Instructions to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1", dtype="auto")

llama-cpp-python

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1",
	filename="gguf/q2_k_gguf/Qwen3.5-0.8B.Q2_K.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Use Docker

docker model run hf.co/Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

LM Studio
Jan

vLLM

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

SGLang

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Ollama:
```
ollama run hf.co/Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
```

Unsloth Studio

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 to start chatting

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Docker Model Runner
How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Docker Model Runner:
```
docker model run hf.co/Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M
```

Lemonade

How to use Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1:Q4_K_M

Run and chat with the model

lemonade run user.qwen3.5-0.8b-unslop-good-lora-v1-Q4_K_M

List all available models

lemonade list

Oysiyl commited on Mar 29

Commit

788ae2b

verified ·

1 Parent(s): 4807369

Unsloth Model Card

Browse files

Files changed (1) hide show

README.md +12 -95

README.md CHANGED Viewed

@@ -1,104 +1,21 @@
 ---
-language:
-- en
-license: apache-2.0
-base_model: Qwen/Qwen3.5-0.8B
-library_name: transformers
 tags:
 - unsloth
 - qwen3_5
-- lora
-- rewriting
-- style-transfer
-- unslop
-pipeline_tag: text-generation
 ---
-# qwen3.5-0.8b-unslop-good-lora-v1
-A Qwen 3.5 0.8B fine-tune for unslop rewriting: taking AI-sounding passages and attempting to rewrite them into cleaner, more natural prose while preserving meaning.
-This run is the Qwen 3.5 0.8B text-model lane in the post-30B follow-up series: meant to test whether a stronger newer family can produce a meaningful quality jump without going all the way back to the largest hardware tier.
-## How it was trained
-- Base model: `Qwen/Qwen3.5-0.8B`
-- Training path: vanilla Transformers/TRL/PEFT fine-tuning on Hugging Face Jobs
-- Dataset: `N8Programs/unslop-good`
-- Rows used: 1000 (full training split)
-- Objective: direct rewrite / style cleanup
-## Training shape
-- hardware: A10G 24GB (`a10g-large`)
-- max_seq_length: 2048
-- num_train_epochs: 2
-- batch_size: 1
-- gradient_accumulation_steps: 1
-- learning_rate: 1e-4
-- scheduler: cosine
-- warmup_steps: 50
-- LoRA rank: 8
-- LoRA alpha: 20
-- LoRA dropout: 0.0
-- 4-bit loading
-- bf16 training
-## Training outcome
-This run is deployment-backed and live on Modal.
-- Modal endpoint: healthy
-- base model: `Qwen/Qwen3.5-0.8B`
-- output repo: `Oysiyl/qwen3.5-0.8b-unslop-good-lora-v1`
-- train_runtime: 5554s
-- train_loss: 2.504
-- final step: 500
-Inference settings used on the live endpoint:
-- enable_thinking: `false`
-- temperature: `0.7`
-- top_p: `0.8`
-- top_k: `20`
-- min_p: `0.0`
-- presence_penalty: `1.5`
-- repetition_penalty: `1.0`
-## Modal-backed evaluation
-Short sanity sample observed output:
-> I need to check my progress on this request.
-The long fiction passage is still rewritten coherently, but this size is clearly the least stable.
-Held-out fiction sample observed output:
-> The wind surged upward abruptly, as if the mountain itself had taken a deep breath. She stood at
-> the precipice of the pass, her cloak fraying like a torn banner trailing behind her. Her gaze
-> fixed on the valley where the last few lights flickered and died one by one. Each extinguishing
-> flame felt intentional—a response to a question she had avoided asking. Below, bells began to
-> toll, not in alarm, but in sorrow. They already knew. "Don't," he whispered, breathless, his
-> hand gripping hers. His fingers touched the fabric, then the air. "If you step over that
-> threshold, there's no turning back." She did not look away. The biting cold stung her throat,
-> sharpening her mind. "There was never a way back," she replied quietly. "Only a path forward you
-> were willing to pretend never existed." Thunder crashed down, heavy and ancient. In a flash, the
-> sky cracked open into white fire; in that split second, he saw her face—resolved, terrified,
-> blazing with purpose. It wasn't the girl he had sworn to guard, but the woman the world had
-> forced upon herself. "You'll die," he said. She laughed, finally meeting him in the center of
-> the scene,
-## Judgment
-Blunt judgment: this is the weakest of the trio. It does show real rewrite ability on long prose, but the short hook still collapses into a meta/no-op, so it is not yet a reliable production rewrite model.
-## Training loss vs training progress
-See the normalized 9B curve plus 0.8B vs 2B vs 4B comparison plot below.
-![Normalized training loss comparison: 9B vs 0.8B vs 2B vs 4B](./training_loss_vs_progress_comparison_9b_0_8b_vs_2b_vs_4b.svg)
-The 9B line is now the actual later 9B run, not a dotted baseline.
-## Comparison across the trio
-- 0.8B: the roughest one; good long-form cleanup, but the short hook is still too brittle
-- 2B: a noticeably better balance of fluency and fidelity
-- 4B: the most polished of the three and the strongest default candidate here
-- 9B: now the actual curve in the plot, and the best train_loss result in the family
-## Conclusion
-This repo is now a real post-run artifact with deployment-backed evaluation notes. The smaller sizes are useful as cheaper pilots, but the 4B lane is the best first-choice rewrite candidate in this set.

 ---
+base_model: unsloth/Qwen3.5-0.8B
 tags:
+- text-generation-inference
+- transformers
 - unsloth
 - qwen3_5
+license: apache-2.0
+language:
+- en
 ---
+# Uploaded finetuned  model
+- **Developed by:** Oysiyl
+- **License:** apache-2.0
+- **Finetuned from model :** unsloth/Qwen3.5-0.8B
+This qwen3_5 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
+[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)