2024-04-01 23:14:53 +00:00
|
|
|
package convert
|
|
|
|
|
|
|
|
import (
|
|
|
|
"fmt"
|
|
|
|
"io"
|
|
|
|
"log/slog"
|
|
|
|
"strings"
|
|
|
|
|
|
|
|
"github.com/pdevine/tensor"
|
|
|
|
"github.com/pdevine/tensor/native"
|
|
|
|
|
|
|
|
"github.com/ollama/ollama/llm"
|
|
|
|
)
|
|
|
|
|
|
|
|
type GemmaModel struct {
|
|
|
|
ModelData
|
|
|
|
}
|
|
|
|
|
|
|
|
func addOnes(data []float32, vectorSize int) ([]float32, error) {
|
|
|
|
n := tensor.New(tensor.WithShape(vectorSize), tensor.WithBacking(data))
|
|
|
|
ones := tensor.Ones(tensor.Float32, vectorSize)
|
|
|
|
|
2024-05-17 19:11:49 +00:00
|
|
|
n, err := n.Add(ones)
|
2024-04-01 23:14:53 +00:00
|
|
|
if err != nil {
|
2024-05-17 19:11:49 +00:00
|
|
|
return nil, err
|
2024-04-01 23:14:53 +00:00
|
|
|
}
|
|
|
|
|
2024-05-17 19:11:49 +00:00
|
|
|
ts, err := native.SelectF32(n, 0)
|
2024-04-01 23:14:53 +00:00
|
|
|
if err != nil {
|
2024-05-17 19:11:49 +00:00
|
|
|
return nil, err
|
2024-04-01 23:14:53 +00:00
|
|
|
}
|
|
|
|
|
2024-05-17 19:11:49 +00:00
|
|
|
var f32s []float32
|
|
|
|
for _, t := range ts {
|
|
|
|
f32s = append(f32s, t...)
|
2024-04-01 23:14:53 +00:00
|
|
|
}
|
|
|
|
|
2024-05-17 19:11:49 +00:00
|
|
|
|
|
|
|
return f32s, nil
|
2024-04-01 23:14:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
func (m *GemmaModel) GetTensors() error {
|
2024-04-15 18:26:42 +00:00
|
|
|
t, err := m.Format.GetTensors(m.Path, m.Params)
|
2024-04-01 23:14:53 +00:00
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
|
2024-04-15 18:26:42 +00:00
|
|
|
slog.Debug(fmt.Sprintf("Total tensors: %d", len(t)))
|
2024-04-01 23:14:53 +00:00
|
|
|
for _, l := range t {
|
|
|
|
if strings.HasSuffix(l.Name, "norm.weight") {
|
|
|
|
wt := l.WriterTo.(safetensorWriterTo)
|
2024-05-17 19:11:49 +00:00
|
|
|
wt.repacker = m.Repack
|
2024-04-01 23:14:53 +00:00
|
|
|
l.WriterTo = wt
|
|
|
|
}
|
|
|
|
m.Tensors = append(m.Tensors, l)
|
|
|
|
}
|
|
|
|
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
|
|
|
func (m *GemmaModel) LoadVocab() error {
|
2024-04-15 18:26:42 +00:00
|
|
|
v, err := LoadSentencePieceTokens(m.Path, m.Params)
|
2024-04-01 23:14:53 +00:00
|
|
|
if err != nil {
|
|
|
|
return err
|
|
|
|
}
|
|
|
|
m.Vocab = v
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
|
2024-05-17 19:11:49 +00:00
|
|
|
func (m *GemmaModel) Repack(_ string, data []float32, shape []uint64) ([]float32, error) {
|
|
|
|
return addOnes(data, int(shape[0]))
|
|
|
|
}
|
|
|
|
|
2024-04-12 20:55:12 +00:00
|
|
|
func (m *GemmaModel) WriteGGUF(ws io.WriteSeeker) error {
|
2024-04-01 23:14:53 +00:00
|
|
|
kv := llm.KV{
|
|
|
|
"general.architecture": "gemma",
|
|
|
|
"general.name": m.Name,
|
|
|
|
"gemma.context_length": uint32(m.Params.ContextSize),
|
|
|
|
"gemma.embedding_length": uint32(m.Params.HiddenSize),
|
|
|
|
"gemma.block_count": uint32(m.Params.HiddenLayers),
|
|
|
|
"gemma.feed_forward_length": uint32(m.Params.IntermediateSize),
|
|
|
|
"gemma.attention.head_count": uint32(m.Params.AttentionHeads),
|
|
|
|
"gemma.attention.head_count_kv": uint32(m.Params.KeyValHeads),
|
|
|
|
"gemma.attention.layer_norm_rms_epsilon": float32(m.Params.NormEPS),
|
|
|
|
"gemma.attention.key_length": uint32(m.Params.HeadDimension),
|
|
|
|
"gemma.attention.value_length": uint32(m.Params.HeadDimension),
|
|
|
|
"general.file_type": uint32(1),
|
|
|
|
"tokenizer.ggml.model": "llama",
|
|
|
|
|
|
|
|
"tokenizer.ggml.tokens": m.Vocab.Tokens,
|
|
|
|
"tokenizer.ggml.scores": m.Vocab.Scores,
|
|
|
|
"tokenizer.ggml.token_type": m.Vocab.Types,
|
|
|
|
|
|
|
|
"tokenizer.ggml.bos_token_id": uint32(m.Params.BoSTokenID),
|
|
|
|
"tokenizer.ggml.eos_token_id": uint32(m.Params.EoSTokenID),
|
|
|
|
"tokenizer.ggml.padding_token_id": uint32(m.Params.PaddingTokenID),
|
|
|
|
"tokenizer.ggml.unknown_token_id": uint32(3),
|
|
|
|
"tokenizer.ggml.add_bos_token": true,
|
|
|
|
"tokenizer.ggml.add_eos_token": false,
|
|
|
|
}
|
|
|
|
|
2024-04-12 20:55:12 +00:00
|
|
|
return llm.NewGGUFV3(m.Params.ByteOrder).Encode(ws, kv, m.Tensors)
|
2024-04-01 23:14:53 +00:00
|
|
|
}
|