2c2db19c0ff7a6cf3cda1324d0a044c5e3db7bb692bd3",
"blk.10.attn_v.weight": "c1247ff7092babd2ed979883095b9aa022b2996cab1c77fb9e6176ddc1498d16",
"blk.10.ffn_down.weight": "fda7544965dc9af874f1062c22151c6cefc8ba08cbe15dc67aa89979e77b2de4",
"blk.10.ffn_gate.weight": "9f2632b1dee7304d10c70bd38d85bb1f148a628a8468f894f57975b8a2f1d945",
"blk.10.ffn_norm.weight": "94f8cbd6b17a4d5aabd93fa32930a687db3b11f086142f1cd71c535c11adcad4",
"blk.10.ffn_up.weight": "8dc2f8db0474939a277a3d89db34c3bcc3381cfea57bd05a8426a164634d9112",
"blk.11.attn_k.bias": "3b8e5a662b19411e3f6530714b766aad2ee41eebc8161bec9db0bc82d383a6e0",
"blk.11.attn_k.weight": "2c29f1ed1ce53ce9604e9ea3663c2c373157e909a0d6064a8920005f6d15dad9",
"blk.11.attn_norm.weight": "48f68a99c3da4ab4c9e492677b606d1b8e0e3de1fdbf6a977523f97b8c21ec31",
"blk.11.attn_output.weight": "5859f3838a94898b020c23040941ed88f4fcb132db400d0849f30a01f62c0f1c",
"blk.11.attn_q.bias": "c5ad89a5628f2bd81252ef44ef6bbcbff15c33ad16fba66435509b959c2af6d3",
"blk.11.attn_q.weight": "d102104e5d61c1e3219564f1d0149fd593db6c6daa9f3872460c84403323cfef",
"blk.11.attn_v.bias": "8653f7d48c5f75a5b55630819f99ecf01c932f12d33fd1a3ee634613e70edde8",
"blk.11.attn_v.weight": "e0a7c7d89b9f2d0d781ce85330022229126e130a8600a09d4a5f920f0bbd50b2",
"blk.11.ffn_down.weight": "4a22b3361eba8bbe1d9a6fda1812618e894c49f13bcacb505defa9badb6b96a6",
"blk.11.ffn_gate.weight": "484698b206760d3fd8df68b252a3c5bae65c8bf6392fb53a5261b021b6f39144",
"blk.11.ffn_norm.weight": "da69e96338cbe30882cf5a9544004387f5bbc0bcb6038e61ba2baabbd2623bac",
"blk.11.ffn_up.weight": "26ec74f1f504d1281715680dfbcc321db4e9900c53932fa40955daceb891b9aa",
"blk.12.attn_k.bias": "f94b49ec3e498f14f6bc3ebefe1f82018935bbe594df03253bfffae36bc20751",
"blk.12.attn_k.weight": "ae6323d0bbcfcea01f598d308993d1a7530317e78c1f64923e36d4b1649e9e73",
"blk.12.attn_norm.weight": "3784536a7611a839a42a29a5cc538c74ee4f9793092e5efe1b227b48f8c4d37f",
"blk.12.attn_output.weight": "46826c00b066829355db78293ab216e890f5eaaed3a70499ee68785189a6b0d9",
"blk.12.attn_q.bias": "b14db2d327ce0deec97beda7d3965a56c43e1e63dc9181840fb176b114cf643a",
"blk.12.attn_q.weight": "30f67df52ced06f76b6c85531657584276a454d6ec9bb7d0c7d2ca8f067f5551",
"blk.12.attn_v.bias": "57ab4b7e43f4fc5853bca7bfbb2702f8c2c391a49252a760abbb7b26330dc4aa",
"blk.12.attn_v.weight": "3ccd9da0cfe241cd33a63310f3ca6d81c5bc5a50d200bfea6612ac376166aca2",
"blk.12.ffn_down.weight": "a095774413198a83c549ce132d7c9684c0baef33145eaa889be370ef9c881c81",
"blk.12.ffn_gate.weight": "bb3b2bbdfb065d2a0a795909c53beec327781a4a7e974bf9f99c436cea459991",
"blk.12.ffn_norm.weight": "3b486c6cd97eb4b17967d9d6c0cc3821a1a6ad73d96b4d8fbf980101b32b8dab",
"blk.12.ffn_up.weight": "d020b82dd39a5d5a9d3881397bf53a567790a07f395284e6eb0f5fe0fef53de3",
"blk.13.attn_k.bias": "69381f8254586eba3623eceb18697fe79f9b4d8f2c30136acb10d5926e3ba1d0",
"blk.13.attn_k.weight": "c4d7a31495d71269f81b586203a50abea3a9e2985667faf258c9306ec6030f1d",
"blk.13.attn_norm.weight": "907da11075d16eda668dabe548af3cfd794df26b8ab53939af1344d91bec6fba",
