Instructions to use sandeshrajx/qwopus-lora-tests with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use sandeshrajx/qwopus-lora-tests with Transformers:

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("sandeshrajx/qwopus-lora-tests", device_map="auto")

llama-cpp-python

How to use sandeshrajx/qwopus-lora-tests with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="sandeshrajx/qwopus-lora-tests",
	filename="gguf-merged/Flagstone8878-Qwen3.5-18B-REAP-A3B-Coding-qwopus-merged-Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = "No input example has been defined for this model task."
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use sandeshrajx/qwopus-lora-tests 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 sandeshrajx/qwopus-lora-tests:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf sandeshrajx/qwopus-lora-tests:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf sandeshrajx/qwopus-lora-tests:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf sandeshrajx/qwopus-lora-tests: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 sandeshrajx/qwopus-lora-tests:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf sandeshrajx/qwopus-lora-tests: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 sandeshrajx/qwopus-lora-tests:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf sandeshrajx/qwopus-lora-tests:Q4_K_M

Use Docker

docker model run hf.co/sandeshrajx/qwopus-lora-tests:Q4_K_M

LM Studio
Jan
Ollama
How to use sandeshrajx/qwopus-lora-tests with Ollama:
```
ollama run hf.co/sandeshrajx/qwopus-lora-tests:Q4_K_M
```

Unsloth Studio

How to use sandeshrajx/qwopus-lora-tests 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 sandeshrajx/qwopus-lora-tests 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 sandeshrajx/qwopus-lora-tests to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for sandeshrajx/qwopus-lora-tests to start chatting

How to use sandeshrajx/qwopus-lora-tests with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sandeshrajx/qwopus-lora-tests: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": "sandeshrajx/qwopus-lora-tests:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use sandeshrajx/qwopus-lora-tests with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sandeshrajx/qwopus-lora-tests: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 sandeshrajx/qwopus-lora-tests:Q4_K_M

Run Hermes

hermes

Atomic Chat new

OpenClaw new

How to use sandeshrajx/qwopus-lora-tests with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf sandeshrajx/qwopus-lora-tests: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 "sandeshrajx/qwopus-lora-tests: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 sandeshrajx/qwopus-lora-tests with Docker Model Runner:
```
docker model run hf.co/sandeshrajx/qwopus-lora-tests:Q4_K_M
```

Lemonade

How to use sandeshrajx/qwopus-lora-tests with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull sandeshrajx/qwopus-lora-tests:Q4_K_M

Run and chat with the model

lemonade run user.qwopus-lora-tests-Q4_K_M

List all available models

lemonade list

sandeshrajx commited on Apr 5

Commit

f95cbaa

verified ·

1 Parent(s): 812d21e

Unsloth Model Card

Browse files

Files changed (1) hide show

README.md +14 -97

README.md CHANGED Viewed

@@ -1,104 +1,21 @@
 ---
-license: apache-2.0
-base_model: Flagstone8878/Qwen3.5-18B-REAP-A3B-Coding
 tags:
-  - qwen
-  - qwen3
-  - qwen3.5
-  - lora
-  - unsloth
-  - code
-  - reasoning
-  - gguf
-library_name: peft
-pipeline_tag: text-generation
 ---
-# sandeshrajx/qwopus-lora-tests
-LoRA adapters and GGUF exports from an experimental Qwen 3.5 18B REAP-A3B coding fine-tune on Opus-distilled reasoning and coding data.
-## What This Repo Contains
-This repo is being used to compare a base checkpoint against a lightweight supervised fine-tune focused on coding-style outputs with explicit `<think>...</think>` reasoning scaffolds.
-It contains:
-- A LoRA adapter trained from `Flagstone8878/Qwen3.5-18B-REAP-A3B-Coding`
-- Matching tokenizer files required to load the adapter
-- GGUF exports for llama.cpp-compatible testing
-## Training Setup
-- Base model: `Flagstone8878/Qwen3.5-18B-REAP-A3B-Coding`
-- Dataset: `nohurry/Opus-4.6-Reasoning-3000x-filtered`
-- Run name: `qwen35-18b-reap-a3b-coding-opus-msl8192-e2_0-lr2e4`
-- Context length: `8192`
-- Epochs: `2.0`
-- Learning rate: `0.0002`
-- Effective batch size: `24`
-- Observed steps: `194`
-- Final reported train loss: `0.5748463515768346`
-## Goal
-The current goal is not to claim a production-quality coding model. This is an iterative testing repo for:
-- checking whether the LoRA improves structure and responsiveness versus the base model
-- measuring how well the model handles coding prompts with reasoning traces
-- comparing adapter behavior and GGUF behavior across different inference stacks
-## Prompt Format
-Training examples were rendered in Qwen chat format. Assistant messages were structured like:
-```text
-<think>
-...reasoning...
-</think>
-...final answer...
-```
-Response-only loss masking was applied starting at:
-```text
-<|im_start|>assistant
-<think>
-```
-That means the opening `<think>` token itself was used as the masking boundary during training.
-## Current Quality Notes
-This is an experimental checkpoint. Early prompt testing shows that the LoRA can produce shorter and more coherent outputs than the untouched base model on some coding prompts, but the generated code is still not consistently correct or production-ready.
-Use it for:
-- comparative testing
-- prompt behavior inspection
-- llama.cpp / GGUF experiments
-Do not assume it is a fully validated coding model.
-## Loading The LoRA
-Use the adapter with the original base model:
-```python
-from transformers import AutoTokenizer
-from peft import AutoPeftModelForCausalLM
-model = AutoPeftModelForCausalLM.from_pretrained("sandeshrajx/qwopus-lora-tests", trust_remote_code=True)
-tokenizer = AutoTokenizer.from_pretrained("sandeshrajx/qwopus-lora-tests", trust_remote_code=True, use_fast=False)
-```
-If your runtime expects the original base model separately, load `Flagstone8878/Qwen3.5-18B-REAP-A3B-Coding` and then attach the adapter from this repo.
-## GGUF Notes
-The GGUF files in this repo are intended for downstream testing in llama.cpp-compatible runtimes. They are provided as convenience artifacts for experimentation and may differ from Transformers / PEFT behavior depending on the backend and quantization format.
-## Provenance
-Artifacts were produced from local Modal training and export scripts in the `qwen-opus-finetune` workspace, then uploaded to Hugging Face for external evaluation.

 ---
+base_model: sandeshrajx/qwopus-lora-tests
 tags:
+- text-generation-inference
+- transformers
+- unsloth
+- qwen3_5_moe
+license: apache-2.0
+language:
+- en
 ---
+# Uploaded finetuned  model
+- **Developed by:** sandeshrajx
+- **License:** apache-2.0
+- **Finetuned from model :** sandeshrajx/qwopus-lora-tests
+This qwen3_5_moe 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)