Instructions to use aelgendy/QModel with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use aelgendy/QModel 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 aelgendy/QModel:Q8_0
# Run inference directly in the terminal:
llama cli -hf aelgendy/QModel:Q8_0

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf aelgendy/QModel:Q8_0
# Run inference directly in the terminal:
llama cli -hf aelgendy/QModel:Q8_0

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf aelgendy/QModel:Q8_0
# Run inference directly in the terminal:
./build/bin/llama cli -hf aelgendy/QModel:Q8_0

Use Docker

docker model run hf.co/aelgendy/QModel:Q8_0

Llama new
LM Studio
Jan
Ollama
How to use aelgendy/QModel with Ollama:
```
ollama run hf.co/aelgendy/QModel:Q8_0
```
Unsloth Desktop

How to use aelgendy/QModel with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aelgendy/QModel:Q8_0

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "aelgendy/QModel:Q8_0"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use aelgendy/QModel with Docker Model Runner:
```
docker model run hf.co/aelgendy/QModel:Q8_0
```

Lemonade

How to use aelgendy/QModel with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull aelgendy/QModel:Q8_0

Run and chat with the model

lemonade run user.QModel-Q8_0

List all available models

lemonade list

Hermes Agent

How to use aelgendy/QModel with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aelgendy/QModel:Q8_0

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 aelgendy/QModel:Q8_0

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use aelgendy/QModel with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf aelgendy/QModel:Q8_0

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 "aelgendy/QModel:Q8_0" \
  --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"

QModel / .env.example

aelgendy

QModel v8.5: accuracy release (code, 3,598-fatwa index, quran.com links)

c634c96 15 days ago

Raw History Blame Contribute Delete

8.83 kB

# QModel 8.5 Configuration Template
# ==================================
# Copy this to .env and update values for your environment

# LLM Backend Selection
# Options: "gguf" (local GGUF file), "ollama", "hf" (HuggingFace), or "lmstudio"
LLM_BACKEND=gguf

# ─────────────────────────────────────────────────────────────────────
# GGUF BACKEND (if LLM_BACKEND=gguf) — self-contained, no external daemon
# ─────────────────────────────────────────────────────────────────────
GGUF_MODEL_PATH=./models/Qwen2.5-14B-Instruct-Q8_0.gguf
GGUF_N_CTX=8192              # Context window size
GGUF_N_GPU_LAYERS=-1         # -1 = offload all layers to GPU (Metal on Mac)

# ─────────────────────────────────────────────────────────────────────
# OLLAMA BACKEND (if LLM_BACKEND=ollama)
# ─────────────────────────────────────────────────────────────────────
# OLLAMA_HOST=http://localhost:11434
# OLLAMA_MODEL=qwen2.5:14b
# Available models: llama3.1, mistral, neural-chat, openhermes

# ─────────────────────────────────────────────────────────────────────
# HUGGINGFACE BACKEND (if LLM_BACKEND=hf)
# ─────────────────────────────────────────────────────────────────────
# HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
# HF_DEVICE=auto              # Options: auto, cuda, cpu
# HF_MAX_NEW_TOKENS=2048
# Popular models:
#   - Qwen/Qwen2.5-14B-Instruct (recommended)

# ─────────────────────────────────────────────────────────────────────
# LM STUDIO BACKEND (if LLM_BACKEND=lmstudio)
# ─────────────────────────────────────────────────────────────────────
# LMSTUDIO_URL=http://localhost:1234
# LMSTUDIO_MODEL=qwen2.5-14b-instruct   # Model loaded in LM Studio

# ─────────────────────────────────────────────────────────────────────
# EMBEDDING MODEL (shared by all backends)
# ─────────────────────────────────────────────────────────────────────
EMBED_MODEL=intfloat/multilingual-e5-large

# ─────────────────────────────────────────────────────────────────────
# DATA FILES
# ─────────────────────────────────────────────────────────────────────
FAISS_INDEX=QModel.index
METADATA_FILE=metadata.json

# ─────────────────────────────────────────────────────────────────────
# RETRIEVAL SETTINGS
# ─────────────────────────────────────────────────────────────────────
TOP_K_SEARCH=20         # Candidate pool size
TOP_K_RETURN=5          # Final results returned to user

# ─────────────────────────────────────────────────────────────────────
# GENERATION SETTINGS
# ─────────────────────────────────────────────────────────────────────
TEMPERATURE=0.2         # 0.0=deterministic, 1.0=creative
MAX_TOKENS=2048         # Max output length

