Instructions to use posttrainllm/qwen3-4b-file-ops-distilled with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use posttrainllm/qwen3-4b-file-ops-distilled with Transformers:

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

pipe = pipeline("text-generation", model="posttrainllm/qwen3-4b-file-ops-distilled")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("posttrainllm/qwen3-4b-file-ops-distilled")
model = AutoModelForCausalLM.from_pretrained("posttrainllm/qwen3-4b-file-ops-distilled", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

MLX

How to use posttrainllm/qwen3-4b-file-ops-distilled with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("posttrainllm/qwen3-4b-file-ops-distilled")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

vLLM

How to use posttrainllm/qwen3-4b-file-ops-distilled with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "posttrainllm/qwen3-4b-file-ops-distilled"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "posttrainllm/qwen3-4b-file-ops-distilled",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/posttrainllm/qwen3-4b-file-ops-distilled

SGLang

How to use posttrainllm/qwen3-4b-file-ops-distilled 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 "posttrainllm/qwen3-4b-file-ops-distilled" \
    --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": "posttrainllm/qwen3-4b-file-ops-distilled",
		"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 "posttrainllm/qwen3-4b-file-ops-distilled" \
        --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": "posttrainllm/qwen3-4b-file-ops-distilled",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

How to use posttrainllm/qwen3-4b-file-ops-distilled with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "posttrainllm/qwen3-4b-file-ops-distilled"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "posttrainllm/qwen3-4b-file-ops-distilled"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use posttrainllm/qwen3-4b-file-ops-distilled with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "posttrainllm/qwen3-4b-file-ops-distilled"

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 posttrainllm/qwen3-4b-file-ops-distilled

Run Hermes

hermes

OpenClaw new

How to use posttrainllm/qwen3-4b-file-ops-distilled with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "posttrainllm/qwen3-4b-file-ops-distilled"

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 "posttrainllm/qwen3-4b-file-ops-distilled" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

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

MLX LM

How to use posttrainllm/qwen3-4b-file-ops-distilled with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "posttrainllm/qwen3-4b-file-ops-distilled"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "posttrainllm/qwen3-4b-file-ops-distilled"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "posttrainllm/qwen3-4b-file-ops-distilled",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Docker Model Runner
How to use posttrainllm/qwen3-4b-file-ops-distilled with Docker Model Runner:
```
docker model run hf.co/posttrainllm/qwen3-4b-file-ops-distilled
```

Qwen3-4B File-Ops Distilled

Summary

This is the first TinyGPT specialist package for a model we actually built: a fused Qwen3-4B-Instruct-2507 bf16 HF/MLX safetensors directory distilled for GorillaFileSystem multi-turn file-operation tasks.

It is a routed specialist, not the general Pace planner.

Artifact

Package id: qwen3-4b-file-ops-distilled
Public artifact: sarthakagrawal927/qwen3-4b-file-ops-distilled
Public storage target: Hugging Face Hub model repo
Format: HF/MLX safetensors directory
Base: Qwen/Qwen3-4B-Instruct-2507
Precision: bf16
Training method: frontier/gold trajectory distillation rendered in the student's native tool-calling chat template

Measured Result

Suite	Stock 4B	Distilled 4B
File-ops hard gate	58%	100%
File-ops hardgen held-out	-	95%
Out-of-domain breadth	59.6%	42.3%

The file-ops domain saturated at 4B: the distilled model matched frontier on the hard and veryhard file-ops gates. The same training caused negative transfer outside that domain, so this package is only correct behind a router.

Recommended Use

Use this model when the router has already identified a file-operation task with derivable arguments: paths, file names, directories, moves, creates, deletes, and navigation through a file-system backend.

Do not use it as a general planner. For general multi-domain planning, use the planner lock in docs/planner-lock-2026-06-19.md: stock Qwen3-4B-Instruct-2507 bf16 with the plan-then-execute prompt.

Known Limits

The specialist regresses on non-filesystem BFCL multi-turn backends.
It was not validated as a broad Pace planner.
It depends on correct routing. Bad routing turns a narrow win into a broad regression.
The artifact is multi-GB and is published on Hugging Face Hub, not committed to this repository or required to remain in local cache.

References

docs/learn/tool-calling-frontier-parity.md sections 8.1-8.5
docs/planner-lock-2026-06-19.md

Downloads last month: 389

Safetensors

Model size

4B params

Tensor type

BF16

MLX

Hardware compatibility

Quantized

Model tree for posttrainllm/qwen3-4b-file-ops-distilled

Base model

Qwen/Qwen3-4B-Instruct-2507

Finetuned

(1859)

this model