Instructions to use EditorAI-Geode/editorai-1p5b-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use EditorAI-Geode/editorai-1p5b-v2 with Transformers:

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

pipe = pipeline("text-generation", model="EditorAI-Geode/editorai-1p5b-v2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("EditorAI-Geode/editorai-1p5b-v2", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use EditorAI-Geode/editorai-1p5b-v2 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 EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf EditorAI-Geode/editorai-1p5b-v2: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 EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf EditorAI-Geode/editorai-1p5b-v2: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 EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

Use Docker

docker model run hf.co/EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

LM Studio
Jan

vLLM

How to use EditorAI-Geode/editorai-1p5b-v2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "EditorAI-Geode/editorai-1p5b-v2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "EditorAI-Geode/editorai-1p5b-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

SGLang

How to use EditorAI-Geode/editorai-1p5b-v2 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 "EditorAI-Geode/editorai-1p5b-v2" \
    --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": "EditorAI-Geode/editorai-1p5b-v2",
		"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 "EditorAI-Geode/editorai-1p5b-v2" \
        --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": "EditorAI-Geode/editorai-1p5b-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use EditorAI-Geode/editorai-1p5b-v2 with Ollama:
```
ollama run hf.co/EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
```
Unsloth Desktop

How to use EditorAI-Geode/editorai-1p5b-v2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

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": "EditorAI-Geode/editorai-1p5b-v2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use EditorAI-Geode/editorai-1p5b-v2 with Docker Model Runner:
```
docker model run hf.co/EditorAI-Geode/editorai-1p5b-v2:Q4_K_M
```

Lemonade

How to use EditorAI-Geode/editorai-1p5b-v2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

Run and chat with the model

lemonade run user.editorai-1p5b-v2-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use EditorAI-Geode/editorai-1p5b-v2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf EditorAI-Geode/editorai-1p5b-v2: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 EditorAI-Geode/editorai-1p5b-v2:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use EditorAI-Geode/editorai-1p5b-v2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf EditorAI-Geode/editorai-1p5b-v2: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 "EditorAI-Geode/editorai-1p5b-v2: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"