"blk.13.attn_output.weight": "ca01cf6d2b8ece2fb3b0f56f1eb76194471ac27b54fe264f99c909f5eb7fef4a",
"blk.13.attn_q.bias": "2f5ecebafe03b1d485b93c41cff756ca57fb65b02e9d8336f14a3d26ab5d159a",
"blk.13.attn_q.weight": "f557f8acad7f0fa62da06b5da134182fe04a5bed8bdb269e316f970c9cc440fb",
"blk.13.attn_v.bias": "a492a88ae131e95714b092545a8752eaea7c7d2f9cb77852628ca8296c415525",
"blk.13.attn_v.weight": "d1220b1fe9f1cc0a5a88ee239d65fec900f5eaf6c448b6c2cbe74c81e15ed333",
"blk.13.ffn_down.weight": "53184e33440b49848a896304eb16a983efbc6b8bee0b93de8c8de716e1585fcb",
"blk.13.ffn_gate.weight": "684bf8896f148c851506c62717e45c426921b93c10d536ecdeb0fb28259a106d",
"blk.13.ffn_norm.weight": "6cb4e547ad8665eb7c174855c08afe1e5490fece66122522c1e9e8132d9064eb",
"blk.13.ffn_up.weight": "c64107897e38c06727075aba4ea7940b2cdd0e278b5c555dffb2790ef553bb57",
"blk.14.attn_k.bias": "2814ca9b160b16ae39557c9b629482fbe3a7592d372c1e1bf1ac59a2d578fde1",
"blk.14.attn_k.weight": "3377177396463afba667742972920ebb45dfdc37e9950e1f0e1d60a2f936b27d",
"blk.14.attn_norm.weight": "5cae870477d51dd35a6d22aaeacfce4dff218ffba693820ede6a4e11f02afd6d",
"blk.14.attn_output.weight": "3cfe9ccf3d48ae9e95b93a132a1c6240189a277d764f58590fb36fdbb714cad0",
"blk.14.attn_q.bias": "6a75acc2f090b2e67bfc26f7fca080ae8bd7c7aa090ec252e694be66b8b8f038",
"blk.14.attn_q.weight": "5ef45c86d7dda1df585aa1b827b89823adf679a6bb9c164bd0f97b2aa6eb96f1",
"blk.14.attn_v.bias": "5534480443e10ed72c31a917f3d104b0f49df5e6dbfa58d0eb5e7318120e3aee",
"blk.14.attn_v.weight": "58f45cf3240c4623626ec415c7d5441eaa8d2fb184f101aba973f222989422d1",
"blk.14.ffn_down.weight": "2dc82a0f20c05b77512458738130d8d05ce150cc078680ae7ee6dd7ed68d955d",
"blk.14.ffn_gate.weight": "d4a6c6f0fcccddfd1fcaa074846622f4a74cb22b9a654ab497abdc1d0dde9450",
"blk.14.ffn_norm.weight": "777e444932a0212ff3feac98442444e17bd8a98cb758ea3356697d0846d12c56",
"blk.14.ffn_up.weight": "6b75f6bd00195198447b69a417ed9d98f8ca28b3cb8be82f4bad908be0777d57",
"blk.15.attn_k.bias": "2d07211a58e6c2f23aa3a6dc03c80a7d135dfb28726b60b0e0fdd0f35ea5c37b",
"blk.15.attn_k.weight": "e77f3c0075a1810e70df956cc51fd08612f576cc09b6de8708dcae5daedb0739",
"blk.15.attn_norm.weight": "379a10d90609a5d5ba67d633803eda1424fc61ba5cca8d3bffe70c8b18b58ebf",
"blk.15.attn_output.weight": "402751c12ee9dbc9db5e3bf66a7b23ebe7d36c0500e0be67be4c8b1c4357fa62",
"blk.15.attn_q.bias": "acb37fc409ee725ceedf7a3a41b40106086abc47b76780728f781942c5120208",
"blk.15.attn_q.weight": "89cd3047a09b46ed2bb57c69dd687f67a1f0235149b30376fa31b525898e4a55",
"blk.15.attn_v.bias": "f081a37289cbe811978feb4da3ef543bdeb7355414d476f44e09b498da10cb2c",
"blk.15.attn_v.weight": "8404f242a11e6d512c9ead9b2f083cda031e9b269f8a0a83f57ee4c56934764e",
"blk.15.ffn_down.weight": "93438f43ee8cc4f1a7fd3840a6afdd5f02123e76db4f0d9474430c0100d148fc",
"blk.15.ffn_gate.weight": "ff935a2698843e87fad9dbf7125f53e460190ec71ee128b650b3fc027fe37bfc",
"blk.15.ffn_norm.weight": "4be80f199841cba831982e988451e1833c3c938a4d6ca1169319087bf0bd723e",
"blk.15.ffn_up.weight": "ee9ba63c66d71053e33551ddd519878bb30b88eeb03cfe047119c5c4000fb0a6",
