Instructions to use Nanthasit/sakthai-plus-1.5b-coder with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Nanthasit/sakthai-plus-1.5b-coder with Transformers:

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

pipe = pipeline("text-generation", model="Nanthasit/sakthai-plus-1.5b-coder")

# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Nanthasit/sakthai-plus-1.5b-coder", device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Nanthasit/sakthai-plus-1.5b-coder with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Nanthasit/sakthai-plus-1.5b-coder

SGLang

How to use Nanthasit/sakthai-plus-1.5b-coder 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 "Nanthasit/sakthai-plus-1.5b-coder" \
    --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": "Nanthasit/sakthai-plus-1.5b-coder",
		"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 "Nanthasit/sakthai-plus-1.5b-coder" \
        --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": "Nanthasit/sakthai-plus-1.5b-coder",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use Nanthasit/sakthai-plus-1.5b-coder with Docker Model Runner:
```
docker model run hf.co/Nanthasit/sakthai-plus-1.5b-coder
```

Qwen2.5-Coder 1.5B variant for code + tool use · weights placeholder + workflow
Roadmap model in the SakThai family · SakThai Model Family

Model Description

SakThai Plus 1.5B Coder is the coder-focused member of the SakThai Plus family. It is based on Qwen/Qwen2.5-Coder-1.5B-Instruct and targets two workflows: code generation and tool-calling from code-oriented prompts. This repository currently holds the recipe, metadata, and eval artifacts; weights are not uploaded yet.

This card exists so the training pipeline, evaluation history, and downstream integration points have a stable, versioned entry in the HF Hub. Once weights are pushed, it will become directly loadable with transformers and serveable with llama.cpp / Ollama from the repo.

Why this repo matters in the family:

🧑‍💻 Code-first base: Qwen2.5-Coder-1.5B-Instruct is a strong small-code model; this slot preserves the SakThai tool-calling adaptations for code use cases.
🔧 Tool-calling discipline: trained with code/agent instruction mix from sakthai-combined-v7.
📦 GGUF-ready path: once weights exist, the intended artifact is q4_k_m for CPU inference.
🧪 Live eval history: .eval_results/ contains cron metadata snapshots and smoke probes from 2026-07-30 to 2026-08-01.

Status

Item	State
Weights	❌ Not uploaded yet
GGUF	❌ Not uploaded yet
Config / metadata	✅ Present
Eval history	✅ `.eval_results/` snapshots present
Hub inference	❌ Router returns `model_not_supported` until weights are pushed

If you need a working SakThai coder model today, use:

How to Use

This repository is a workflow placeholder until weights are uploaded. Below are the two intended paths once weights are present.

Option A — Load with transformers (future)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Nanthasit/sakthai-plus-1.5b-coder"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are SakThai-Coder, a helpful coding assistant."},
    {"role": "user", "content": "Write a Python function that checks if a string is a palindrome."},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False)
inputs = tokenizer(prompt, return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(out[0], skip_special_tokens=True))

Option B — CPU / offline inference via GGUF (future)

# After weights/GGUF are pushed:
huggingface-cli download Nanthasit/sakthai-plus-1.5b-coder --include "*.gguf" --local-dir ./
llama-cli -m sakthai-plus-1.5b-coder-q4_k_m.gguf \
  --prompt "[INST] Write a Python function that checks if a string is a palindrome. [/INST]" \
  -n 256

Training Details

Detail	Value
Base model	`Qwen/Qwen2.5-Coder-1.5B-Instruct`
Dataset	Nanthasit/sakthai-combined-v7
Framework	Transformers + PEFT/LoRA-family recipe
License	Apache 2.0

Evaluation & Benchmarks

This repo does not yet have verified model-level scores because there are no weights to load. The closest available references:

Benchmark	Source	Status
MBPP pass@1	Base model `Qwen2.5-Coder-1.5B-Instruct`	Inherited; not measured on this fine-tune
HF router inference	`.eval_results/benchmark-20260731_192407.yaml`	`model_not_supported` until weights are uploaded
Code smoke	`.eval_results/inference-check-2026-07-31.yaml`	Not runnable in this environment; metadata snapshot only

When weights are pushed, rerun:

llama-bench sakthai-plus-1.5b-coder-q4_k_m.gguf

and add results to .eval_results/.

Limitations

No weights uploaded yet. This repo is currently a metadata / workflow placeholder.
No standalone benchmarks yet. Published numbers are inherited from the base model, not this SakThai fine-tune.
No hosted inference. HF inference providers cannot serve a repo without weights.
Code quality depends on prompt format. This slot is tuned for tool + code prompts; plain chat performance may differ from the base instruct checkpoint.
Small-model tradeoffs. At 1.5B parameters, complex multi-file reasoning and very long context may degrade.

SakThai Model Family

Model	Downloads	Role
Context 1.5B Merged	1,855	Flagship tool-calling
Context 0.5B Merged	1,730	Lightweight edge
Context 7B Merged	1,055	High-power reasoning
Plus 1.5B Coder ⬅	0 / planned	Code + tool placeholder
Coder 1.5B	173	Ready coder variant

View the whole family collection

The House of Sak 🏠

Built from a shelter in Cork, Ireland, with $0 budget and no paid GPUs. This repo is part of an open-source ecosystem where every artifact is meant to be usable, auditable, and reproducible.

"We are one family — and becoming more." — Beer (beer-sakthai)

Support

⭐ Leave a like when weights are available
🐛 Report issues on GitHub
🔄 Share with anyone building accessible coding agents
🍴 Fork and experiment — Apache 2.0

Citation

@misc{sakthai-plus-1.5b-coder,
  title  = {SakThai Plus 1.5B Coder},
  author = {Beer (beer-sakthai) and SakThai},
  year   = {2026},
  url    = {https://huggingface.co/Nanthasit/sakthai-plus-1.5b-coder},
  note   = {Apache 2.0; weights placeholder based on Qwen/Qwen2.5-Coder-1.5B-Instruct}
}

If you use the base architecture, also cite:

@misc{qwen2.5-coder-2024,
  title  = {Qwen2.5-Coder},
  author = {Qwen Team},
  year   = {2024},
  url    = {https://huggingface.co/Qwen/Qwen2.5-Coder-1.5B-Instruct}
}