mirror of
https://github.com/ollama/ollama.git
synced 2026-03-11 17:34:04 -05:00
33ee7168ba
* WIP - MLX backend with gemma3 * MLX: add cmake and go tag build toggles To build the new MLX backend code: cmake --preset MLX cmake --build --preset MLX --parallel cmake --install build --component MLX go build -tags mlx . Note: the main.go entrypoint for the MLX engine will change in a follow up commit. * add experimental image generation runtime * add experimental image generation runtime * MLX: wire up cuda build for linux * MLX: get dependencies correct and dedup This is still too large for a unified github artifact, but is now "correct" for the mlx_cuda_v13 directory. * fix relative link bug in dedup * Add darwin build and readme * add go build tag for mlx dependent code and wire up build_darwin.sh * lint cleanup * macos: build mlx for x86 This will be CPU only. * cuda build instructions and fix drift from mlx bump * stale comment * Delete agent helper doc * Clean up readme.md * Revise README for tokenizer clarity and details Updated README to clarify tokenizer functionality and removed correctness section. --------- Co-authored-by: jmorganca <jmorganca@gmail.com>
59 lines
1.5 KiB
Go
59 lines
1.5 KiB
Go
//go:build mlx
|
|
|
|
package gemma3
|
|
|
|
import (
|
|
"github.com/ollama/ollama/fs"
|
|
"github.com/ollama/ollama/x/ml"
|
|
"github.com/ollama/ollama/x/ml/nn"
|
|
"github.com/ollama/ollama/x/ml/nn/pooling"
|
|
"github.com/ollama/ollama/x/model"
|
|
"github.com/ollama/ollama/x/model/input"
|
|
)
|
|
|
|
type embedModel struct {
|
|
model.Base
|
|
model.SentencePiece
|
|
|
|
*TextModel
|
|
poolingType pooling.Type
|
|
|
|
Dense [2]*nn.Linear `gguf:"dense"`
|
|
}
|
|
|
|
func (m *embedModel) Forward(ctx ml.Context, batch input.Batch) (ml.Tensor, error) {
|
|
hiddenStates := m.TextModel.Forward(ctx, batch, m.Cache)
|
|
hiddenStates = m.poolingType.Forward(ctx, hiddenStates)
|
|
for _, dense := range m.Dense {
|
|
hiddenStates = dense.Forward(ctx, hiddenStates)
|
|
}
|
|
hiddenStates = hiddenStates.L2Norm(ctx, 1e-12)
|
|
return hiddenStates, nil
|
|
}
|
|
|
|
func newEmbedModel(c fs.Config) (model.Model, error) {
|
|
m := &embedModel{
|
|
SentencePiece: model.NewSentencePiece(
|
|
&model.Vocabulary{
|
|
Values: c.Strings("tokenizer.ggml.tokens"),
|
|
Scores: c.Floats("tokenizer.ggml.scores"),
|
|
Types: c.Ints("tokenizer.ggml.token_type"),
|
|
AddBOS: c.Bool("tokenizer.ggml.add_bos_token", true),
|
|
BOS: []int32{int32(c.Uint("tokenizer.ggml.bos_token_id"))},
|
|
AddEOS: c.Bool("tokenizer.ggml.add_eos_token", false),
|
|
EOS: append(
|
|
[]int32{
|
|
int32(c.Uint("tokenizer.ggml.eos_token_id")),
|
|
int32(c.Uint("tokenizer.ggml.eot_token_id", 106)),
|
|
},
|
|
c.Ints("tokenizer.ggml.eos_token_ids")...,
|
|
),
|
|
},
|
|
),
|
|
TextModel: newTextModel(c),
|
|
poolingType: pooling.Type(c.Uint("pooling_type", 0)),
|
|
}
|
|
|
|
return m, nil
|
|
}
|