"blk.16.attn_k.bias": "3f5fbabed4510c620b99d9d542739295fa6a262a7157f3a00a4889253f8341b8",
"blk.16.attn_k.weight": "8ca6eb139b281c257324cddea97a8e9aa7c048b53075cf00153123b967c27ee5",
"blk.16.attn_norm.weight": "290157f005e5aa7dddf4bd60100e7ee7b0baa7f11ec5c2cea5e0ead2aad3a4c6",
"blk.16.attn_output.weight": "b1f4d80a7447f08f1c331712527f750d00147f35c042442ade96fd029dadc5a1",
"blk.16.attn_q.bias": "e3e4e442ad4416791b468cad8de0d0d2d68c7e7df8d06002f4d49b4da9cb25e4",
"blk.16.attn_q.weight": "cc7392fa5bb1107d3816e7e7363de252d37efd4165d065e258806291ce0a147b",
"blk.16.attn_v.bias": "a7629830f2f6293e018916849614636d40b1bcd11245f75dbc34d38abae8f324",
"blk.16.attn_v.weight": "b6c7856c7d594437630929c8cf3b31d476e817875daf1095334ec08e40c5e355",
"blk.16.ffn_down.weight": "f9c0a777a00170990a4982d5a06717511bf9b0dd08aeaab64d9040d59bcbebba",
"blk.16.ffn_gate.weight": "ed88f11bc3176c9f22004e3559ccb9830a278b75edd05e11971d51c014bd5cd2",
"blk.16.ffn_norm.weight": "ab24abdcc4957895e434c6bb3a5237a71ff5044efb9f76c1a9e76e280c128410",
"blk.16.ffn_up.weight": "99f594dc8db37f554efa606e71d215fbc3907aa464a54038d6e40e9229a547ff",
"blk.17.attn_k.bias": "f236625676f9b2faa6781c7184d12d84c089c130d2a9350a6cf70210990f6bf1",
"blk.17.attn_k.weight": "c2a4f20cd3e98538308a13afe9cc5880bdd90d543449c6072dedd694b511ee1a",
"blk.17.attn_norm.weight": "5a9da4ee168311f487a79fc9d065a035432c6cafa8adb963a84954cf32f57a2a",
"blk.17.attn_output.weight": "d5df7031e354186ce65dc09d6f8a92eb721c0319816f8596b0c8a5d148ed0a2a",
"blk.17.attn_q.bias": "3212d5eeaa7ed7fac93cc99e16544de93c01bb681ae9391256ed4a8671fc6b00",
"blk.17.attn_q.weight": "d18cd9aa7ee10c551cb705549fa1ae974aea233f86471c9a19022dc29b63d0d5",
"blk.17.attn_v.bias": "a74ad11a1f8357742f80e2a0c0b3a2578fc8bbaf14c8223000767e07a5d79703",
"blk.17.attn_v.weight": "da18ac0e90884436a1cb0ad6a067f97a37f321b03c70b8b03bf481339fef5c80",
"blk.17.ffn_down.weight": "81a8a5d7a194fb53d976558e0347efbe9fdb1effffde9634c70162e1a20eff51",
"blk.17.ffn_gate.weight": "72870d83ab62f2dcd45f593924e291a45e4ae1b87f804b5b88aa34cfd76dd15e",
"blk.17.ffn_norm.weight": "cae39ac69b9bdaeefab7533796fdf11dbb7a4bdbdeed601e20f209503aafe008",
"blk.17.ffn_up.weight": "e7cb40b0842468507cec0e502bbed8a86428b51d439e3466bc12f44b2754e28f",
"blk.18.attn_k.bias": "8bfc02b94f9587aa125e2d8bbc2b15f0a5eb8f378d8b3e64a8150ae0a8ca3df2",
"blk.18.attn_k.weight": "434bc3b3332ea48afee890aa689eb458a75c50bc783492b0cbf64d42db40e8ad",
"blk.18.attn_norm.weight": "d6ffc09396c42a70d1f0e97d81113eee704d3bfc9eeae2bed022075a5dd08075",
"blk.18.attn_output.weight": "133f001f81f3b082468a7de67cb2e7a76508fce34bcc4dee7f0858e06eee082c",
"blk.18.attn_q.bias": "758d0e28bf5e660b3090aafb70e2a3191b4f3bb218d65e9139a086ceacaf599f",
"blk.18.attn_q.weight": "12d7b86fc1b09b9fa7f8b7ed43d8a410892cec8672d0c752f8346f6193343696",
"blk.18.attn_v.bias": "9efd15bab0519462431d6c6e8a5b7dd4e151dc449468097ee0ddca369c0ecc2e",
"blk.18.attn_v.weight": "f631231a79d4a2e9730fb2e386d8c18621eb3fb7900fbfdff5e6d52cc42db122",
"blk.18.ffn_down.weight": "874a2dddf456f3ab56b958b0860d71c8c680a6f89322c9bf6b2f32a113592300",
