Remove deprecated Flash preview and legacy MTP
This commit is contained in:
@@ -109,8 +109,8 @@ struct Shape {
|
||||
original_context: u64,
|
||||
}
|
||||
|
||||
const FLASH: Shape = Shape {
|
||||
model: ModelChoice::DeepSeekV4Flash,
|
||||
const DEEPSEEK_BASE: Shape = Shape {
|
||||
model: ModelChoice::DeepSeekV4Flash0731,
|
||||
family: ModelFamily::DeepSeek,
|
||||
layers: 43,
|
||||
embd: 4096,
|
||||
@@ -152,10 +152,7 @@ const FLASH: Shape = Shape {
|
||||
original_context: 65_536,
|
||||
};
|
||||
|
||||
const FLASH_0731: Shape = Shape {
|
||||
model: ModelChoice::DeepSeekV4Flash0731,
|
||||
..FLASH
|
||||
};
|
||||
const FLASH_0731: Shape = Shape { ..DEEPSEEK_BASE };
|
||||
|
||||
const PRO: Shape = Shape {
|
||||
model: ModelChoice::DeepSeekV4Pro,
|
||||
@@ -168,7 +165,7 @@ const PRO: Shape = Shape {
|
||||
ff_expert: 3072,
|
||||
indexer_top_k: 1024,
|
||||
expert_weight_scale: 2.5,
|
||||
..FLASH
|
||||
..DEEPSEEK_BASE
|
||||
};
|
||||
|
||||
const GLM: Shape = Shape {
|
||||
@@ -236,7 +233,6 @@ impl Model {
|
||||
pub(crate) fn open(settings: &EngineSettings) -> Result<Self, String> {
|
||||
validate_engine_artifacts(
|
||||
settings.model,
|
||||
settings.artifacts.mtp.is_some() && !settings.speculative.dspark,
|
||||
settings.speculative.dspark,
|
||||
&settings.artifacts,
|
||||
)?;
|
||||
@@ -244,7 +240,7 @@ impl Model {
|
||||
if settings.execution.warm_weights {
|
||||
model.main.warm()?;
|
||||
}
|
||||
if let Some(path) = &settings.artifacts.mtp {
|
||||
if let Some(path) = &settings.artifacts.support {
|
||||
let support = Gguf::open(path)?;
|
||||
let kind = validate_support(&support, &model.shape)?;
|
||||
if settings.execution.warm_weights {
|
||||
@@ -1958,13 +1954,12 @@ mod sampling_tests {
|
||||
fn metal_executes_real_flash_token() {
|
||||
configure_metal_sources().unwrap();
|
||||
let path = crate::model::engine_artifacts(
|
||||
ModelChoice::DeepSeekV4Flash,
|
||||
false,
|
||||
ModelChoice::DeepSeekV4Flash0731,
|
||||
false,
|
||||
&crate::app::models_path(),
|
||||
)
|
||||
.model;
|
||||
let model = Model::open_main(&path, ModelChoice::DeepSeekV4Flash).unwrap();
|
||||
let model = Model::open_main(&path, ModelChoice::DeepSeekV4Flash0731).unwrap();
|
||||
let tokens = model.render_prompt(
|
||||
"You are a helpful assistant",
|
||||
"Hello",
|
||||
@@ -2008,7 +2003,7 @@ mod sampling_tests {
|
||||
.0;
|
||||
drop(executor);
|
||||
|
||||
let model = Model::open_main(&path, ModelChoice::DeepSeekV4Flash).unwrap();
|
||||
let model = Model::open_main(&path, ModelChoice::DeepSeekV4Flash0731).unwrap();
|
||||
let mut restored = metal::Executor::open(model, 32_768, false, 0).unwrap();
|
||||
assert!(restored.load_checkpoint(&checkpoint, &mut |_| {}).unwrap());
|
||||
assert_eq!(restored.position(), tokens.len() as u32);
|
||||
|
||||
Reference in New Issue
Block a user