Instructions to use itriedcoding/Satari-Code-1.5B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use itriedcoding/Satari-Code-1.5B with Transformers:

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

pipe = pipeline("text-generation", model="itriedcoding/Satari-Code-1.5B")

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("itriedcoding/Satari-Code-1.5B", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use itriedcoding/Satari-Code-1.5B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "itriedcoding/Satari-Code-1.5B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "itriedcoding/Satari-Code-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/itriedcoding/Satari-Code-1.5B

SGLang

How to use itriedcoding/Satari-Code-1.5B 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 "itriedcoding/Satari-Code-1.5B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "itriedcoding/Satari-Code-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "itriedcoding/Satari-Code-1.5B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "itriedcoding/Satari-Code-1.5B",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use itriedcoding/Satari-Code-1.5B with Docker Model Runner:
```
docker model run hf.co/itriedcoding/Satari-Code-1.5B
```

Satari Code 1.5B

Status: training pipeline and model card are published; weight files are uploaded after the first GPU fine-tune completes. Check the training repository for the exact recipe. No benchmark numbers are claimed before measurement.

Satari Code is a 1.5B-parameter coding model produced by QLoRA supervised fine-tuning of Qwen2.5-Coder-1.5B on public code-instruction data.

Base: Qwen/Qwen2.5-Coder-1.5B (Apache-2.0)
Method: 4-bit QLoRA, r=32, alpha=32, response-only loss
Data: ise-uiuc/Magicoder-OSS-Instruct-75K
Context: 2,048 tokens during fine-tuning
Licence: Apache-2.0

Training

Setting	Value
Precision	4-bit NF4 + bf16/fp16 compute
LoRA rank	32
LoRA alpha	32
Target modules	q, k, v, o, gate, up, down
Epochs	2
Effective batch	8
Learning rate	2e-4 (cosine)
Max sequence	2,048
Hardware	1× 16 GB GPU (T4 / A10G)

Reproduce with training/train_sft.py: see the Satari_Code_SFT.ipynb notebook.

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "itriedcoding/Satari-Code-1.5B"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto", device_map="auto")

messages = [
    {"role": "system", "content": "You are Satari Code, a precise coding assistant."},
    {"role": "user", "content": "Write a Python LRU cache with tests."},
]
inputs = tok.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(model.device)
out = model.generate(inputs, max_new_tokens=512, do_sample=True, temperature=0.2, top_p=0.95)
print(tok.decode(out[0][inputs.shape[1]:], skip_special_tokens=True))

A 4-bit GGUF (Q4_K_M) is available at itriedcoding/Satari-Code-1.5B-GGUF for CPU inference with llama.cpp.

Evaluation

Benchmark numbers are not claimed here. Run them yourself with your preferred harness once the weights exist; this card will be updated with measured results and the exact evaluation command.

Limitations

1.5B parameters: weaker than 7B+ models on long-horizon reasoning and large refactors.
Fine-tuned on 2,048-token samples; very large files will be truncated.
Can produce insecure or subtly incorrect code. Review all output before running it.
Primarily English prompts.

Licence

Apache-2.0, inherited from the base model.

Downloads last month: -; Downloads are not tracked for this model. How to track

Model tree for itriedcoding/Satari-Code-1.5B

Base model

Qwen/Qwen2.5-1.5B

Finetuned

Qwen/Qwen2.5-Coder-1.5B

Finetuned

(68)

this model