aziz9788

qwen35-saudidraft-full-final-adapter

README

License: apache-2.0

aziz9788/qwen35-saudidraft-full-final-adapter

PEFT LoRA adapter only (not merged weights).

Table with columns: Field, Value
Field	Value
Base	`unsloth/Qwen3.5-4B`
Rank `r`	32
Saved `lora_alpha`	32
`use_rslora`	True

Continued from SaudF/qwen35-4b-sft-rebuild-v1 checkpoint-1650. The previous eval operating point overwrote 1650 lora_alpha to 22 at merge time.

On Vast this repo is downloaded and merged onto unsloth/Qwen3.5-4B. Two recipes are compared:

direct — Unsloth + this adapter at lora_alpha=22
stacked — Unsloth + checkpoint-1650 at alpha 22, then this adapter on top (saved alpha 32)

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Model Details

Model Provider

aziz9788

Model Tree

Base

unsloth/Qwen3.5-4B

Adapter

this model

Input Modalities

TextImageVideo

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Explore FriendliAI today

Get started Talk to an engineer