"blk.18.ffn_gate.weight": "4549ef8976c345a511df4a7133bdaf6fe387335f52dfd8a4605a8ae3f728c403",
"blk.18.ffn_norm.weight": "80c258a2536a860e19bfcbd9f29afa13214fbb4c34bde0d4da51287d354e9a59",
"blk.18.ffn_up.weight": "8b03308a581457a3c038b7a086f3cdf14941d7ad4107c4bd6d9d6b062fd00d73",
"blk.19.attn_k.bias": "e77f7b0c8e3e0a9b0d61918cd88371047752a1b02b1576936f4ec807d4d870ee",
"blk.19.attn_k.weight": "a2a318e93355230c0d0f95c441b080bf9c4914507255f363fb67a5e771d4d1e6",
"blk.19.attn_norm.weight": "9a4bdeb3970be21ac74a94c2c81eb36986533db81b78db6edec48d9802910d59",
"blk.19.attn_output.weight": "2369b103dd3947e2cef02b2669b405af5957fb3a7f9d0ff40646078c4b4317ad",
"blk.19.attn_q.bias": "e20bf427bef69059ae84a5d9f98f7d688489627f198fb6153def018ff9fd2e34",
"blk.19.attn_q.weight": "45a3bb3bdfd2f29dd76e5f78ddae73678b9a2a85dfaf609e460240ef5b7be2ad",
"blk.19.attn_v.bias": "a441f58a3e02ed86ee1819eefc9bd4e8b70d11b864a929d58a2c2ac0aeb8203d",
"blk.19.attn_v.weight": "30b0b04480c510450a7abb2ce9fa05c65b150a3cc4dc76f8916bf8d013f1b6be",
"blk.19.ffn_down.weight": "eebb9ab8fdb6a6efcfff8cf383adac9ec2d64aeeff703d16ed60d3621f86c395",
"blk.19.ffn_gate.weight": "3fef1493029298378886586478410b3d2e4e879f6aa83c07e210a7ce6481817f",
"blk.19.ffn_norm.weight": "e1be99ea1e8fb9678f7b8ba200f3f37e03878f3574d65d57bcd3a9fd796e2112",
"blk.19.ffn_up.weight": "f07cf25e09394fb69fe3ef324bdc0df9a4cecf3dc53070b8acc39e6d1689bf82",
"blk.2.attn_k.bias": "b29baa8221f125eff6b8ac1a950fa1d7cfc1bce7bdc636bf3df7d4065ab6466c",
"blk.2.attn_k.weight": "4bd0c179bced8bc37a09f5748c394e0cf50273942fb38a866e5cf50b6c96c437",
"blk.2.attn_norm.weight": "07b3edc6a6325c3428aa12f29bcae0be0de363ce61a6af487bc5c93fb8c468d9",
"blk.2.attn_output.weight": "056b5b31dbc81087c81b9d41c25960aa66c7190004c842ba343979644d7f4d88",
"blk.2.attn_q.bias": "479b6212401e097767c9d52b12a1adb8961c0fce9fcaaab81f202a9d85744376",
"blk.2.attn_q.weight": "f89196076f446a6dd8a9eee017f303504f9c03094c326449cee5a7fc0a97fade",
"blk.2.attn_v.bias": "ef9b1b986dbd9d7291027a88b67dc31434435b20e76e4f1e9d6273ebd31224f0",
"blk.2.attn_v.weight": "9322f4f00e85f8c0936845c51ca64b202a93df104f36886986a8452a8e4967a5",
"blk.2.ffn_down.weight": "7beac0d2440dc49af33ededb85a6cc3ba23ab33ad3ffa5760714b2ef84d94f6e",
"blk.2.ffn_gate.weight": "818a93864a5890c1f4dc66429004fad07645a50142350e9bff9a68fe24608a52",
"blk.2.ffn_norm.weight": "152c924d5514942ad274aafb8cc91b35c1db3627c3d973d92f60ff75f3daf9ba",
"blk.2.ffn_up.weight": "9c9579e600f209546db6015c9acfeda4f51b6d3cca6e8db4d20a04285fe61a37",
"blk.20.attn_k.bias": "fd22bfeffb63d818ce2ff1ea2ace0db5d940f7a9489b6bfc1ec4a5398848d7fe",
"blk.20.attn_k.weight": "f74439bc74c2f9252130c9c28384fd7352368b58bb7ce3f2444cf0288dfff861",
"blk.20.attn_norm.weight": "5c15d2613df87be6495fb7546b7dcedd2801d12fa5ecc02c877df889330e8f37",
"blk.20.attn_output.weight": "6731a39286a67f6859832f96695732e579e14e0c36956eccd1edce3db11595b8",
"blk.20.attn_q.bias": "04466e5a3f454a19b9b433fc2585396feac780027ece7ccb4e4bb3e406fc14d8",
"blk.20.attn_q.weight": "ead4c71daaeb17bf20d014a34c88b97f238456488e815ae0f281a5daf6fc99b8",