# ─────────────────────────────────────────────────────────────────────
# SAFETY & QUALITY
# ─────────────────────────────────────────────────────────────────────
# Confidence gate: skip the LLM and return "not found" unless retrieval found
# something genuinely relevant (direct reference / verbatim text match, or):
# - USE_RERANKER=true  → best cross-encoder relevance (0-1) >= RERANK_MIN_SCORE
# - USE_RERANKER=false → best dense cosine similarity >= DENSE_MIN_SCORE
#   (multilingual-e5-large scale; recalibrate if EMBED_MODEL changes)
# Calibrate with: python -m eval.run_eval --show-failures
# On the golden set the reranker gate refuses 92% of off-topic questions vs 36% for
# the dense gate, for ~1 s per answer; set false only where the model can't be loaded.
USE_RERANKER=true

# Rewrite queries without the LLM (regex intent + thesaurus terms), ~2 s faster per
# question. Follow-ups with chat history always use the LLM rewrite.
FAST_QUERY_REWRITE=true
RERANK_MIN_SCORE=0.05
DENSE_MIN_SCORE=0.80

# Hadith boost: Score bonus when intent=hadith
# Prevents Quran verses from outranking relevant Hadiths
HADITH_BOOST=0.08

# Quran boost: Score bonus when intent=tafsir/quran
# Prevents Hadiths from outranking relevant Quran verses
QURAN_BOOST=0.12

# ─────────────────────────────────────────────────────────────────────
# RANKING
# ─────────────────────────────────────────────────────────────────────
RERANK_ALPHA=0.6        # 60% dense (embedding), 40% sparse (BM25)

# ─────────────────────────────────────────────────────────────────────
# CACHING
# ─────────────────────────────────────────────────────────────────────
CACHE_SIZE=512          # Max cache entries
CACHE_TTL=3600          # Cache expiry in seconds

# ─────────────────────────────────────────────────────────────────────
# SECURITY
# ─────────────────────────────────────────────────────────────────────
ALLOWED_ORIGINS=*       # CORS origins (restrict in production: origin1.com,origin2.com)

# ─────────────────────────────────────────────────────────────────────
# USAGE EXAMPLES
# ─────────────────────────────────────────────────────────────────────
#
# Self-contained (GGUF, e.g. HF Spaces):
#   LLM_BACKEND=gguf
#   GGUF_MODEL_PATH=./models/Qwen2.5-14B-Instruct-Q8_0.gguf
#
# Development (Ollama):
#   LLM_BACKEND=ollama
#   OLLAMA_HOST=http://localhost:11434
#   OLLAMA_MODEL=qwen2.5:14b
#
# Production (HuggingFace GPU):
#   LLM_BACKEND=hf
#   HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
#   HF_DEVICE=cuda
#
# Production (HuggingFace CPU):
#   LLM_BACKEND=hf
#   HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
#   HF_DEVICE=cpu

	# QModel 8.5 Configuration Template
	# ==================================
	# Copy this to .env and update values for your environment

	# LLM Backend Selection
	# Options: "gguf" (local GGUF file), "ollama", "hf" (HuggingFace), or "lmstudio"
	LLM_BACKEND=gguf

	# ─────────────────────────────────────────────────────────────────────
	# GGUF BACKEND (if LLM_BACKEND=gguf) — self-contained, no external daemon
	# ─────────────────────────────────────────────────────────────────────
	GGUF_MODEL_PATH=./models/Qwen2.5-14B-Instruct-Q8_0.gguf
	GGUF_N_CTX=8192 # Context window size
	GGUF_N_GPU_LAYERS=-1 # -1 = offload all layers to GPU (Metal on Mac)

	# ─────────────────────────────────────────────────────────────────────
	# OLLAMA BACKEND (if LLM_BACKEND=ollama)
	# ─────────────────────────────────────────────────────────────────────
	# OLLAMA_HOST=http://localhost:11434
	# OLLAMA_MODEL=qwen2.5:14b
	# Available models: llama3.1, mistral, neural-chat, openhermes

	# ─────────────────────────────────────────────────────────────────────
	# HUGGINGFACE BACKEND (if LLM_BACKEND=hf)
	# ─────────────────────────────────────────────────────────────────────
	# HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
	# HF_DEVICE=auto # Options: auto, cuda, cpu
	# HF_MAX_NEW_TOKENS=2048
	# Popular models:
	# - Qwen/Qwen2.5-14B-Instruct (recommended)

