Implementation:Unslothai Unsloth Push To Hub Merged

Knowledge Sources	Unsloth HuggingFace Hub
Domains	Model_Deployment, Model_Sharing
Last Updated	2026-02-07 00:00 GMT

Overview

Concrete tool for merging LoRA weights and uploading models to HuggingFace Hub provided by the Unsloth library.

Description

model.push_to_hub_merged merges LoRA adapters, saves the model locally to a temporary directory, then uploads all files to HuggingFace Hub. It also auto-generates a model card. A companion function model.push_to_hub_gguf handles GGUF-format uploads.

Usage

Call on a trained PeftModel after training. Requires a HuggingFace token. Pass private=True for private repositories.

Code Reference

Source Location

• Repository: unsloth
• File: unsloth/save.py
• Lines: L1378-1505 (push_to_hub_merged), L2060-2526 (push_to_hub_gguf), L1506-1594 (upload_to_huggingface helper)

Signature

def unsloth_push_to_hub_merged(
    self,
    repo_id: str,
    tokenizer = None,
    save_method: str = "merged_16bit",
    use_temp_dir: Optional[bool] = None,
    commit_message: Optional[str] = "Trained with Unsloth",
    private: Optional[bool] = None,
    token: Union[bool, str, None] = None,
    max_shard_size: Union[int, str, None] = "5GB",
    create_pr: bool = False,
    safe_serialization: bool = True,
    revision: str = None,
    commit_description: str = "Upload model trained with Unsloth 2x faster",
    tags: Optional[List[str]] = None,
    temporary_location: str = "_unsloth_temporary_saved_buffers",
    maximum_memory_usage: float = 0.75,
) -> None:
    """
    Merges LoRA weights and pushes to HuggingFace Hub.

    Args:
        repo_id: HuggingFace repo ID (e.g., "username/model-name").
        save_method: "merged_16bit", "merged_4bit", or "lora".
        private: Create private repository. Default None (public).
        token: HuggingFace auth token.
        commit_message: Git commit message for the upload.
    """

Import

# Called as a method on the model instance:
model.push_to_hub_merged("username/my-model", tokenizer=tokenizer, token="hf_xxx")

I/O Contract

Inputs

Name	Type	Required	Description
repo_id	str	Yes	HuggingFace Hub repository ID (user/model-name)
tokenizer	PreTrainedTokenizer	No	Tokenizer to upload alongside model
save_method	str	No	"merged_16bit", "merged_4bit", or "lora" (default: "merged_16bit")
private	bool	No	Create private repo (default: None/public)
token	str	No	HuggingFace authentication token

Outputs

Name	Type	Description
Hub repository	Remote	Model uploaded to HuggingFace Hub with model card, config, weights, tokenizer

Usage Examples

Push Merged Model

model.push_to_hub_merged(
    "myuser/llama-3.2-finetuned",
    tokenizer=tokenizer,
    save_method="merged_16bit",
    token="hf_your_token",
)

Push GGUF to Hub

model.push_to_hub_gguf(
    "myuser/llama-3.2-finetuned-GGUF",
    tokenizer=tokenizer,
    quantization_method=["q4_k_m", "q8_0"],
    token="hf_your_token",
)

Related Pages

Implements Principle

• Principle:Unslothai_Unsloth_Hub_Upload

Requires Environment

• Environment:Unslothai_Unsloth_Python_Transformers