"blk.20.attn_v.bias": "adcc848e043025de9bd55ccb14dd8fb6343e8b5185ed07e12964be41d0faf99f",
"blk.20.attn_v.weight": "81bfc23f83526386a4761c2c16b6a93cd0bbf9d846c1a51b82c71f1474a465f1",
"blk.20.ffn_down.weight": "9bf660af3bafad919d03173c89a65fc9c89440a76c42c9e55e4d171076f3c17f",
"blk.20.ffn_gate.weight": "c04b4f3ccce44917ee228b998e2c19dd702aef10a43413afb152e808b5ac5c42",
"blk.20.ffn_norm.weight": "3d5b555d7746a71220143c6b8fff5ce4eb63283d9d9c772f1233d848f69f4ff4",
"blk.20.ffn_up.weight": "d7a196505c39e5469dfc7c6958bdbb54e93629ac1a047a6663ed96b318753094",
"blk.21.attn_k.bias": "4db1f48e5c6a3bc5720a5da813bbef08283e6269e12d83f8a9c54e52715d8011",
"blk.21.attn_k.weight": "c687b2f0e132a5e220a2a059b61aa2a537f37d8a674d7709f87880637b263b31",
"blk.21.attn_norm.weight": "ec23b0ff847a4b45585ab8e04f10fc20bb1637c5f1fbcdc4d73f336bcb5d1bd0",
"blk.21.attn_output.weight": "01255390576316c1731ef201e32c6e934eba356c28438cd06d9027ac6a3ff84f",
"blk.21.attn_q.bias": "3098f37205a15418e1681e407c82b7ce7c6fda6c6826b0590a13e1b68a38a1ea",
"blk.21.attn_q.weight": "30ea62cbb702a5359229dc96819df17ee535e2e9988d044b005c73ea536e1005",
"blk.21.attn_v.bias": "7bbedb2c22a04737f21993115701d4a06b985b7ca3b64681f53cd1be8d7ea39e",
"blk.21.attn_v.weight": "e11905e63579e36fbee978062af7599339ae29633765a4835628d79a795ec8df",
"blk.21.ffn_down.weight": "84def2ffd8aca766f9ce12ed9ac76919ab81eb34bdeae44fa4224417c38af527",
"blk.21.ffn_gate.weight": "4e99f05377b4a0b8d875045530a5c59dee6a46ac8a45597f6579f6fdfa800787",
"blk.21.ffn_norm.weight": "af48f13d03fba38ff8794a5f5005e666e501f971ca2e30bbded2777a8096f37d",
"blk.21.ffn_up.weight": "a29541c39a6acbc364be86994632a5bf55d701027cb7f23320f8c6d55ee42c91",
"blk.22.attn_k.bias": "c97f84db6c75422df6ef5768676d4e9abefaa3b8337aa2730ff260f8fc350480",
"blk.22.attn_k.weight": "af9a0c56f68779513e95be11611b7be6175ddae27d48bee9dd72fdbf05f6cbfa",
"blk.22.attn_norm.weight": "1c7518eb5bcff4a202c6f4a2827f14abd76f9bcc64ce75fe9db60b69437a5c9c",
"blk.22.attn_output.weight": "1abcf1f3caa2f59dd018646b93f9cf8fd30d49e98a473e6a8704419a751be46f",
"blk.22.attn_q.bias": "7221e01cb692faf2f7f8c2eb6e2fac38a1b751a9c9fdb6a21a0a936eb0bf4b96",
"blk.22.attn_q.weight": "faaf8fb7b6c19f343d47f3ea6b57151fb46c787e0b3bd2c292fd327d3d4d8e35",
"blk.22.attn_v.bias": "3ec05942e82d735de99dfd0d8228d8425e63e2fc584da98b3326bdef89ecb2e5",
"blk.22.attn_v.weight": "42e7b0ad06db76227837da9d4e74b2db97f3df4050ecb3a87cb9b55e08dfcb42",
"blk.22.ffn_down.weight": "87ef98ad2d0e824b0fa5ad8aa18787162922e527c9b1b721a99bc07d3bf97c82",
"blk.22.ffn_gate.weight": "562d6e5a1654b03aaa0e33864d23c10297fd4bcaa72d30fac69fb771ee1df9d6",
"blk.22.ffn_norm.weight": "f8a405dee467749d59427ce05cdd4b9c11bb18934a89258ea461f013b7d251f5",
"blk.22.ffn_up.weight": "90e1f4ae4062649d4d838399eb353e8bb8d56a49982b6a7f64aa3945377f7187",
"blk.23.attn_k.bias": "9ad22178a85f3be7e25f5aff462f31627466364f2f5e92f265cc91db0da9a8a8",
"blk.23.attn_k.weight": "d813beffb10f03278f5b58eea0f9d73cdcb7b5b4045ae025c379592e854f7dfd",
"blk.23.attn_norm.weight": "f583c9836044bdb056d6f8911088ac28add68e500043ae1f97b5d9158fe3d769",
"blk.23.attn_output.weight": "02789911ac3b97f6b761e958b7dd6dc7da61a46a1be92bd0b346039ca7ecd2b2",
"blk.23.attn_q.bias": "38c4970fb9b4f7e4a139258a45639d848653814b4bc89ea9849709b13f16414b",
"blk.23.attn_q.weight": "eb694be9a5ab5858b8dab064ee4cce247dc757424e65282989bd4d015b8580ce",
"blk.23.attn_v.bias": "0a25f6533aa7e7a152a4b198cf6c411c2408a34afa4f161bb4d5ffba2f74e33f",
"blk.23.attn_v.weight": "187e1bac6b70f74e6364de226565aa8275ee2854d09cbe5895451a689596049e",
"blk.23.ffn_down.weight": "88880dd9ba7ee80ade972927f810b5d2c30a69520c615190b27f9daabc0a8c5a",
"blk.23.ffn_gate.weight": "5abec63197935ab3eb8e6de0a5307396ec46cdb1cc5de25d87c845f3c4a3e887",
"blk.23.ffn_norm.weight": "60e1f5e6310c3a531c554a6bb7cd883aed58db1e51853f739436ea461c1843d7",
"blk.23.ffn_up.weight": "3d7f502771743f4a634188dfcd8b8a384fb07467ca8528366aee59ddb25b7bce",
"blk.3.attn_k.bias": "0b6b442ebbac29c8c4b67e8e3876d0382dd2dc52efdf4ab0ebbc6f71b6252393",
"blk.3.attn_k.weight": "480f40584fbda692c26f2cee45f5923780b236f8b4e8ec7bbee0237777a0918d",
"blk.3.attn_norm.weight": "39872be2af31bc9cd6b583ebba6fb759f621d586d66e5a2fc0b85991615a8923",
"blk.3.attn_output.weight": "924b2c80d8513bf637f8ebb3756a340d9cf2243de723fd08d7f5dccd46b3f8b6",
"blk.3.attn_q.bias": "863c9d848156847a3fe9bbc44415a4395245b5d13e95673c014fdb71e494ab0a",
"blk.3.attn_q.weight": "bff73ee5de92fba8f6c089bbb19ce57e17ab3c9c29295712804bb752711b882e",
"blk.3.attn_v.bias": "e1b6fea126e86189112fcdfee79ffc66a087461527bc9c2dc52dc80f3b7de95e",
"blk.3.attn_v.weight": "7812b7f5133636f06cdbb4dcc48ef7803206538641b6c960777b37f60a8e6752",
"blk.3.ffn_down.weight": "00b393d6a7e3ad9b5224211ccdbc54a96aae151f24ed631764ac224972a6bc82",
"blk.3.ffn_gate.weight": "cfd63fa3a038af05dc53c6eeb3c192f1602f26ff24cb840bcf1510fcb37b5513",
"blk.3.ffn_norm.weight": "7389fc240a282949580ea2f5b0d7973ac79f32f76dc0155b537bb6b751f8e27a",
"blk.3.ffn_up.weight": "2a945f47090df9cb16f92f1f06c520f156f8e232182eaaed09f257b8947a2a62",
"blk.4.attn_k.bias": "62533c31f0de498187593f238c6597503fef2a92e920cd540a96bc5311b3b2a0",
"blk.4.attn_k.weight": "93e829868bffd980a8e589b9c4566cd81e6ce4296a5f357a2ae93febe1284156",
"blk.4.attn_norm.weight": "9e0aaa4bbdd1389890f8abec20533f3ab16d61b872b1a8dbd623023921c660a9",
"blk.4.attn_output.weight": "74467d6f44357d67f452ac49da861468b38e98057017bd38bc9a449f9d3538e6",
"blk.4.attn_q.bias": "8e6d9026fd69b314c1773c5946be2e11daf806ef22a5d91d744344fd30c58c59",
"blk.4.attn_q.weight": "e5bfbafd94a4d530f3769f5edbba8cc08d9b5bee8f66ebf4cb54e69bc0b7f63b",
"blk.4.attn_v.bias": "20c570f92022d9905eb85c0e41d1fdb30db22007a9628b51f512f8268d6c34a2",
"blk.4.attn_v.weight": "9638d459d61da03c9dd34dad985e03c43b4f8a5bc9701a82153478329b0517e0",
"blk.4.ffn_down.weight": "9d91b06e89d52f4365dece7eaeec50f81e52cb2407b333248a81e6e2f84c05b8",
"blk.4.ffn_gate.weight": "bf6350a79c6a6ee9146edfd788b88d4a4c2b54db1aa0adcc1464dbba8a84b646",
"blk.4.ffn_norm.weight": "11a70a6b9f7ce336292f4e3a2c6c92d366d4ee4306ad4fdb1870fde107e9cc31",
