mirror of
https://github.com/NousResearch/atropos.git
synced 2026-04-22 16:48:57 +00:00
[pre-commit.ci] auto fixes from pre-commit.com hooks
for more information, see https://pre-commit.ci
This commit is contained in:
parent
d2ea8cd612
commit
91afc9e46e
4 changed files with 91 additions and 65 deletions
|
|
@ -114,15 +114,17 @@ class TrainingConfig(BaseModel):
|
|||
wandb_group: Optional[str] = Field(None, description="Wandb group name")
|
||||
|
||||
# === Training Mode Configuration ===
|
||||
weight_bridge_mode: Literal["shared_vllm", "lora_only", "lora_restart", "none"] = Field(
|
||||
"none",
|
||||
description=(
|
||||
"How to synchronize weights with inference server. "
|
||||
"'shared_vllm': attach to vLLM's shared memory tensors and update in-place. "
|
||||
"'lora_only': keep base model frozen, train/swap LoRA adapters via HTTP (slow, needs --enforce-eager). "
|
||||
"'lora_restart': LoRA training with vLLM restarts (fast, CUDA graphs enabled). "
|
||||
"'none': legacy mode, restart vLLM with new checkpoint files."
|
||||
),
|
||||
weight_bridge_mode: Literal["shared_vllm", "lora_only", "lora_restart", "none"] = (
|
||||
Field(
|
||||
"none",
|
||||
description=(
|
||||
"How to synchronize weights with inference server. "
|
||||
"'shared_vllm': attach to vLLM's shared memory tensors and update in-place. "
|
||||
"'lora_only': keep base model frozen, train/swap LoRA adapters via HTTP (slow, needs --enforce-eager). "
|
||||
"'lora_restart': LoRA training with vLLM restarts (fast, CUDA graphs enabled). "
|
||||
"'none': legacy mode, restart vLLM with new checkpoint files."
|
||||
),
|
||||
)
|
||||
)
|
||||
train_layer_indices: Optional[List[int]] = Field(
|
||||
None,
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue