A LoRA fine-tune of Qwen/Qwen3.8-27B on the digitalpipelines/samantha-1.1-uncensored dataset, giving the model Samantha's warm, empathetic, emotionally-present conversational persona.

What's in this repo

File	Description
`adapter_model.safetensors`	The LoRA adapter (PEFT format), ~304 MB
`adapter_config.json`	PEFT config (r=16, α=32, all-linear)
`Qwen3.8-27B-Samantha-LoRA-qwen35.gguf`	LoRA adapter in GGUF, arch-tagged `qwen35` for llama.cpp
`tokenizer*.json`, `chat_template.jinja`	Tokenizer + chat template

Related repositories

Repo	What it is
Lathly/Qwen3.8-27B-Samantha	The full merged model (bf16 safetensors) — this adapter baked into the base weights, ready to run with Transformers / vLLM / SGLang
Lathly/Qwen3.8-27B-Samantha-GGUF	GGUF builds for llama.cpp — f16 (for rolling your own quants) + Q5_K_M (tested production build) + mmproj for multimodal

If you want the persona without managing an adapter, grab the merged model or a GGUF. Additional community GGUF quantizations are welcome.

Usage

🤗 Transformers + PEFT

python

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.8-27B", torch_dtype="bfloat16", device_map="auto")
model = PeftModel.from_pretrained(base, "Lathly/Qwen3.8-27B-Samantha_Uncensored_1.1_LoRA")
tok = AutoTokenizer.from_pretrained("Qwen/Qwen3.8-27B")

🦙 llama.cpp (GGUF)

You need a GGUF of the base model (e.g. a community Qwen3.8-27B Q5_K_M) plus this LoRA GGUF:

bash

llama-server \
  -m Qwen3.8-27B-Q5_K_M.gguf \
  --lora Qwen3.8-27B-Samantha-LoRA-qwen35.gguf \
  -ngl 99 --host 0.0.0.0 --port 8080

Qwen LoRA merge fix

llama.cpp branch: qwen-lora-vhead-fix" Commit: f839835a3 Description: convert : reorder V heads for LoraTorchTensor Required to successfully merge certain Qwen LoRA adapters.

Important: the adapter GGUF here is already tagged with general.architecture = qwen35 to match the base GGUF. If you regenerate the adapter yourself with an older convert_lora_to_gguf.py, it may tag as qwen3 and llama.cpp will refuse to load it with model arch and LoRA arch mismatch. Retag general.architecture to qwen35 to fix.

Training

Fine-tuned with Unsloth (QLoRA, 4-bit) on dual RTX 5070 Ti. Settings were chosen to closely match the reference Samantha training spec. The adapter was trained on text-only conversational data with a sequence length of 2048; this does not change the base model's supported context length, but Samantha-style behavior at long context has not been systematically evaluated:

Parameter	Value
LoRA rank (r)	16
LoRA alpha	32
Target modules	all-linear (q, k, v, o, gate, up, down)
LoRA dropout	0.05
Bias	none
Task type	CAUSAL_LM
Optimizer	8-bit Paged AdamW
Learning rate	2e-5
LR scheduler	linear
Epochs	1
Effective batch size	32 (grad accumulation)
Sequence length	2048
Warmup steps	50
Weight decay	0.01
Final train loss	~1.42

License & Attribution

Released under Apache-2.0, inheriting from both parents:

Dataset: digitalpipelines/samantha-1.1-uncensored — Apache-2.0
The original Samantha persona/dataset concept is by Eric Hartford.

This is a modified derivative (LoRA fine-tune) of the above. No warranty. Outputs are the model's own; use responsibly.

Disclaimer

This is an "uncensored" persona fine-tune intended for open conversational use. It may produce content that some find objectionable. You are responsible for how you use it and for compliance with applicable laws and the base-model/dataset licenses.