"blk.4.ffn_up.weight": "64f23f493d02b147a72a59605e6b7dd1c4c74f6813a38a2a60818bd66f697347",
"blk.5.attn_k.bias": "f6c2c279c0ed686f298ad1e5514b5cd882199341f896abbb2c2129d4c64ce9c5",
"blk.5.attn_k.weight": "0e682f75870abf9efaca10dac5f04c580f42820ecf4e234d43af967019acb86f",
"blk.5.attn_norm.weight": "01efae7653705e741932fcd79dff3be643d7e97f4b5719b887835dffe44b3a82",
"blk.5.attn_output.weight": "69e841d00d196acc489cd70bc5ffbbb63530ac5fabb169d40c4fb3a32ebb8ed8",
"blk.5.attn_q.bias": "f3304d76ccd44fed887565857c8e513b1211d89a5d3e81782de507ab3f6fc045",
"blk.5.attn_q.weight": "98612a6b7920a247853ada95c240807d4ca8e43604279e7a2fc9bb265ae40469",
"blk.5.attn_v.bias": "39940a9b353ceed3edfd4a39b985c9520490aa1b9f11749c94fdf6d879d1a259",
"blk.5.attn_v.weight": "839f84b828cf83aecf479a0dc7bc86cce05145ef77dcf29916dc3e0680f5b665",
"blk.5.ffn_down.weight": "1f48cbb0960f15e06ab8a3754ade792995a655856389ddbca629c07e89d1b114",
"blk.5.ffn_gate.weight": "33d8219fce3189e1aab376039896eebd4ad36ebd26a8278cd19b26e4357e4f81",
"blk.5.ffn_norm.weight": "0f4a0f83d37127fa4483f2905cb4f38ef6ddc71584b6cb05632c62a9af313dda",
"blk.5.ffn_up.weight": "22a64a11e5f0a1ff45ca327bf9e1efa258f085ff6a96edc398b7474f725b4514",
"blk.6.attn_k.bias": "baa91df99d4df2d25e8d590bca4e334b97f2d9aa3df8e748fedc8a6188499111",
"blk.6.attn_k.weight": "121f3b9f4b9491996499392e2688a929cafe102a67920b4cb2a039349c43d8eb",
"blk.6.attn_norm.weight": "b4cf987e923d71f2f84c58d20ea8af7576b225bf61952145b489fdd395e3d411",
"blk.6.attn_output.weight": "a112642150a138d54b2a4038042fd33619035a35694771e966f3575856c635d6",
"blk.6.attn_q.bias": "a97ea10469cdfa3fdddf8bad6de683ef99f6170eb8d29d15dcf6bf4bce37c5a3",
"blk.6.attn_q.weight": "d80c787019317a87361de6bbc7df6701357216bdd9b404522cede34a719a5500",
"blk.6.attn_v.bias": "d846269db9cd77ae28da26ba0914cace1b6754bd5301af9c44607085dfcbd2d7",
"blk.6.attn_v.weight": "06567c433e8a391647633291b50828a076ad7c2436106bb9278c60a3f8fccb3b",
"blk.6.ffn_down.weight": "f15f66f56b3c474eac8c6315c5fff07c3e29c6e483d7efd4d303c7f43814be91",
"blk.6.ffn_gate.weight": "47768f89c6da8eefb29adb766ff4eb38c9dfd79320bbc1386248319fcbcf567f",
"blk.6.ffn_norm.weight": "7f8195e6b148212967145fc9d86ce36b699cff0de026042245c2d344f1ef8510",
"blk.6.ffn_up.weight": "53d7707ae4347aadb445289f9f87a008b72df5cb855b00080a605442fdd8edf3",
"blk.7.attn_k.bias": "63e274df3217dde25b8369a383e480fe4f6b403a74385f15ac0b5db71dce2744",
"blk.7.attn_k.weight": "f6fce88602f5945eee09767acbcad387d132614e6da39ae359f2bbf380d94b1f",
"blk.7.attn_norm.weight": "bbf5dc7336c0f9a511afef6bf5efeffd78f1b83940850c3eb7eb20c621b75656",
"blk.7.attn_output.weight": "d9fb907a138396a859cecbfcb377927308dc93c24c7fb52dba5eb59265feadec",
"blk.7.attn_q.bias": "f02ba1318346af77e309f40aee716e2de7ee8cab67e67b17636db9bf40894fb0",
"blk.7.attn_q.weight": "54a691e824be287a61c35c172edc01922ed792d2addeee029afc17ba6c7e11b9",
"blk.7.attn_v.bias": "3a4f182f51e84ce862d558fb2751b91802b65d74596bb14d624808513a8a83ec",
"blk.7.attn_v.weight": "a142fe6e106d3ab484e2dc6f9c72b8fc0a385279dde08deb1ad1fd05ac25deb1",
"blk.7.ffn_down.weight": "8daf7e8c430d183a4d6ab3eb575fafa4b5e31689f68b290c8b370411ad9d0f12",