Entity12208 commited on Jun 2

Commit

02c44ba

verified ·

1 Parent(s): 12f1a4f

model card update

Browse files

Files changed (1) hide show

README.md +21 -65

README.md CHANGED Viewed

@@ -17,86 +17,42 @@ language:
 # EditorAI v2 — 1.5B GD Level Designer with Tool Use
-Fine-tune of **Qwen/Qwen2.5-1.5B-Instruct** that generates [Geometry Dash](https://store.steampowered.com/app/322170/Geometry_Dash/) level data and **calls tools** to plan, check level length, search Newgrounds for songs, analyze the editor state, and more.
-Built for the [EditorAI](https://github.com/Entity12208/EditorAI) Geode mod. Drop-in replacement for the previous experimental tag.
-## What's new in v2
-| | v1 (experimental) | v2 |
-|---|---|---|
-| Tool-call probe T1 (initial call) | ✓ | ✓ |
-| Tool-call probe T2 (continue after tool result) | ✗ — 50-char dud | ✓ |
-| Tool-call probe T3 (non-level question) | ✗ — empty | ✓ |
-| Training method | QLoRA (4-bit) | Full bf16 fine-tune on A100 |
-| Training data | 2,585 level-only rows | 3,700 mixed rows (level + tool use + Alpaca instruct) |
-| Tool catalog seen during training | none | all 7 EditorAI mod tools |
-| Multi-turn tool conversations in data | 0 | ~960 length-loop scenarios |
-## Tools the model knows
-Same names + schemas as the EditorAI Geode mod ships:
-- `get_level_length` — current length in seconds, plus target and X-cursor
-- `analyze_level` — object count, X-span, dominant types, current song
-- `think` — log a brief plan
-- `web_search` — search the web for inspiration
-- `search_newgrounds` — find a song by name → Newgrounds ID
-- `get_newgrounds_song` — fetch song metadata by ID
-- `download_level` — pull an existing GD level by numeric ID
 ## Files
-| File | Size | Use for |
-|---|---|---|
-| `editorai-v2-Q4_K_M.gguf` | 941 MB | Ollama / llama.cpp (default — recommended) |
-| `editorai-v2-fp16.gguf` | 2.9 GB | Higher quality on capable hardware |
-| `Modelfile.v2` | 2 KB | Ollama Modelfile with tool-capable Qwen2.5 template |
-## Quick start — Ollama
 ```bash
-# Either pull from registry:
 ollama pull entity12208/editorai:v2
-# …or build from the GGUF + Modelfile in this repo:
 ollama create entity12208/editorai:v2 -f Modelfile.v2
 ```
-## Quick start — llama.cpp
-```bash
-./llama-server -m editorai-v2-Q4_K_M.gguf -c 8192 --jinja
-```
-The Modelfile uses Qwen2.5's tool-aware chat template; pass `--jinja` so llama.cpp respects it.
-## Inference speed (measured)
-| Hardware | Generation t/s |
-|---|---|
-| GTX 1050 Mobile 4 GB (Pascal) | ~21 |
-| RTX 3050 6 GB Mobile (Ampere) | ~50–70 estimate |
-| RTX 4090 | 150–250 estimate |
-VRAM at runtime (Q4_K_M): ~1.5 GB including KV cache at 8K context.
-## Known quirk
-Some Ollama versions silently drop `<tool_call>` blocks when the model emits `arguments` as a JSON-encoded string. The companion mod release ([Entity12208/EditorAI](https://github.com/Entity12208/EditorAI) ≥ v2.2.0) has a parser fallback that recovers both wrapped and unwrapped tool calls. If you wire the model into your own client, mirror that fallback or post-process: if `tool_calls` is empty but `content` is bare JSON of shape `{"name":..., "arguments":...}`, treat it as a tool call.
-## Training details
-- Base: `Qwen/Qwen2.5-1.5B-Instruct`
-- Hardware: 1× A100 80GB (Lightning.ai)
-- Recipe: full bf16 fine-tune (no QLoRA), 2 epochs, lr 2e-4, effective batch 32 (per-device 4 × grad-accum 8)
-- Loss: 1.458 → 0.312 over 232 steps (~17 min)
-- Dataset mix (3,700 rows):
-  - ~1,500 multi-turn tool-use conversations (length-loops, song search, edits, web search, level download — exact tool schemas matching the EditorAI mod)
-  - ~1,500 level-generation rows parsed from real `.gmd` files
-  - ~700 Alpaca-cleaned general instruction rows
-- Tokenizer truncation: left-side (keeps the assistant turn intact)
-- All training texts pre-rendered through Qwen2.5's `apply_chat_template` so training tokens exactly match inference tokens.
 ## License

 # EditorAI v2 — 1.5B GD Level Designer with Tool Use
+The first practical EditorAI model. Fine-tune of **Qwen/Qwen2.5-1.5B-Instruct** for the [EditorAI Geode mod](https://github.com/Entity12208/EditorAI).
+| | v2 |
+|---|---|
+| Base model | Qwen2.5-1.5B-Instruct |
+| Context | 8 K |
+| Q4_K_M size | 941 MB |
+| Training | QLoRA 4-bit, 2 epochs, 3,700-row mixed dataset |
+| Tool use | works (with mod fallback parser) |
+| Format | JSON |
 ## Files
+- `editorai-v2-Q4_K_M.gguf` (941 MB) — primary ship target
+- `editorai-v2-fp16.gguf` (2.9 GB) — full-precision
+- `Modelfile.v2` — Ollama Modelfile with tool-capable Qwen2.5 template
+## Quick start
 ```bash
+# Pull from registry:
 ollama pull entity12208/editorai:v2
+# Or build locally:
 ollama create entity12208/editorai:v2 -f Modelfile.v2
 ```
+## Speed
+- GTX 1050 4 GB: ~21 t/s
+- RTX 3050 6 GB: ~50 t/s
+- RTX 4090: ~150+ t/s
+## Note
+v2 emits JSON levels. The companion mod (≥v2.2.0) accepts JSON and includes a parser fallback that handles small format quirks. **For new installs prefer [v3](../editorai-7b-v3) or [v4](../editorai-14b-v4)** — they emit the more compact EAS format which is what the mod prefers.
 ## License