	# ─────────────────────────────────────────────────────────────────────
	# LM STUDIO BACKEND (if LLM_BACKEND=lmstudio)
	# ─────────────────────────────────────────────────────────────────────
	# LMSTUDIO_URL=http://localhost:1234
	# LMSTUDIO_MODEL=qwen2.5-14b-instruct # Model loaded in LM Studio

	# ─────────────────────────────────────────────────────────────────────
	# EMBEDDING MODEL (shared by all backends)
	# ─────────────────────────────────────────────────────────────────────
	EMBED_MODEL=intfloat/multilingual-e5-large

	# ─────────────────────────────────────────────────────────────────────
	# DATA FILES
	# ─────────────────────────────────────────────────────────────────────
	FAISS_INDEX=QModel.index
	METADATA_FILE=metadata.json

	# ─────────────────────────────────────────────────────────────────────
	# RETRIEVAL SETTINGS
	# ─────────────────────────────────────────────────────────────────────
	TOP_K_SEARCH=20 # Candidate pool size
	TOP_K_RETURN=5 # Final results returned to user

	# ─────────────────────────────────────────────────────────────────────
	# GENERATION SETTINGS
	# ─────────────────────────────────────────────────────────────────────
	TEMPERATURE=0.2 # 0.0=deterministic, 1.0=creative
	MAX_TOKENS=2048 # Max output length

	# ─────────────────────────────────────────────────────────────────────
	# SAFETY & QUALITY
	# ─────────────────────────────────────────────────────────────────────
	# Confidence gate: skip the LLM and return "not found" unless retrieval found
	# something genuinely relevant (direct reference / verbatim text match, or):
	# - USE_RERANKER=true → best cross-encoder relevance (0-1) >= RERANK_MIN_SCORE
	# - USE_RERANKER=false → best dense cosine similarity >= DENSE_MIN_SCORE
	# (multilingual-e5-large scale; recalibrate if EMBED_MODEL changes)
	# Calibrate with: python -m eval.run_eval --show-failures
	# On the golden set the reranker gate refuses 92% of off-topic questions vs 36% for
	# the dense gate, for ~1 s per answer; set false only where the model can't be loaded.
	USE_RERANKER=true

	# Rewrite queries without the LLM (regex intent + thesaurus terms), ~2 s faster per
	# question. Follow-ups with chat history always use the LLM rewrite.
	FAST_QUERY_REWRITE=true
	RERANK_MIN_SCORE=0.05
	DENSE_MIN_SCORE=0.80

	# Hadith boost: Score bonus when intent=hadith
	# Prevents Quran verses from outranking relevant Hadiths
	HADITH_BOOST=0.08

	# Quran boost: Score bonus when intent=tafsir/quran
	# Prevents Hadiths from outranking relevant Quran verses
	QURAN_BOOST=0.12

	# ─────────────────────────────────────────────────────────────────────
	# RANKING
	# ─────────────────────────────────────────────────────────────────────
	RERANK_ALPHA=0.6 # 60% dense (embedding), 40% sparse (BM25)

	# ─────────────────────────────────────────────────────────────────────
	# CACHING
	# ─────────────────────────────────────────────────────────────────────
	CACHE_SIZE=512 # Max cache entries
	CACHE_TTL=3600 # Cache expiry in seconds

	# ─────────────────────────────────────────────────────────────────────
	# SECURITY
	# ─────────────────────────────────────────────────────────────────────
	ALLOWED_ORIGINS=* # CORS origins (restrict in production: origin1.com,origin2.com)

	# ─────────────────────────────────────────────────────────────────────
	# USAGE EXAMPLES
	# ─────────────────────────────────────────────────────────────────────
	#
	# Self-contained (GGUF, e.g. HF Spaces):
	# LLM_BACKEND=gguf
	# GGUF_MODEL_PATH=./models/Qwen2.5-14B-Instruct-Q8_0.gguf
	#
	# Development (Ollama):
	# LLM_BACKEND=ollama
	# OLLAMA_HOST=http://localhost:11434
	# OLLAMA_MODEL=qwen2.5:14b
	#
	# Production (HuggingFace GPU):
	# LLM_BACKEND=hf
	# HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
	# HF_DEVICE=cuda
	#
	# Production (HuggingFace CPU):
	# LLM_BACKEND=hf
	# HF_MODEL_NAME=Qwen/Qwen2.5-14B-Instruct
	# HF_DEVICE=cpu