"blk.7.ffn_gate.weight": "a2a786b45eb660994254b48e2aaf22f3e9821cfb383dee0ba04cc4350a2f8e72",
"blk.7.ffn_norm.weight": "73828bbc8c9610cc139fcf03e96272648cdc291263251fe3a67367408deb69e1",
"blk.7.ffn_up.weight": "e85dd0f63fed449ce16893c5795ea6a050a2d7a66d9534410a227e22c905dafa",
"blk.8.attn_k.bias": "91a752a6e2c364e5ee6a015770fe289aece4911ae6c6bbfe74ac52f465465f93",
"blk.8.attn_k.weight": "99c069e92c43a2efb74e23188256b3cabbbe06399878e681ce203a05d5da378a",
"blk.8.attn_norm.weight": "c76d36d3cc06aa2a9edb1abf9f602bb7ed61ac9d61f8ef7ed736a1e619abe717",
"blk.8.attn_output.weight": "ee5ff156a2625e1f203f65e69b514f9df04bd9a5e82b28e3876e16cf1c6f65c5",
"blk.8.attn_q.bias": "8fbd868a93b330c8b0418b488c5301f42a7eb0c58445a4e515d56777f1d96ed5",
"blk.8.attn_q.weight": "9f20ef86e80098ba52a3a31ebcc315bea3a614dac9cba7ac1db02f156db9b577",
"blk.8.attn_v.bias": "c4813571d5d618742183a7890c0b89cd7f18e210c758f63aad564659bc38a26d",
"blk.8.attn_v.weight": "ea88e1a4cf8bd56e9a88ada427d2b0cd352234827640757ee2a9ed594fb67a53",
"blk.8.ffn_down.weight": "b0d1a7495811580b189aaa3e20ea871d6d01ed7b6c23e59825078ef786944ff2",
"blk.8.ffn_gate.weight": "0a17c0caa0b06721c49b59b2a63a5dcbf744dd1cffa55962b404ba910c658a62",
"blk.8.ffn_norm.weight": "f15f109d4a8e9d1ff7c71fa5bc6373df7ee80c5f7d1de3fa0d4849d747e36bcb",
"blk.8.ffn_up.weight": "bbf4c5c4c5c8a0f9ae8b88e3cc8b86f81b98148722d5a350995af176c0b774f2",
"blk.9.attn_k.bias": "a7f60d962686b8ca60f69643e0e0fa8614688be738fb0b1c6bd54de35c2beb5e",
"blk.9.attn_k.weight": "dd80ce4adb00e338fc04b307e4c18a27071f4ba4397184a24d765e6e4a268ef4",
"blk.9.attn_norm.weight": "721e6487547e2b3986ab4b4e2500ceade59d908bccf4436e1e8031f246deb2bd",
"blk.9.attn_output.weight": "5a800af39107b363861e5f5173483cdcd644d8ac3b0c8a443b9c759d71285db8",
"blk.9.attn_q.bias": "0a19b4925ea8ca8067acc909b058adc327de3874cfc94cc9eb4a106d3f370123",
"blk.9.attn_q.weight": "93e84906684c0c7ede79967236d9fc8344da84a9f1daa04e8295c2c9b6b26a24",
"blk.9.attn_v.bias": "615421f812f821e230ecde4e6da35d868823248355ce7e4e51e2d650ead565f9",
"blk.9.attn_v.weight": "7f4913e289aefd9ceecbdaf9767b1e95303f5d59dd67ecb2cc15768477f4d08e",
"blk.9.ffn_down.weight": "95d1b3933221e87dc4af70dd566daec9498bf358070b8d26f1fc70766a84a152",
"blk.9.ffn_gate.weight": "530f2d04f6a1fbffaaa5f2fbc3a328ebed7b330e3af14b4fc7d8a51b13ad8d42",
"blk.9.ffn_norm.weight": "28077de416217ea1df94b96017bef4cc562ab62e51b1a03a671c70abc29ce52a",
"blk.9.ffn_up.weight": "b87b6190778aaee4695938e24ac6c90dbbee6dce7c5c2ab5bc26ba4564581822"
}
convert/tokenizer.go
View file @ f6f37130
......@@ -100,6 +100,8 @@ func parseTokenizer(fsys fs.FS, specialTokenTypes []string) (*Tokenizer, error)
t.Pre = "deepseek-llm"
case "21cde974d587f0d54dc8d56b183cc1e6239600172035c68fbd6d4b9f8da0576e":
t.Pre = "deepseek-coder"
case "1ff7f41064896984db5d1bb6ff64fa4bc29007d08c1b439e505b7392777a319e":
t.Pre = "qwen2"
case "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855":
// noop, empty pretokenizer
default:
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment