mirror of
https://github.com/ollama/ollama.git
synced 2026-11-10 15:02:13 -06:00
a00d083bb5
* mlxrunner: restore System One scoring
Carry the MLX implementation and its tests from the original System One
commit. The API/compiler and llama-server implementation already exist on
the release branch, so retain those versions. Adapt the optional model test
to the renamed decision package and its explicit rendering/context setup.
(cherry picked from commit 26e546ddb3)
* Enable MLX System One and add regression coverage
* mlxrunner: add Laya support for System One
Import Laya checkpoints and implement bidirectional decision scoring.
* mlxrunner: release cached buffers after System One scoring
* decision: match Tev1 and Nimble publisher prompt formats
Select Tev1 prompts using renderer metadata and fix JSON formatting, with regression coverage for GGUF and MLX.
* mlx: support Clef and Clef Flash decision models
Add model import, quantization safeguards, and text/image scoring with the Clef joint decision head.
* support vision capabilities for decision models
* Make Clef scoring cancellable between prefill chunks
Preserve joint-head inputs and test cancellation, cleanup, and recovery.
* mlx: separate decision preparation and readout from forward
Let serial and batched scoring share model-owned preparation and readout
without duplicating architecture-specific logic in the runner.
* mlx: add Strands Decider support
* implement caching
---------
Co-authored-by: ParthSareen <parth.sareen@ollama.com>
51 lines
1.9 KiB
Go
51 lines
1.9 KiB
Go
package create
|
|
|
|
import (
|
|
"encoding/json"
|
|
"fmt"
|
|
"os"
|
|
"path/filepath"
|
|
)
|
|
|
|
// Laya ships an encoder config and a decision config rather than a root HF
|
|
// config. Keep both verbatim and add only the descriptor Ollama uses to dispatch.
|
|
func readLayaConfig(dir string) (sourceModelConfig, json.RawMessage, error) {
|
|
var decision struct {
|
|
Encoder string `json:"encoder"`
|
|
MaxLen int `json:"max_len"`
|
|
HeadLayers *int `json:"head_layers"`
|
|
}
|
|
data, err := os.ReadFile(filepath.Join(dir, "rl_agent_config.json"))
|
|
if err != nil {
|
|
return sourceModelConfig{}, nil, err
|
|
}
|
|
if err := json.Unmarshal(data, &decision); err != nil {
|
|
return sourceModelConfig{}, nil, fmt.Errorf("parse rl_agent_config.json: %w", err)
|
|
}
|
|
var encoder struct {
|
|
ModelType string `json:"model_type"`
|
|
HiddenSize int `json:"hidden_size"`
|
|
NumHiddenLayers int `json:"num_hidden_layers"`
|
|
NumAttentionHeads int `json:"num_attention_heads"`
|
|
VocabSize int `json:"vocab_size"`
|
|
}
|
|
data, err = os.ReadFile(filepath.Join(dir, "encoder", "config.json"))
|
|
if err != nil {
|
|
return sourceModelConfig{}, nil, err
|
|
}
|
|
if err := json.Unmarshal(data, &encoder); err != nil {
|
|
return sourceModelConfig{}, nil, fmt.Errorf("parse encoder/config.json: %w", err)
|
|
}
|
|
if decision.Encoder == "" || decision.MaxLen <= 0 || decision.HeadLayers == nil || encoder.ModelType != "modernbert" {
|
|
return sourceModelConfig{}, nil, fmt.Errorf("unsupported Laya source configuration")
|
|
}
|
|
cfg := sourceModelConfig{ModelType: "laya", Architectures: []string{"LayaForDecision"}}
|
|
data, err = json.Marshal(map[string]any{
|
|
"architectures": cfg.Architectures, "model_type": cfg.ModelType,
|
|
"hidden_size": encoder.HiddenSize, "num_hidden_layers": encoder.NumHiddenLayers,
|
|
"num_attention_heads": encoder.NumAttentionHeads, "vocab_size": encoder.VocabSize,
|
|
"max_position_embeddings": decision.MaxLen,
|
|
})
|
|
return cfg, data, err
|
|
}
|