silvermete0r/qwen2.5-nano-function-master

Qwen2.5-0.5B - Nano Function Master

A lightweight, fast function-calling model fine-tuned from Qwen/Qwen2.5-0.5B-Instruct on the Salesforce/xlam-function-calling-60k dataset.

Part of the Fast Nano SLMs project - optimizing inference and memory efficiency for function-calling tasks in private/resource-constrained environments.

Project repository: silvermete0r/nano-slms-for-local-function-calling-assistance

---

Training Details

Hyperparameter	Value
LoRA rank (r)	16
LoRA alpha	32
Target modules	q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj, down_proj
LoRA dropout	0.05
Training samples	9,000
Eval samples	1,000
Max steps	600
Effective batch size	16 (4 × 4 grad accumulation)
Learning rate	2e-4
LR scheduler	Cosine with 50 warmup steps
Optimizer	AdamW (fused)
Weight decay	0.01
Gradient clipping	1.0
Max sequence length	1024
Packing	❌ disabled
Dataset	Salesforce/xlam-function-calling-60k

---

Benchmark Results

Evaluated on 1,000 held-out samples from xlam-function-calling-60k:

Before fine-tuning | After fine-tuning ~ Results

Metric                           Before      After      Delta
--------------------------------------------------------------
json_valid_pct                     45.4       98.7      +53.3
name_match_pct                     15.0       93.3      +78.3
args_keys_match_pct                11.8       81.0      +69.2
args_exact_pct                      8.9       69.1      +60.2

Estimated during the final evaluation inference stage of the fine-tuned model.

Metric	Value
Average TTFT (s)	0.0475
Average Tokens per Second	154.0545
Average Tokens Generated	51.199
Average VRAM Delta (MB)	0.0094
Peak VRAM Reserved (MB)	3120.0
Average RAM Delta (MB)	-0.0666
Average CPU Usage (%)	69.2192
Throughput (samples/s)	2.06
Estimated CO₂ Emissions (g CO₂eq)	5.4236

---

Usage

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch, json

model_id = "silvermete0r/qwen2.5-nano-function-master"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, torch_dtype=torch.float16, device_map="auto")

tools = [
    {
        "name": "get_weather",
        "description": "Get the current weather for a location",
        "parameters": {
            "location": {"type": "string", "description": "City name"},
            "unit":     {"type": "string", "enum": ["celsius", "fahrenheit"]}
        }
    }
]

messages = [
    {
        "role": "system",
        "content": (
            "You are a helpful assistant with access to the following functions. "
            f"Use them if required:\n{json.dumps(tools)}\n\n"
            "Respond strictly with a JSON array of function calls."
        )
    },
    {"role": "user", "content": "What's the weather in Aktobe?"},
]

inputs = tokenizer.apply_chat_template(
    messages, return_tensors="pt", add_generation_prompt=True).to(model.device)

outputs = model.generate(inputs, max_new_tokens=128, do_sample=False)
response = tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True)
print(json.loads(response))
# → [{"name": "get_weather", "arguments": {"location": "Aktobe", "unit": "celsius"}}]

Batched inference (faster)

tokenizer.padding_side = "left"
tokenizer.pad_token = tokenizer.eos_token

prompts = [
    tokenizer.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
    for msgs in list_of_message_lists
]
inputs = tokenizer(prompts, return_tensors="pt", padding=True, truncation=True,
                   max_length=1024).to(model.device)

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=256, do_sample=False,
                             pad_token_id=tokenizer.pad_token_id)

responses = [
    tokenizer.decode(out[inputs["input_ids"].shape[1]:], skip_special_tokens=True)
    for out in outputs
]

---

Intended Use

• ✅ Local function-calling assistants (offline / air-gapped)
• ✅ Edge deployment (laptops, Raspberry Pi 5, Jetson Nano)
• ✅ Privacy-sensitive environments (no API calls)
• ✅ Structured JSON output generation
• ❌ Not intended for open-ended conversational use

References

@misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
}

@article{qwen2,
      title={Qwen2 Technical Report},
      author={An Yang and Baosong Yang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Zhou and Chengpeng Li and Chengyuan Li and Dayiheng Liu and Fei Huang and Guanting Dong and Haoran Wei and Huan Lin and Jialong Tang and Jialin Wang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Ma and Jin Xu and Jingren Zhou and Jinze Bai and Jinzheng He and Junyang Lin and Kai Dang and Keming Lu and Keqin Chen and Kexin Yang and Mei Li and Mingfeng Xue and Na Ni and Pei Zhang and Peng Wang and Ru Peng and Rui Men and Ruize Gao and Runji Lin and Shijie Wang and Shuai Bai and Sinan Tan and Tianhang Zhu and Tianhao Li and Tianyu Liu and Wenbin Ge and Xiaodong Deng and Xiaohuan Zhou and Xingzhang Ren and Xinyu Zhang and Xipin Wei and Xuancheng Ren and Yang Fan and Yang Yao and Yichang Zhang and Yu Wan and Yunfei Chu and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zhihao Fan},
      journal={arXiv preprint arXiv:2407.10671},
      year={2024}
}

@article{liu2024apigen,
  title={APIGen: Automated Pipeline for Generating Verifiable and Diverse Function-Calling Datasets},
  author={Liu, Zuxin and Hoang, Thai and Zhang, Jianguo and Zhu, Ming and Lan, Tian and Kokane, Shirley and Tan, Juntao and Yao, Weiran and Liu, Zhiwei and Feng, Yihao and others},
  journal={arXiv preprint arXiv:2406.18518},
  year={2024}
}